feat(analysis): report_families.py - riepilogo strategie/famiglie per anno + integrazione

Report consolidato: (A) Ret%/anno per famiglia (FADE/HONEST/PAIRS/TSM01) e portafogli,
(B) Ret%/anno per ogni strategia singola (15 sleeve), (C) analisi di integrazione delle
nuove famiglie nel MASTER (+pairs/+TSM01/esteso) con correlazioni, (D) numeri sobri e
raccomandazione (leva 2x, cap pairs 30-35%).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

.gitignore

@@ -16,3 +16,4 @@ data/processed/
 *.pt
 *.pth
 notebooks/.ipynb_checkpoints/
+data/paper_trades/

CLAUDE.md

@@ -9,29 +9,50 @@ Progetto di ricerca: riconoscimento pattern frattali per trading algoritmico su
 - **Linguaggio:** Python 3.11+
 - **Package manager:** uv (dipendenze in `pyproject.toml`, lock in `uv.lock`)
 - **Dati:** Parquet in `data/raw/` (non committati, ~70 MB)
-- **ML:** scikit-learn (GradientBoosting), PyTorch (LSTM)
+- **ML:** scikit-learn (GradientBoostingClassifier)
 - **Analisi:** numpy, pandas, scipy
 - **API dati:** Cerbero MCP su `cerbero-mcp.tielogic.xyz` (Deribit, Bybit, Hyperliquid), ccxt/Binance come fallback
+- **Config:** pyyaml per `strategies.yml`
 
 ## Struttura
 
 ```
-src/data/         → download e caricamento dati (downloader.py)
-src/fractal/      → indicatori frattali (patterns.py, indicators.py, similarity.py)
-src/backtest/     → engine di backtesting (engine.py)
-scripts/          → analisi e strategie numerate 01–13
-docs/diary/       → diario di ricerca giornaliero
-data/raw/         → file .parquet OHLCV (gitignored)
-data/processed/   → modelli salvati (gitignored)
+src/data/              → download e caricamento dati (downloader.py)
+src/fractal/           → indicatori frattali (patterns.py, indicators.py, similarity.py)
+src/backtest/          → engine di backtesting (engine.py)
+src/strategies/        → classe base Strategy ABC + indicatori condivisi
+  base.py              → Strategy, Signal, BacktestResult, YearlyStats
+  indicators.py        → keltner_ratio, detect_squeezes, ema, atr, rv, correlation
+src/live/              → paper trading live multi-strategia
+  multi_runner.py      → orchestratore: carica YAML, fetch candele, tick worker
+  strategy_worker.py   → worker indipendente: capital, trade log, stato persistente.
+                         Exit guidati da strategia (TP/SL/max_bars via Signal.metadata),
+                         fallback hold_bars/stop -2%. Usa fee_rt della strategia.
+  strategy_loader.py   → import dinamico classi Strategy da scripts/strategies/
+  cerbero_client.py    → client HTTP per Cerbero MCP (Deribit testnet)
+  signal_engine.py     → squeeze + ML real-time (legacy ML01, ora in waste) + validazione OOS
+  telegram_notifier.py → notifiche Telegram per trade
+scripts/strategies/    → strategie con edge validato OOS (solo MR01_bollinger_fade)
+scripts/waste/         → strategie scartate (W01-W28 + famiglia squeeze SQ/MT/ML/AD/CM/PD)
+scripts/analysis/      → ricerca/validazione OOS fee-aware (strategy_research, oos_validation, ...)
+strategies.yml         → config multi-strategy paper trader
+docs/diary/            → diario di ricerca giornaliero
+docs/specs/            → specifiche di design
+data/raw/              → file .parquet OHLCV (gitignored)
 ```
 
 ## Comandi
 
 ```bash
-uv sync                                    # installa dipendenze
-uv run python -m src.data.downloader       # scarica dati storici
-uv run python scripts/13_squeeze_ml_hybrid.py  # strategia vincente
-uv run pytest                              # test
+uv sync                                              # installa dipendenze
+uv run python -m src.data.downloader                  # scarica dati storici
+uv run python scripts/strategies/MR01_bollinger_fade.py        # strategia attiva (mean-reversion)
+uv run python scripts/analysis/strategy_research.py            # ricerca strategie fee-aware OOS
+uv run python scripts/analysis/oos_validation.py               # perche' la famiglia squeeze e' scartata
+uv run python scripts/analysis/validate_worker_mr01.py         # replay worker reale su MR01
+uv run python -m src.live.multi_runner                # paper trading live multi-strategia
+docker compose up -d                                  # deploy Docker
+uv run pytest                                         # test
 ```
 
 ## Dati storici
@@ -47,22 +68,136 @@ df = load_data("ETH", "15m")  # carica un asset/timeframe
 Fonte primaria: Cerbero MCP (endpoint `/mcp-deribit/tools/get_historical`).
 Token observer: nel file `secrets/observer.token` del progetto CerberoSuite.
 
-## Strategia vincente
+## Strategie attive
 
-**Squeeze + ML ibrida** (script 13):
+> **LEZIONE CRITICA (2026-05-28).** L'intera famiglia squeeze-breakout (SQ01-04,
+> MT01, ML01, AD01, CM01, PD01) è stata **scartata in `scripts/waste/`**: le
+> accuratezze storiche 76-82% erano un **artefatto di look-ahead**. Quei backtest
+> decidono la direzione con `sign(close[i]-close[i-1])` (la candela di breakout `i`)
+> ma entrano a `close[i-1]` — cioè comprano *prima* della candela che usano per
+> scegliere la direzione. Dal vivo il worker scopre il breakout solo a `close[i]`
+> ed entra lì: l'edge sparisce (win-rate ~47%, lancio di moneta). Sotto ingresso
+> onesto e fee reali **tutte perdono, anche a fee zero**. Inoltre i breakout
+> *rientrano* (mean-reversion > continuation). Vedi `scripts/analysis/oos_validation.py`
+> e `intrabar_test.py`.
 
-1. Rileva squeeze di volatilità (Bollinger dentro Keltner)
-2. Al rilascio dello squeeze, estrai feature strutturali dalla finestra
-3. GradientBoosting predice direzione con walk-forward training
-4. Trade solo se modello ha confidenza ≥ 70%
+Tutte le strategie estendono `src.strategies.base.Strategy`
+(`generate_signals() → backtest()`). Le strategie mean-reversion condividono
+`src.strategies.fade_base.FadeStrategy` (backtest intrabar TP/SL/max_bars).
+**Strategie con edge netto validato OOS fee-aware (tutte fade/mean-reversion):**
 
-Configurazione migliore: ETH 15m, BBw=14, squeeze threshold=0.8, breakout=3 barre, leva 3x, position 15%.
+| Codice | Nome | Meccanismo | Edge OOS netto (1h, fee 0.10% RT) | DD | Note |
+|--------|------|-----------|-----------------------------------|----|------|
+| **MR01** | Bollinger Fade | banda std attorno a SMA | BTC +201% / ETH +1238% | 15-72% | Fada la banda, TP alla media, SL ad ATR |
+| **MR02** | Donchian Fade | estremi canale H/L | BTC +172% / ETH enorme | 30-42% | Fada la rottura del canale, TP al centro |
+| **MR07** | Return Reversal | z dei rendimenti di barra | BTC +105% / ETH +195% | 25-46% | Fada il movimento estremo, exit in ATR; esposizione ~8% |
 
-Risultato backtestato: 76.9% accuracy, 118% annuo, 4.2% drawdown, €13.78/giorno da €1.000.
+> **MR03 Keltner Fade** spostata in `scripts/waste/`: era la fade più debole
+> (BTC Sharpe 1.22, il filtro trend la peggiorava) e ridondante con MR01 (stessa
+> idea di banda). Rimuoverla dal portafoglio ne ha *migliorato* le metriche.
+> La funzione `keltner_fade` resta in `strategy_research_v2.py` come record.
+
+**Lezione confermata:** l'edge è sempre *mean-reversion* (i breakout rientrano).
+Il trend-following (Donchian trend, RSI cross) e gli oscillatori senza filtro
+(RSI revert, ADX-filtered fade) perdono netti → restano scartati.
+
+Ogni strategia è robusta su **tutta** la sua griglia parametri (entrambi gli asset
+→ tutte positive OOS) e su **tutte** le fee 0.00-0.20% RT (margine ampio).
+MR01 validato col worker reale: BTC +196% / ETH +251% OOS (nov 2023→mag 2026).
+Ricerca completa: `scripts/analysis/strategy_research.py` (MR01) e
+`scripts/analysis/strategy_research_v2.py` (MR02/MR03/MR07).
+Validazione live-path: `scripts/analysis/oos_validation.py`.
+
+**Filtro trend (riduzione DD + aumento Acc).** Tutte le fade accettano i parametri
+opzionali `trend_max` / `ema_long`: saltano i segnali quando il prezzo è troppo
+esteso rispetto al trend di fondo (`|close − EMA(ema_long)| / ATR(14) > trend_max`),
+cioè quando si starebbe fadando un trend/crollo estremo. Con `trend_max=3.0`,
+`ema_long=200` (default in `strategies.yml`): accuratezza su tutti gli sleeve
+e DD giù drasticamente su ETH (MR01 71%→26%, MR02 42%→25%, MR03 66%→34%,
+MR07 46%→21%), edge OOS confermato (vedi `scripts/analysis/risk_management.py`).
+Unica eccezione: MR03 BTC, dove il filtro peggiora entrambe → lasciato disattivo.
+Leva non robusta scartate: vol-target sizing e skip-alta-volatilità (peggiorano).
+
+**Portafoglio.** Diversificare su sotto-conti indipendenti equipesati (le 4 strategie
+× BTC/ETH, pos 0.15 ciascuno) abbatte il DD aggregato: ~14% full / ~10% OOS sul
+paniere di 8 sleeve, contro il 20-70% del singolo. È la vera leva anti-drawdown.
+
+**Combinare le due famiglie (fade + honest).** Le fade (reversione intraday 1h) e le
+honest (DIP/TR/ROT trend+rotazione multi-crypto) sono **quasi scorrelate**
+(correlazione cross-famiglia ~0.05). Combinarle in un unico portafoglio migliora il
+rischio/rendimento rispetto a ciascuna famiglia da sola: equal-weight dei 9 sleeve
+→ DD 5.2% full / 4.7% OOS e Sharpe 4.23 full / 4.33 OOS (vs honest-only 12.6% DD /
+2.20 Sharpe e fade-only 8.2% DD / 4.09 Sharpe), CAGR ~47% mantenuta. Studio in
+`scripts/analysis/combine_portfolio.py`.
+
+**ROT02 — riduzione DD (top_k 2→3).** La rotazione dual-momentum honest concentrava
+il book su 2 soli asset (DD 40%). Diversificare su 3 (`top_k=3`) dimezza quasi il DD
+(40%→26%) e *alza* pure il ritorno full (+1095%→+1303%, ret/DD da 27 a 50); il
+vol-target abbassa il DD ma sacrifica ritorno, quindi si tiene top_k=3 senza VT.
+Applicato a `ROT02_dual_momentum.py` e a `_rot_daily_equity` (usata dai portafogli).
+
+**Portafogli pronti (artefatti accorpati e migliorati).** Oltre a `PORT01` (solo
+honest), due script in `scripts/strategies/`:
+- `PORT02_fade_master.py` — le 3 fade × BTC/ETH accorpate (6 sleeve, filtro trend),
+  equal-weight daily: DD ~8.2% full / 5.9% OOS, Sharpe 3.95/4.09, CAGR ~46%.
+- `PORT03_all_master.py` — portafoglio MASTER (fade + honest, 9 sleeve). Due varianti:
+  `equal` (massimo Sharpe: DD 5.2%/4.7% OOS, Sharpe 4.23/4.33) e `5050` fra le due
+  famiglie (minimo DD: 5.0% full / 4.5% OOS). È la configurazione consigliata.
+Come `PORT01`, sono meta-portafogli (script `run()` di report), non `Strategy` con
+`generate_signals`, quindi non nel `strategy_loader`.
+
+**Esplorazione famiglie alternative (branch `strategy_explore`, 2026-05-29).** Esplorate
+9 famiglie nuove con agenti paralleli su harness onesto condiviso
+(`scripts/analysis/explore_lab.py`). 7 sono rumore (rifiutate: stagionalità oraria/mensile,
+cross-sectional reversal, opening-range breakout, lead-lag BTC→alt, continuation intraday —
+quest'ultima riconferma la dominanza mean-reversion). Due edge reali:
+- **PR01 Pairs** (`scripts/strategies/PR01_pairs_reversion.py`): spread reversion
+  market-neutral sul log-ratio z-score, **config UNIVERSALE** `n=50 z_in=2.0 z_exit=0.75
+  max_bars=72` (anti-overfit, niente tuning per-coppia). **5 coppie robuste**: ETH/BTC
+  (Sharpe 4.36), LTC/ETH (3.08), ADA/ETH (2.69), BTC/LTC (2.36, robusta anche 4h), ETH/SOL
+  (1.96, la più debole). Pattern: sempre alt-liquido vs major. Plateau confermato
+  (heatmap 20/20 Sharpe>1) + walk-forward (ETH/BTC 11/12 finestre+). **BNB/ETH scartata**
+  (overfit). Corr col mercato ~0.02-0.08. Fee su **2 gambe** (worker da estendere). Verifica: `pairs_research.py`.
+- **TSM01** (`scripts/analysis/tsmom_research.py`): TSMOM multi-orizzonte 3/6/12m + risk-off,
+  **gross 0.30**, distinto da ROT02 (corr 0.62), DD 15-22%, mai un anno negativo. Robusto
+  (36/36 config OOS+) ma diversificatore, non motore di ritorno (rende meno di ROT02).
+
+Aggiungere i **5 pairs** al MASTER (quasi scorrelati, ~0.02-0.09) è il free-lunch più
+grande (`scripts/analysis/combine_v2.py`). **Numeri sobri onesti** (l'OOS singolo 2024-25
+è regime calmo → ottimistico ~50%): worst-DD su 90g rolling **~6%** (non 2.3%), Sharpe
+atteso **~5** (mediana semestrale), ogni anno positivo dal 2021, regge **leva 2x +
+slippage doppio** (CAGR 36%, Sharpe 5.1). Config robusta raccomandata: **MASTER-esteso
+equal-weight, leva 2x, cap pairs ~30-35%** (i pairs sono ~57% del rischio e non ancora
+validati col worker live a 2 gambe). La confluenza multi-TF è stata SCARTATA (overfit).
+
+**Metodologia obbligatoria per ogni nuova strategia** (per non ripetere l'errore squeeze):
+1. Ingresso eseguibile: direzione e prezzo decisi con dati **fino a `close[i]`**, mai `close[i-1]` con direzione da `i`.
+2. Backtest **NETTO** dopo fee realistiche Deribit (**0.10% RT** taker, non 0.20%) + leva.
+3. Validazione **out-of-sample** (held-out) + robustezza su griglia parametri + sweep fee.
+4. Crea script in `scripts/strategies/`, aggiungi a `MODULE_MAP` (`strategy_loader.py`) e a `strategies.yml`.
+
+Strategie scartate storiche in `scripts/waste/` (W01-W28 + la famiglia squeeze).
+
+**Verso €50/giorno.** Con 4 strategie indipendenti (MR01/MR02/MR03/MR07) × 2 asset
+(BTC/ETH) su €1000 ciascuna, il PnL medio storico aggregato è ben oltre €50/giorno;
+ma quei numeri sono backtest a leva 3x su 8 anni e includono anni eccezionali (es.
+ETH 2024). Stima onesta: il target è *plausibile* su un portafoglio diversificato di
+queste fade, ma va confermato col paper trader live prima di rischiare capitale reale.
+
+## Multi-Strategy Paper Trader
+
+Orchestratore che esegue N strategie in parallelo su dati live Cerbero, ognuna con €1000 USDC virtuali indipendenti.
+
+**Config:** `strategies.yml` — lista strategie con asset, tf, sizing, parametri. Attualmente solo MR01 (BTC/ETH 1h).
+**Persistenza:** `data/paper_trades/{strategy}___{asset}__{tf}/` con `trades.jsonl` (append-only) + `status.json` (resume al restart, include tp/sl/max_bars).
+**Hot-add:** aggiungi riga YAML → `docker compose restart` → storico intatto.
+**Exit strategia:** se un `Signal` porta `tp`/`sl`/`max_bars` in `metadata` (come MR01), il worker esce su take-profit/stop-loss/time-limit a quei livelli; altrimenti usa il fallback `hold_bars`/stop -2%.
+**Notifiche:** Telegram per ogni trade (richiede `.env` con `TELEGRAM_BOT_TOKEN` e `TELEGRAM_CHAT_ID`).
 
 ## Convenzioni
 
-- Script numerati progressivamente (`01_`, `02_`, …). Ogni script è autocontenuto.
+- Strategie in `scripts/strategies/` con codice univoco (MR01, ...).
+- Script scartati in `scripts/waste/` (W01-W28 + famiglia squeeze).
 - Diario in `docs/diary/YYYY-MM-DD.md`. Aggiornare dopo ogni esperimento significativo.
 - Nessun dato sensibile nei commit (token, chiavi API). Usare `.gitignore`.
 - Verificare sempre assenza di data leakage prima di fidarsi dei risultati. In particolare: `returns[i-w : i]` include `close[i]` che è un candle nel futuro — usare `returns[i-w : i-1]`.
@@ -70,5 +205,8 @@ Risultato backtestato: 76.9% accuracy, 118% annuo, 4.2% drawdown, €13.78/giorn
 ## Attenzione
 
 - **Data leakage:** è stata trovata e corretta nello script 05. Ogni volta che si usano rendimenti logaritmici (`np.diff(np.log(close))`), ricordare che `returns[k]` usa `close[k+1]`. I feature devono fermarsi a `returns[i-2]` se il prezzo corrente è `close[i-1]`.
-- **Fee:** sempre 0.1% per lato (0.2% round-trip). Includere nel backtest.
+- **Fee:** Deribit perp reale = taker ~0.05%/lato (**0.10% round-trip**), maker ~0%. Usare 0.10% RT come baseline (lo 0.20% storico era pessimista 2x). Includere SEMPRE nel backtest: sono vincolo di prim'ordine, molte operazioni = morte per fee. Il worker usa `strategy.fee_rt` (MR01 = 0.001).
 - **Leva:** testato con 3x. Aumentare a 5x migliora i rendimenti ma raddoppia il drawdown.
+- **GBM:** GradientBoostingClassifier di scikit-learn. Ensemble di alberi decisionali sequenziali. Walk-forward per evitare leakage temporale.
+- **Cerbero `get_historical` (fix 2026-05-28):** `end_date` come data nuda è inclusivo dell'intera giornata fino all'ultima candela chiusa (es. `end=oggi` arriva fino ad ora, non più a mezzanotte); accettati anche timestamp con orario (`...T14:00:00`, naive=UTC); nessun cap a ~5000 righe (paginazione interna). Il client passa già `end=oggi`, ora corretto. Prima del fix il paper trader restava a zero trade perché il feed era fermo a mezzanotte.
+- **Dati ETH Deribit 15m:** 14-30%/anno di candele *flat* (O=H=L=C, volume 0, run fino a ~54h) per bassa liquidità del perpetuo. Verificato (2026-05-28): escluderle NON cambia i backtest (Δacc ≤0.5pp) → edge robusto. Resta un caveat operativo (slippage/fill in trading reale, irrilevante per paper). BTC pulito eccetto picco ~8% nel 2024.

Dockerfile

@@ -8,7 +8,9 @@ COPY pyproject.toml uv.lock ./
 RUN uv sync --frozen --no-dev
 
 COPY src/ src/
+COPY scripts/strategies/ scripts/strategies/
+COPY strategies.yml strategies.yml
 
 VOLUME /app/data
 
-CMD ["uv", "run", "python", "-m", "src.live.paper_trader"]
+CMD ["uv", "run", "python", "-m", "src.live.multi_runner"]

README.md

@@ -8,80 +8,189 @@ Partendo da un capitale iniziale di €1.000, raggiungere un profitto medio di
 
 ## Risultati
 
-Tredici strategie testate su dati storici 2018–2026 (BTC e ETH, timeframe 5m / 15m / 1h). Le migliori cinque:
+> ⚠️ **Revisione 2026-05-28.** La famiglia squeeze-breakout (SQ/MT/ML/AD/CM/PD, con
+> accuracy storiche dichiarate 76-82%) è stata **scartata**: quei numeri erano un
+> **artefatto di look-ahead**. I backtest decidevano la direzione dalla candela di
+> breakout `close[i]` ma entravano a `close[i-1]` — impossibile dal vivo. Sotto
+> ingresso onesto (`close[i]`) e fee reali, l'edge sparisce e tutte perdono, anche
+> a fee zero. Dettagli e prove: `scripts/analysis/oos_validation.py`.
 
-| # | Strategia | Accuracy | ROI annuo | Max DD | €/giorno |
-|---|-----------|----------|-----------|--------|----------|
-| 1 | ETH 15m Squeeze + ML ibrida | 76.9% | 118% | 4.2% | €13.78 |
-| 2 | ETH 1h Squeeze + Vol | 83.9% | 22% | 2.0% | €0.71 |
-| 3 | BTC 15m Squeeze + ML ibrida | 78.8% | 69% | 7.0% | €5.51 |
-| 4 | ETH 1h Squeeze (BBw=30) | 82.8% | 47% | 3.2% | €1.77 |
-| 5 | ETH Walk-Forward ML | 57.7% | 38% | 47% | €3.12 |
+Dopo una validazione **out-of-sample, fee-aware** di tutte le famiglie, l'unica con
+edge netto reale è il **mean-reversion** (i breakout *rientrano*, non continuano):
 
-La strategia vincente (#1) opera su ETH a 15 minuti con ~1 trade al giorno, leva 3x e drawdown contenuto al 4.2%.
+| Codice | Strategia | Mercato | Edge OOS netto | Max DD | Robustezza |
+|--------|-----------|---------|----------------|--------|------------|
+| **MR01** | Bollinger Fade (mean-reversion) | BTC 1h | **+196 / +201%** | 15% | ✅ |
+| **MR01** | Bollinger Fade (mean-reversion) | ETH 1h | **+251%** | ~25% | ⚠️ DD alto |
+
+Netto dopo **fee realistiche Deribit 0.10% RT** (taker), leva 3x, pos 15%, su finestra
+held-out (nov 2023→mag 2026). MR01 è positivo su **tutta** la griglia parametri
+(`n∈{14,20,30,50}` × `k∈{2.0,2.5,3.0}`) e per **ogni** livello di fee 0.00-0.20% RT —
+margine di sicurezza ampio, niente parametro fortunato. Ri-validato col worker live reale.
 
 ## Come funziona
 
-### Volatility Squeeze Breakout
+### MR01 — Bollinger Fade (mean-reversion)
 
-Il meccanismo centrale sfrutta i cicli naturali di compressione ed espansione della volatilità:
+La strategia attiva sfrutta il fatto, emerso dai dati, che su BTC/ETH a 1h gli estremi
+di prezzo **rientrano verso la media** più di quanto proseguano:
 
-1. **Compressione** — le Bollinger Bands entrano dentro i Keltner Channel (il prezzo si muove sempre meno, accumulando "energia").
-2. **Breakout** — le bande escono dal canale. Un impulso direzionale parte.
-3. **Conferma ML** — un modello GradientBoosting, addestrato su feature strutturali e frattali della finestra precedente, conferma la direzione e filtra i segnali deboli.
+1. **Bollinger Bands** (window `n`, `k` deviazioni standard) sul close.
+2. **Entry** — quando il close esce *sotto* la banda inferiore → **long** (o *sopra* la superiore → **short**). Ingresso a `close[i]`, eseguibile dal vivo.
+3. **Take-profit** alla media mobile (il rientro atteso).
+4. **Stop-loss** a `sl_atr × ATR` oltre l'estremo; **time-limit** a `max_bars`.
 
-### Feature frattali
+Nessun look-ahead: direzione e livelli sono calcolati con dati fino a `close[i]`.
 
-- Rapporti body/shadow normalizzati su finestre multiple (12, 24, 48 candele)
-- Momentum, volatilità, skewness, kurtosis dei rendimenti logaritmici
-- Autocorrelazione lag-1
-- Profilo volumetrico e spike detection
-- Durata della fase di squeeze e rapporto di espansione Keltner
-- Posizione del prezzo rispetto al range recente e ATR normalizzato
+### Perché lo squeeze breakout è stato abbandonato
+
+L'ipotesi originale era opposta — *continuazione* dopo la compressione di volatilità
+(Bollinger dentro Keltner → breakout direzionale). Su dati storici sembrava dare
+76-82% di accuracy, ma era un **artefatto di look-ahead**: il backtest entrava a
+`close[i-1]` con direzione decisa da `close[i]`. Replicando l'esecuzione reale
+(ingresso a `close[i]`) l'edge collassa al ~47% (lancio di moneta) e i costi fanno
+il resto. Il test sui breakout intra-barra a 5m conferma che il movimento *rientra*
+subito (mean-reversion), giustificando MR01. Tutta la famiglia squeeze è in `scripts/waste/`.
+
+### Lezione metodologica
+
+Ogni nuova strategia deve passare: (1) **ingresso eseguibile** senza look-ahead,
+(2) backtest **netto** dopo fee realistiche (0.10% RT Deribit), (3) validazione
+**out-of-sample** + robustezza su griglia parametri + sweep fee. Strumenti in
+`scripts/analysis/` (`strategy_research.py`, `oos_validation.py`, `intrabar_test.py`).
 
 ## Struttura progetto
 
 ```
 PythagorasGoal/
 ├── src/
-│   ├── data/          # Download e gestione dati storici (Cerbero MCP + Binance)
-│   ├── fractal/       # Indicatori frattali: Hurst, Higuchi FD, self-similarity
-│   ├── backtest/      # Motore di backtesting con fee e metriche
-│   ├── strategies/    # (predisposto per strategie modulari)
-│   ├── nn/            # (predisposto per reti neurali)
-│   └── utils/
-├── scripts/           # Script di analisi e test (01–13)
+│   ├── data/              # Download e gestione dati (Cerbero MCP + Binance)
+│   ├── fractal/           # Indicatori frattali: Hurst, Higuchi FD, self-similarity
+│   ├── backtest/          # Motore di backtesting con fee e metriche
+│   ├── strategies/        # Classe base Strategy ABC + indicatori condivisi
+│   │   ├── base.py        # Strategy, Signal, BacktestResult, YearlyStats
+│   │   └── indicators.py  # keltner_ratio, detect_squeezes, ema, atr, rv, corr
+│   └── live/              # Paper trading live su Deribit testnet
+│       ├── multi_runner.py     # Orchestratore multi-strategia
+│       ├── strategy_worker.py  # Worker indipendente con stato persistente
+│       ├── strategy_loader.py  # Import dinamico classi Strategy
+│       ├── cerbero_client.py   # Client HTTP per Cerbero MCP
+│       ├── signal_engine.py    # Squeeze + ML real-time (legacy) + validazione OOS
+│       └── telegram_notifier.py
+├── scripts/
+│   ├── strategies/        # Strategie con edge validato OOS (solo MR01_bollinger_fade)
+│   ├── waste/             # Strategie scartate (W01-W28 + famiglia squeeze SQ/MT/ML/AD/CM/PD)
+│   └── analysis/          # Ricerca/validazione OOS fee-aware (strategy_research, oos_validation, ...)
+├── strategies.yml         # Config multi-strategy paper trader
 ├── data/
-│   ├── raw/           # Parquet OHLCV (non committati, ~70 MB)
-│   └── processed/     # Modelli salvati
+│   └── raw/               # Parquet OHLCV (gitignored, ~70 MB)
 ├── docs/
-│   └── diary/         # Diario di ricerca giornaliero
-├── tests/
-├── pyproject.toml
-└── README.md
+│   ├── diary/             # Diario di ricerca giornaliero
+│   └── specs/             # Specifiche di design
+├── Dockerfile
+├── docker-compose.yml
+└── pyproject.toml
 ```
 
+## Strategie attive
+
+Tutte le strategie estendono `src.strategies.base.Strategy` (`generate_signals() → backtest()`).
+
+| Codice | Script | Tipo | Descrizione |
+|--------|--------|------|-------------|
+| **MR01** | `MR01_bollinger_fade.py` | Mean-reversion | Fada la banda di Bollinger, TP alla media, SL ad ATR. Unica con edge netto validato OOS. |
+
+La famiglia squeeze (SQ01-04, ML01, MT01, PD01, CM01, AD01) è in `scripts/waste/`:
+edge storico = artefatto di look-ahead (vedi sezione *Come funziona*).
+
+Per eseguire il backtest della strategia:
+
+```bash
+uv run python scripts/strategies/MR01_bollinger_fade.py
+```
+
+Per la ricerca/validazione fee-aware out-of-sample:
+
+```bash
+uv run python scripts/analysis/strategy_research.py     # screening famiglie + deep-dive MR01
+uv run python scripts/analysis/oos_validation.py        # perche' la famiglia squeeze e' scartata
+uv run python scripts/analysis/validate_worker_mr01.py  # replay del worker live su MR01
+```
+
+## Paper Trading Live
+
+Il multi-strategy runner esegue N strategie in parallelo su dati live da Cerbero MCP, ognuna con €1000 USDC virtuali indipendenti. Se un `Signal` porta `tp`/`sl`/`max_bars` in `metadata` (come MR01), il worker chiude su take-profit alla media / stop-loss ad ATR / time-limit; altrimenti usa il fallback `hold_bars`/stop -2%.
+
+### Avvio
+
+```bash
+# Locale
+uv run python -m src.live.multi_runner
+
+# Docker
+docker compose up -d
+```
+
+### Configurazione
+
+Le strategie attive sono definite in `strategies.yml`:
+
+```yaml
+defaults:
+  capital: 1000
+  position_size: 0.15
+  leverage: 3
+
+strategies:
+  - name: MR01_bollinger_fade
+    asset: BTC
+    tf: 1h
+    enabled: true
+    params:
+      bb_window: 50
+      k: 2.5
+      sl_atr: 2.0
+      max_bars: 24
+```
+
+Per aggiungere una strategia: nuova riga in `strategies.yml`, poi `docker compose restart`. Lo storico delle strategie esistenti rimane intatto.
+
+### Persistenza
+
+Ogni strategia ha la sua directory in `data/paper_trades/`:
+
+```
+data/paper_trades/
+  MR01_bollinger_fade__BTC__1h/
+    trades.jsonl    # Storico trade append-only
+    status.json     # Stato corrente (resume al restart, include tp/sl/max_bars)
+```
+
+Notifiche Telegram per ogni trade (richiede `TELEGRAM_BOT_TOKEN` e `TELEGRAM_CHAT_ID` in `.env`).
+
 ## Setup
 
 ```bash
-# Clona il repository
+# Clona e installa
 git clone <repo-url> && cd PythagorasGoal
-
-# Installa dipendenze (richiede uv)
 uv sync
 
-# Scarica dati storici (~70 MB, richiede connessione)
+# Scarica dati storici (~70 MB)
 uv run python -m src.data.downloader
 
-# Esegui la strategia ibrida vincente
-uv run python scripts/13_squeeze_ml_hybrid.py
+# Backtest strategia attiva
+uv run python scripts/strategies/MR01_bollinger_fade.py
+
+# Paper trading live
+uv run python -m src.live.multi_runner
 ```
 
 ### Requisiti
 
 - Python ≥ 3.11
 - [uv](https://docs.astral.sh/uv/) come package manager
-- Accesso a Cerbero MCP (`cerbero-mcp.tielogic.xyz`) per i dati Deribit, oppure Binance via ccxt come fallback
+- Accesso a Cerbero MCP (`cerbero-mcp.tielogic.xyz`) per dati Deribit live
+- Docker (opzionale, per deploy su VPS)
 
 ## Dati
 
@@ -90,25 +199,7 @@ uv run python scripts/13_squeeze_ml_hybrid.py
 | BTC | 5m / 15m / 1h | 883K / 294K / 74K | 2018-01 → oggi |
 | ETH | 5m / 15m / 1h | 882K / 294K / 74K | 2018-01 → oggi |
 
-Fonte primaria: Deribit perpetual via Cerbero MCP. Fallback per il periodo antecedente: Binance spot via ccxt. Formato: Apache Parquet.
-
-## Strategie testate
-
-| Script | Approccio | Esito |
-|--------|-----------|-------|
-| 01 | Pattern candlestick discreti (U/D/0) | Nessun edge |
-| 02 | DTW pattern matching | Troppo lento, edge minimo |
-| 03 | Proiezione FFT (ispirata al paper) | Random (49.8%) |
-| 04 | GBM su feature frattali (Hurst, FD) | 63.6% a soglia 0.65 |
-| 05 | GBM multi-window (corretto data leakage) | 58.9% |
-| 06 | GBM su feature strutturali normalizzate | 58.6%, +57.5% return |
-| 07 | LSTM su sequenze candele | 58.4%, comparabile a GBM |
-| 08 | Ensemble multi-timeframe (1h + 15m) | 59.2% (consensus 2/3) |
-| 09 | Walk-forward ML | 57.7%, Sharpe 7.4, €3.12/day |
-| 10 | Ensemble 5 modelli alta precisione | In corso |
-| 11 | **Volatility Squeeze Breakout** | **83.9%**, approccio strutturale |
-| 12 | Report finale e simulazione crescita | — |
-| 13 | **Squeeze + ML ibrida** | **76.9%**, 118% ann, €13.78/day |
+Fonte primaria: Deribit perpetual via Cerbero MCP. Fallback: Binance spot via ccxt. Formato: Apache Parquet.
 
 ## Riferimenti
 

docker-compose.yml

@@ -1,14 +1,17 @@
 services:
   paper-trader:
     build: .
-    container_name: pythagoras-paper
+    container_name: pythagoras-multi
     restart: unless-stopped
     volumes:
       - ./data:/app/data
+      - ./strategies.yml:/app/strategies.yml:ro
+    env_file:
+      - .env
     environment:
       - PYTHONUNBUFFERED=1
     healthcheck:
-      test: ["CMD", "python", "-c", "import json; s=json.load(open('/app/data/paper_trades/status.json')); assert s['last_update']"]
+      test: ["CMD", "python", "-c", "import os; assert any(f.endswith('status.json') for r,d,fs in os.walk('/app/data/paper_trades') for f in fs)"]
       interval: 120s
       timeout: 10s
       retries: 3

docs/diary/2026-05-28-honest-strategies.md

@@ -0,0 +1,136 @@
+# 2026-05-28 — Ricerca onesta di nuove strategie (post-squeeze)
+
+## Contesto e mandato
+
+Dopo aver scoperto che l'intera famiglia squeeze-breakout era un artefatto di
+look-ahead (accuratezze 76-82% svanite sotto ingresso eseguibile), il mandato è
+stato: trovare in modo **onesto** almeno 3 strategie attendibili, testate su ~8
+anni e su più criptovalute, con le fee incluse nella valutazione, partendo da
+€1.000 con l'obiettivo (aspirazionale) di €50/giorno. Esplorare anche idee fuori
+dal comune e l'uso combinato di più crypto e timeframe.
+
+## Metodologia (engine onesto)
+
+Tutto il lavoro usa un unico engine condiviso (`scripts/analysis/honest_lab.py`)
+con questi vincoli anti-illusione:
+
+1. **Ingresso eseguibile.** Ogni segnale alla barra `i` usa solo dati fino a
+   `close[i]` e l'ingresso avviene a `close[i]` (ciò che il worker live vede e
+   può eseguire). Disponibile anche l'ingresso più conservativo a `open[i+1]`.
+2. **Uscita realistica.** Take-profit / stop-loss valutati intrabar su `high`/`low`,
+   in modo conservativo (SL prima del TP nello stesso bar), più time-limit.
+   Una posizione per volta (non-overlap), capitale composto.
+3. **Fee di prim'ordine.** Tutto è NETTO dopo fee round-trip realistiche Deribit
+   (0.10% RT) moltiplicate per la leva (3x), con sweep fino a 0.20% RT.
+4. **Validazione severa.** FULL + out-of-sample (ultimo 30%) + conteggio anni
+   positivi + sweep fee + griglia parametri + test su **8 crypto**
+   (BTC, ETH, SOL, BNB, XRP, LTC, DOGE, ADA, 2018→2026).
+
+## Lezione madre
+
+**Shortare le crypto perde OOS in modo sistematico in questo campione.** Sia la
+mean-reversion sul lato short, sia il momentum short, crollano fuori campione: il
+periodo 2018-2026 è net-bull e ogni rialzo "estremo" tende a continuare invece di
+rientrare. Tutte le configurazioni che sopravvivono oneste sono **long-biased**.
+È un fatto da dichiarare: parte della performance OOS è correlata al beta rialzista
+delle crypto. Le strategie aggiungono *timing* sopra quel beta, non lo eliminano.
+
+## Le 3 strategie selezionate (meccanismi distinti)
+
+| Codice | Meccanismo | TF | Asset robusti | OOS netto (fee 0.10% RT) | DD | Anni+ |
+|--------|-----------|----|---------------|--------------------------|----|-------|
+| **DIP01** | Dip-buy z-score reversion (long-only) | 1h | BTC, ETH, SOL | BTC +59% · ETH +224% · SOL +13% | 23-55% | 6-7/9 |
+| **TR01** | EMA 20/100 trend-following (long-only) | 4h | BNB, BTC, DOGE, SOL, XRP | BTC +27% · DOGE +53% · XRP +29% | 29-53% | 4-6/8 |
+| **ROT01** | Rotazione cross-sectional momentum sul paniere | 1d | intero paniere (8) | **+44%** | 53% | 5/7 |
+
+Dettagli e riproducibilità: `scripts/analysis/honest_final.py` (tabella di
+validazione unica), `honest_rotation.py`, `honest_trend.py`, `honest_candidates.py`,
+`honest_diag.py`/`honest_diag2.py` (diagnostica long/short e filtro trend).
+
+### DIP01 — compra le capitolazioni
+Long-only: entra quando lo z-score del prezzo rispetto alla media a 50 barre scende
+sotto −2.5 (capitolazione), prende profitto al rientro verso la media, SL a 2.5·ATR.
+È la versione robusta e onesta della famiglia mean-reversion: regge lo sweep fee
+fino a 0.20% RT (BTC +45% OOS anche a 0.20%). Funziona sui major (BTC/ETH/SOL); sugli
+alt molto parabolici (DOGE/BNB) un dip fisso continua a scendere e non ha edge.
+
+### TR01 — cavalca i trend
+Long-only: in posizione quando EMA(20) > EMA(100) sul 4h, altrimenti cash. Poche
+operazioni (≈200 flip in 8 anni) ⇒ le fee non sono letali. È **complementare** a
+DIP01: guadagna nei regimi di trend, dove la reversione soffre.
+
+### ROT01 — la più affidabile e "fuori dal comune"
+Una sola strategia che usa **tutto il paniere** in un unico book: ogni giorno ordina
+le 8 crypto per momentum (rendimento a 60 giorni) e alloca a parti uguali alle 2
+migliori con momentum positivo, il resto in cash. Cattura la *dispersione* tra
+crypto (gli alt forti corrono molto più di BTC nei bull) senza shortare nulla.
+È **param-insensitive** (tutte le combinazioni lookback/top-k sono positive OOS) e
+regge le fee fino a 0.20% RT (+41% OOS). Risponde direttamente alla richiesta di
+combinare più crypto e un timeframe diverso in un'unica strategia. Per-anno:
+2020 +33% · 2021 +181% · 2022 −29% (bear) · 2023 +43% · 2024 +59% · 2025 +6% · 2026 −10% (YTD).
+
+## Diversificazione
+
+I tre meccanismi coprono regimi diversi e in larga misura anti-correlati:
+reversione (DIP01), momentum di singolo asset (TR01), forza relativa cross-asset
+(ROT01). Eseguirli insieme produce una curva di equity più liscia del singolo.
+
+## Onestà sull'obiettivo €50/giorno
+
+Va detto chiaramente: **€50/giorno su €1.000 in pochi mesi non è raggiungibile a
+rischio sano.** Significa ~€18.250/anno, cioè ~1.825%/anno; gli edge onesti qui
+trovati rendono il 30-60% OOS su orizzonti pluriennali. Le strade per avvicinare
+quel numero sono: (a) far crescere il capitale per anni con interesse composto —
+€50/giorno diventa plausibile solo quando il capitale è molto più grande; (b) alzare
+la leva, che però aumenta proporzionalmente il drawdown (già 23-55%) ed espone a
+rovina; (c) aggiungere capitale. Nessuna di queste è una scorciatoia. La proposta
+onesta è un portafoglio delle 3 strategie a leva moderata, puntando alla
+**sopravvivenza e alla crescita composta**, non al target giornaliero immediato.
+
+## Miglioramenti (alzare Acc, ridurre DD, migliorare PnL)
+
+Leve oneste e documentate, senza tuning sui singoli anni
+(`scripts/analysis/honest_improve.py`, `honest_improve2.py`):
+
+### ROT02 — dual-momentum overlay (migliora TUTTO)
+Alla rotazione cross-sectional di ROT01 si aggiunge un overlay di *absolute
+momentum*: cash quando BTC è sotto la sua media a 100 giorni (mercato risk-off).
+Taglia i bear di sistema (gli unici anni rossi di ROT01).
+
+| | FULL% | OOS% | DD% |
+|---|---|---|---|
+| ROT01 base | +679 | +44 | 53 |
+| **ROT02 (SMA100)** | **+1095** | **+98** | **40** |
+
+PnL su, DD giù: dominanza su tutte e tre le metriche. Param-insensitive (SMA100-150).
+
+### DIP01 — market-gate (variante low-DD)
+Comprare i dip solo quando BTC è risk-on alza l'**Acc** (ETH 52→57%, SOL 49→52%) e
+**dimezza il DD** (ETH 53→23%, SOL 25→13%), al costo di parte della PnL (meno trade).
+È de-risking, non un pasto gratis: utile per chi vuole una curva più liscia. Su BTC
+il gate va evitato (i dip migliori di BTC arrivano proprio quando BTC è sotto la
+propria SMA), quindi DIP01 base resta la versione di riferimento per BTC.
+
+### PORT01 — portafoglio combinato (il vero motore di risk-reduction)
+Equal-weight giornaliero ribilanciato delle 3 sleeve anti-correlate
+(DIP01 BTC + TR01 basket + ROT02). La diversificazione porta il DD del portafoglio
+**sotto** quello della sleeve meno rischiosa, mantenendo una CAGR alta.
+
+| Sleeve | ret% | DD% | CAGR% |
+|--------|------|-----|-------|
+| DIP01 BTC | +322 | 15 | 31 |
+| TR01 basket | +591 | 27 | 43 |
+| ROT02 dual-mom | +771 | 40 | 49 |
+| **PORTAFOGLIO** | **+642** | **12** | **45** |
+
+Per-anno portafoglio: 2021 +203% · 2022 **−1%** (bear neutralizzato, era −30% su ROT) ·
+2023 +47% · 2024 +50% · 2025 +14% · 2026 −2% (YTD). Nessun anno realmente negativo,
+DD massimo 12%, CAGR 45%. È la configurazione di deployment raccomandata.
+
+## Prossimi passi
+
+- Integrare DIP01 nel worker (già compatibile: Signal con tp/sl/max_bars).
+- Trailing-stop ad ATR per TR01 (per alzarne l'Acc e ridurne ulteriormente il DD).
+- Estendere il worker per strategie position-based (TR01) e di portafoglio (ROT01).
+- Backtest del portafoglio combinato con ribilanciamento del capitale.
+- Walk-forward rolling (oltre al singolo split 70/30) per confermare la stabilità.

docs/diary/2026-05-28.md

@@ -0,0 +1,193 @@
+# 2026-05-28 — Giorno 3: Bug dati Cerbero, paper trader fermo, fix MT01 multi-timeframe
+
+### 12:20 — Sintomo: paper trader live a zero trade
+
+**Cosa:** check del container `pythagoras-multi` (multi-strategy paper trader, 6 strategie).
+**Reale:** container healthy da ore, ma **0 trade** su tutte le strategie, tutte FLAT a €1000.
+Primo falso indizio: `last_bar_ts: 0` in tutti gli `status.json`. Indagando il worker,
+quel campo si aggiorna **solo a posizione aperta** (contatore `hold_bars`), non ad ogni
+candela → non è la causa. Il loop era vivo (status.json riscritti ogni 60s).
+**Lezione:** non fidarsi del nome di un campo; verificare nel codice quando viene scritto.
+L'healthcheck del container controlla solo l'esistenza di `status.json`, non la freschezza
+→ un loop bloccato risulterebbe comunque "healthy".
+
+### 12:45 — Causa radice: bug lato Cerbero MCP `get_historical`
+
+**Cosa:** probe dirette all'endpoint `/mcp-deribit/tools/get_historical`.
+**Reale:** due bug lato server:
+1. **`end_date` data-nuda tronca a mezzanotte:** `end=oggi` restituiva candele solo fino a
+   `oggi 00:00`. Il `df` live finiva sempre alla barra di mezzanotte e **non avanzava** durante
+   la giornata → nessun breakout fresco sull'ultima barra → nessun ingresso (condizione worker
+   `last_signal.idx >= last_idx - 1`).
+2. **Cap a ~5000 righe** che ignora `start_date`: una richiesta di 365g a 15m restituiva ~52
+   giorni. Ecco perché ML01 si addestrava su soli 88 samples (overfit, train_acc 100%).
+**Lezione:** lo zero-trade non era nelle strategie ma nel feed dati. Sempre validare la
+freschezza/copertura dei dati prima di sospettare la logica.
+
+### 13:30 — Fix lato Cerbero + verifica
+
+**Cosa:** report passato al dev di `cerbero-mcp`; fix deployato (riavvio container) + doc
+aggiornata in `cerbero-mcp/docs/API_REFERENCE.md`.
+**Reale dopo deploy (verificato con probe):**
+- `end=oggi` (data nuda) → ultima candela = ora corrente (age ~3 min). ✅
+- 365g a 15m → **35.099 candele**, span 365.6g, nessun cap. ✅
+- Supportati anche timestamp con orario (`...T14:00:00`, naive = UTC). ✅
+
+Nostro client (`src/live/cerbero_client.py`) invariato: passa già `end=oggi`, ora corretto.
+**Lezione:** "trust but verify" — la doc dichiarava i fix prima che fossero deployati; solo
+la probe diretta ha confermato cosa era davvero attivo sul server.
+
+### 14:00 — Problema residuo: MT01 usava un trend 1h STANTIO
+
+**Cosa:** check di tutte le strategie sul percorso di codice reale con dati freschi.
+**Reale:**
+- Tutte le 6 strategie girano senza crash; SQ01/SQ02 generano molti segnali.
+- **MT01 leggeva il trend 1h dal parquet statico** (`load_data(asset,"1h")`), non da Cerbero.
+  Il parquet finiva a mezzanotte → per ogni barra 15m di oggi `searchsorted` cadeva oltre la
+  fine e si agganciava sempre alla candela di mezzanotte (gap 14.8h). La conferma
+  multi-timeframe — il cuore di MT01 — era di fatto congelata e il gap cresce ogni giorno.
+- In `data/raw/` mancavano del tutto i parquet **15m** (`btc_15m`, `eth_15m`) → backtest 15m rotti.
+**Lezione:** una strategia live che dipende da un file statico ha un punto cieco temporale;
+il dato live e quello di backtest devono provenire da fonti coerenti.
+
+### 14:30 — Fix MT01: trend 1h live da Cerbero
+
+**Cosa:** modifica al runner perché MT01 prenda l'1h live, non dal parquet.
+- `MT01.generate_signals` accetta un `df_1h` opzionale (fallback al parquet se assente).
+- `StrategyWorker.tick(df, df_1h=None)` lo inoltra ai signal.
+- `multi_runner` fa fetch 1h live (resolution 60) per gli asset MT01 ad ogni poll (`htf_cache`).
+**Reale (verificato a codice montato, pre-rebuild):** gap del trend 1h sull'ultima barra
+**0.75h** (fresco) contro **14.8h** col parquet statico. Segnali invariati sullo storico.
+**Lezione:** isolare la dipendenza dal file statico rende MT01 immune al drift tra un
+`download_all()` e l'altro.
+
+### 14:55 — Rigenerazione dati + rebuild
+
+**Cosa:** `download_asset` per 15m+1h (saltati 1m/5m, lenti e inutilizzati), poi
+`docker compose up -d --build` (il codice `src/` è baked nell'immagine).
+**Reale:** parquet rigenerati con storia completa 2018→2026 e freschi (15m fino alle 14:45,
+1h fino alle 14:00). Container ripartito: 6 strategie attive, ML01 riaddestrato su **534
+samples** (anno pieno), MT01 senza errori, fetch 1h live OK.
+
+### 15:00 — Regressione backtest sui dati rigenerati
+
+**Cosa:** rilanciati i backtest per confermare che i numeri documentati si riproducano sui
+dati ricreati da zero (BTC/ETH 15m, hold=3, fee 0.2% RT, leva 3x, pos 15%).
+**Reale:** accuratezze e drawdown **identici**, solo +1/+3 trade dalle barre recenti in più.
+
+| Strategia | Ottenuto | Documentato | Esito |
+|---|---|---|---|
+| SQ01 BTC 15m | 76.7% / DD 6.7% / 4063t | 76.7% / 6.7% / 4062 | ✓ |
+| SQ01 ETH 15m | 76.4% / 6.2% / 2951t | 76.4% / 6.2% / 2948 | ✓ |
+| SQ02 BTC 15m | 79.7% / 6.5% / 1251t | 79.7% / 6.5% / 1250 | ✓ |
+| SQ02 ETH 15m | 78.6% / 3.4% / 944t | 78.6% / 3.4% / 942 | ✓ |
+| **MT01 BTC 15m (ema20+vol)** | **82.7% / 5.9% / 503t** | 82.7% / 5.9% / 503 | ✓ esatto |
+| MT01 ETH 15m (ema20+vol) | 81.2% / 2.9% / 404t | — | ok |
+
+**Lezione:** l'integrità dei dati rigenerati è confermata — la pipeline di download produce
+risultati riproducibili. La config live di MT01 (ema20+vol) coincide col best documentato.
+
+### Punti aperti
+
+1. **Backtest e drift dati:** MT01 live ora è immune (1h da Cerbero), ma i backtest girano
+   sempre sui dati fino all'ultimo `download_all()`. Per dati di backtest sempre freschi
+   serve uno scheduling del download (cron/job).
+2. **Healthcheck:** valutare un check su mtime di `status.json` (< 180s) per rilevare uno
+   stallo del loop, non solo l'esistenza del file.
+
+---
+
+### 23:00 — 3 nuove strategie con edge OOS fee-aware (branch `strategy_free`)
+
+**Obiettivo:** trovare almeno 3 nuove strategie (oltre MR01), edge netto validato
+out-of-sample e fee-aware, per il target €1.000 → ~€50/giorno.
+
+**Metodologia (invariata dalla lezione squeeze):** ingresso eseguibile a `close[i]`
+(nessun look-ahead), backtest netto dopo fee Deribit 0.10% RT + leva 3x, OOS = ultimo
+30% held-out, robustezza su griglia parametri + sweep fee 0.00–0.20% RT, exit
+TP/SL intrabar o time-limit, una posizione per volta, capitale composto.
+
+**Candidati** (`scripts/analysis/strategy_research_v2.py`), tutti mean-reversion
+(l'edge è sempre il rientro, mai la continuazione):
+
+| Candidato | Esito | Motivo |
+|---|---|---|
+| **MR02 Donchian Fade** | ✅ | Robusto su tutta la griglia `n × sl_atr` e tutte le fee |
+| **MR03 Keltner Fade** | ✅ | Robusto su tutta la griglia `n × k`; banda ATR, indipendente da Bollinger |
+| **MR07 Return Reversal** | ✅ | Intero blocco `tp_atr=2.0` positivo full+OOS; esposizione ~8% |
+| MR04 Z-score Reversion | ⛔ | Robusto ma è MR01 riparametrizzato (stessa banda std): edge non *nuovo* |
+| MR05 Bollinger + filtro ADX | ⛔ | Non robusto: negativo su gran parte della griglia BTC |
+| MR06 RSI(2) Connors | ⛔ | ETH 1h negativo; non robusto su entrambi gli asset |
+
+**Risultati** (netto 0.10% RT, leva 3x, OOS, 1h):
+
+| Codice | Meccanismo | BTC OOS | ETH OOS | DD (full) |
+|---|---|---|---|---|
+| MR02 | estremi canale Donchian H/L | +172% | enorme | 30% / 42% |
+| MR03 | canale ATR su EMA | +112% | +886% | 37% / 66% |
+| MR07 | z dei rendimenti di barra | +105% | +195% | 25% / 46% |
+
+**Validazione live-path** (`oos_validation.py`, legge `strategies.yml`, exit hold
+del worker): tutte e tre positive netto OOS su tutto lo sweep fee, anche al
+pessimistico 0.20% RT → edge robusto pure al meccanismo di exit.
+
+**Verifiche:** equivalenza esatta backtest produzione vs research engine (MR02 BTC:
+2039 trade, DD 29% identici); le 3 classi si caricano dal `strategy_loader`;
+aggiunte a `strategies.yml` (BTC+ETH 1h). Nessuna suite di test nel progetto.
+
+**Onestà sul target:** con 4 fade indipendenti × 2 asset il PnL storico aggregato
+supera €50/giorno, ma sono backtest a leva 3x su 8 anni con annate eccezionali
+(ETH 2024). Plausibile ma da confermare col paper trader live prima del capitale reale.
+DD alto su ETH (MR03 ~66%, come MR01) → leva più bassa consigliata per quell'asset.
+
+**File:** `strategy_research_v2.py`, `src/strategies/fade_base.py`,
+`scripts/strategies/MR0{2,3,7}_*.py` (nuovi); `strategy_loader.py`, `strategies.yml`,
+`CLAUDE.md` (aggiornati).
+**Lezione confermata:** ogni edge robusto trovato finora è mean-reversion; ogni
+variante trend/continuation o oscillatore senza filtro perde netto.
+
+---
+
+### 23:45 — Aumentare Acc e ridurre DD (filtro trend + portafoglio)
+
+**Obiettivo:** alzare accuratezza e abbassare drawdown sulle 4 fade, senza
+distruggere l'edge né overfittare (ogni leva misurata FULL **e** OOS).
+
+**Diagnosi:** perdite/DD concentrati 2018–2021 (bear/covid/caos vol), su ETH DD
+pieno 66–71%. Banco di prova: `scripts/analysis/risk_improvements.py` e
+`risk_portfolio.py`.
+
+**Leve testate:**
+
+| Leva | Esito | Motivo |
+|---|---|---|
+| Sizing vol-target (size ∝ 1/dist-SL) | ⛔ | Over-size sui trade a stop stretto → DD su, ritorno giù |
+| Skip alta volatilità (ATR% in coda alta) | ⛔ | L'alta vol è *positiva* per le fade (più reversione): Acc e ritorno giù |
+| **Filtro trend** (`\|close−EMA200\|/ATR > soglia` → salta) | ✅ | Non fada trend/crolli estremi: Acc↑ ovunque, DD↓ molto su ETH, OOS regge |
+| **Portafoglio** equipesato (sotto-conti indipendenti) | ✅ | Curve poco correlate → DD aggregato 14% (full)/10% (OOS) vs 20-70% singolo |
+
+**Filtro trend — sweep soglia** (assoluta in ATR, regola unica per tutte = niente
+overfit): 3.0 ATR è l'equilibrio (2.0 taglia troppo ritorno). Effetto su config
+deployata (base → filtro):
+
+| Sleeve | Acc | DD |
+|---|---|---|
+| MR01 ETH | 46→55 | **71→26** |
+| MR02 ETH | 49→55 | 42→25 |
+| MR03 ETH | 49→52 | 66→34 |
+| MR07 ETH | 48→54 | 46→21 |
+| MR01 BTC | 51→54 | 32→34* |
+| MR02 BTC | 48→52 | 29→23 |
+| MR07 BTC | 49→53 | 25→18 |
+| MR03 BTC | 47→47 | 37→37 (filtro OFF) |
+
+\*MR01 BTC: DD full +2pt ma Acc +3.7 e DD OOS piatto (14.8→15.0). **MR03 BTC**:
+il filtro peggiora entrambe (unico sleeve) → lasciato disattivo nello yaml.
+
+**Implementazione:** helper `trend_distance()` in `fade_base.py`; param opzionali
+`trend_max`/`ema_long` (default None = retro-compatibile) in tutte le strategie
+(MR01/02/03/07); `strategies.yml` con `trend_max: 3.0, ema_long: 200` (eccetto
+MR03 BTC). Verificato: equivalenza produzione vs ricerca.
+
+**Lezione:** il modo onesto di ridurre il DD non è strozzare il sizing (peggiora),
+ma (a) non opporsi a trend estremi e (b) diversificare su strategie scorrelate.

docs/diary/2026-05-29-exploration.md

@@ -0,0 +1,155 @@
+# Diario — 2026-05-29 — Esplorazione di nuove famiglie di strategie
+
+## Obiettivo
+
+Trovare 5-10 nuove famiglie di strategie, diverse da quelle esistenti, migliori o
+complementari, con DD basso e attenzione alle fee. Esplorazione onesta (no
+look-ahead, netto fee, OOS) condotta con **agenti paralleli**, ognuno su una famiglia
+indipendente, tutti sullo stesso harness condiviso (`scripts/analysis/explore_lab.py`).
+Lavoro sul branch `strategy_explore`.
+
+## Famiglie esplorate (9) ed esito onesto
+
+| Famiglia | Esito | Note |
+|---|---|---|
+| **Pairs / spread reversion** | ✅ **VINCITORE** | Market-neutral, genuinamente nuova, decorrelata |
+| **TSMOM multi-orizzonte** | ✅ diversificatore | Marginale ma distinto (corr 0.53 con ROT02), DD basso |
+| Stagionalità settimanale | ⚠️ marginale/fragile | "Mercoledì-long-24h" 7/8 asset OOS+ ma effetto concentrato a 00:00 UTC |
+| Vol-target BTC | ⚠️ marginale | Sharpe 0.94 vs 0.76 buy&hold, DD ancora 44% |
+| Stagionalità intraday (ora) | ❌ rumore | L'edge orario muore sotto le fee |
+| Stagionalità mensile/turn-of-month | ❌ rumore | Reale in-sample, morto OOS dal 2024 |
+| Cross-sectional reversal | ❌ nessun edge | Perde vs equal-weight, corr 0.98 col momentum |
+| Opening-range breakout | ❌ non generalizza | Solo BTC/ETH, alcuni regimi, fee-fragile |
+| Lead-lag BTC→alt | ❌ nessun edge | Reazione contemporanea (corr lag+1 ≈ 0), non batte buy&hold |
+| Momentum/continuation intraday | ❌ negativo | Conferma: il *fade* (mean-reversion) domina |
+
+7 famiglie su 9 sono rumore — e l'harness le ha rifiutate senza produrre falsi
+positivi (segnale che la metodologia onesta funziona). Due edge reali emergono.
+
+## Vincitore 1 — PAIRS (market-neutral) — `PR01_pairs_reversion.py`
+
+Scommette sul rientro del log-ratio di due cripto verso la media (z-score). Quando
+`z ≤ −2` → long A / short B; `z ≥ +2` → l'opposto; esce al rientro (`|z| ≤ 0.5`) o a
+tempo. Engine onesto verificato in `pairs_research.py` (test esplicito no-look-ahead:
+`z[i]` invariato perturbando il futuro). Fee contate su **2 gambe** (0.20% RT/coppia).
+
+Validazione (netto, leva 3x, OOS = ultimo 30%, 1h):
+
+| Coppia | CAGR | Sharpe | OOS DD | anni+ |
+|---|--:|--:|--:|--:|
+| ETH/BTC | 144% | 4.04 | 17% | 8/9 |
+| LTC/ETH | 71% | 2.52 | 10% | 7/8 |
+| ADA/ETH | 77% | 2.16 | 11% | 7/8 |
+
+Tutte le 10 coppie testate positive FULL+OOS, regge fee 0.40% RT/coppia, correlazione
+col mercato ~0.02 (market-neutral confermato). DD pieno 42-49% (alto), ma OOS DD
+10-17% (buono) e soprattutto **quasi-zero correlazione** col resto → diversificatore
+eccezionale. Limite: 2 gambe (long+short), il worker live va esteso prima del live.
+
+## Vincitore 2 — TSM01 (TSMOM multi-orizzonte) — `tsmom_research.py`
+
+Long-only multi-crypto: tiene equal-weight gli asset con consenso pieno del segno di
+momentum su 3/6/12 mesi, cash se BTC<SMA100. Distinto da ROT02 (persistenza assoluta
+vs ranking relativo), corr 0.53. FULL +169% / OOS +80% / DD 22% / Sharpe 1.07,
+**mai un anno negativo**, regge fee 0.40%. Verificato no-look-ahead (cheat-test
+esplode a +575%). Marginale come stand-alone (rende meno di ROT02) ma utile in ensemble.
+
+## Il payoff — combinare le nuove fonti col MASTER (`combine_v2.py`)
+
+Le nuove sleeve sono quasi scorrelate col MASTER-9 (pairs ~0.02-0.08, TSM01 0.05).
+Aggiungerle migliora nettamente il portafoglio:
+
+| Portafoglio | CAGR | DD% | Sharpe | OOS DD% | OOS Sharpe |
+|---|--:|--:|--:|--:|--:|
+| MASTER-9 (base) | 47 | 5.2 | 4.23 | 4.7 | 4.33 |
+| **MASTER + pairs (12)** | **66** | **3.8** | **5.67** | **3.3** | **6.86** |
+| MASTER + TSM01 (10) | 44 | 4.7 | 4.21 | 4.2 | 4.33 |
+| MASTER esteso (13) | 62 | 3.6 | 5.66 | 3.0 | 6.79 |
+
+I **pairs** sono l'aggiunta decisiva: alzano la CAGR (47→66), **abbassano il DD**
+(5.2→3.8 full, 4.7→**3.3** OOS) e portano lo Sharpe OOS a **6.86** — il free-lunch
+della diversificazione da una fonte market-neutral scorrelata. TSM01 contribuisce
+poco (diluisce il ritorno) ma abbassa lievemente il DD.
+
+## Caveat onesti
+
+- I pairs hanno DD pieno alto (42-49%) sull'1h; il vantaggio sta nella decorrelazione,
+  non nel DD stand-alone. Richiedono esecuzione a 2 gambe (short del perp B) — da
+  verificare shortabilità/liquidità sugli alt e raddoppio fee nel worker.
+- Sharpe combinati 5-7 e CAGR 60%+ sono backtest a leva 3x su finestra 2021-2026 con
+  OOS ~1.6 anni e il 2024 cripto eccezionale: numeri ottimistici, da confermare in
+  paper trading live.
+- TSMOM e le strategie honest condividono l'overlay risk-off SMA100: parte della loro
+  difensività è comune (non perfettamente indipendente).
+
+## Terza ondata — espansione dei meccanismi provati + 2 nuovi sondaggi
+
+Esplorate altre 4 direzioni con agenti paralleli:
+- **Fade su 6 nuovi alt (ADA/BNB/DOGE/LTC/SOL/XRP)**: 0 robuste. La mean-reversion
+  fade vive solo su BTC/ETH (liquidi); sugli alt sparisce o è artefatto di pochi pump
+  (DOGE). Coerente con la lezione del progetto.
+- **Espansione PAIRS** (tutte le 28 coppie): trovate **3 nuove coppie robuste** →
+  BTC/LTC (robusta 1h *e* 4h, Sharpe 2.21, DD 24-34%, concentrazione PnL 9%), ETH/SOL
+  e BNB/ETH (Sharpe 2.4+, solo 1h). Pattern: sempre alt-liquido vs major, mai alt/alt.
+  PR01 ora ha **6 coppie**.
+- **Low-volatility anomaly**: ❌ in cripto è INVERTITA (vince l'alta vol = alta beta),
+  ridondante con EW+risk-off/ROT02. L'anti-test high-vol stravince.
+- **Confluenza multi-timeframe (fade 1h confermato da 4h)**: non crea edge nuovo e non
+  migliora lo Sharpe, ma **dimezza il DD** di MR01 (ETH: stesso Sharpe 3.17 a DD 38% vs
+  63%) e stabilizza l'OOS → utile variante low-DD, non strategia indipendente.
+
+## Bilancio finale e MASTER esteso (6 pairs)
+
+Robusti deployabili: **famiglia PAIRS (6 coppie) + TSM01** (+ confluenza MTF come variante
+low-DD di MR01, + tilt stagionale mercoledì marginale). I 6 pairs sono quasi scorrelati col
+MASTER (corr 0.02-0.08). MASTER + 6 pairs:
+
+| Portafoglio | CAGR | DD% | Sharpe | OOS DD% | OOS Sharpe |
+|---|--:|--:|--:|--:|--:|
+| MASTER-9 (base) | 47 | 5.2 | 4.23 | 4.7 | 4.33 |
+| **MASTER + 6 pairs (15)** | **71** | 5.7 | **5.93** | **2.3** | **7.71** |
+| MASTER esteso +TSM01 (16) | 67 | 5.4 | 5.95 | **2.2** | 7.67 |
+
+Aggiungere i 6 pairs porta l'**OOS DD a 2.2-2.3%** (da 4.7%) con Sharpe OOS ~7.7 e tutti
+gli anni positivi: il guadagno di diversificazione da fonti market-neutral scorrelate.
+
+## Quarto giro — validazione anti-overfitting e irrobustimento
+
+Tre audit scettici paralleli (walk-forward, plateau, stress, scomposizione):
+
+**Pairs — de-overfittati.** Sostituita la config per-coppia (cherry-picking di z_exit/n)
+con **una config universale `n=50 z_in=2.0 z_exit=0.75 max_bars=72`**. Verifiche:
+- plateau (non picco): heatmap n×z_in → 20/20 celle Sharpe>1 su ETH/BTC e BTC/LTC;
+- walk-forward (train 2y / test 6m rolling): ETH/BTC 11/12 finestre positive, BTC/LTC
+  9/10 → edge distribuito su tutta la storia, non un regime singolo;
+- **BNB/ETH scartata** (era robusta solo coi suoi parametri → overfit; crolla con la
+  universale e muore per prima allo stress costi). Famiglia ridotta a **5 pairs**.
+- stress: 5/6 reggono fee+slippage realistici; solo ETH/BTC regge fee 6x (coda fee-fragile).
+
+**Master — numeri sobri.** L'OOS Sharpe 7.7 / DD 2.3% è **ottimistico ~50%** perché l'OOS
+cade nel bull calmo 2024-25. Numeri onesti da usare:
+- worst-DD su finestra mobile 90g (2021-2026) = **5.7%** (bear FTX) → budget DD ~6%, non 2.3%;
+- Sharpe per-semestre: mediana **~5** (min 1.2, max 12) → atteso ~5, non 7.7;
+- ogni anno e ogni semestre dal 2021 positivo (anche il 2022 bear, grazie alle gambe short);
+- equal-weight ≈ inverse-vol (non dipende da pesi fortunati);
+- regge **leva 2x + slippage doppio** (CAGR 36%, Sharpe 5.1);
+- **rischio concentrato: i pairs portano ~57% del rischio** → cap consigliato ~30-35%.
+- Config robusta raccomandata: **MASTER-esteso, equal-weight, leva 2x, cap pairs ~30-35%**.
+
+**TSM01 — confermato robusto** (36/36 config OOS+, walk-forward stabile) ma corr reale con
+ROT02 = **0.62** (non 0.53), e gran parte del DD basso viene dall'overlay risk-off condiviso.
+Tenuto come diversificatore con **gross 0.30** (stesso Sharpe, DD 22%→15%).
+
+**Confluenza multi-TF — SCARTATA: era overfit.** Taglia il 97% dei trade (restano ~40 in
+8 anni = non significativo), distrugge lo Sharpe (1.58→0.27 su BTC) e il caso "bello" non
+sopravvive alle perturbazioni. Per abbassare il DD di MR01 meglio ridurne la leva, non il filtro 4h.
+
+**Risultato del giro:** quanto trovato regge l'esame anti-overfit (NON è l'errore squeeze),
+ma i numeri vanno comunicati sobri (Sharpe ~5, DD ~6%) e con leva 2x + cap pairs. Famiglia
+pairs consolidata a 5 coppie con config universale; confluenza MTF rimossa dai vincitori.
+
+## File creati (branch strategy_explore)
+
+`scripts/analysis/explore_lab.py` (harness onesto condiviso), `pairs_research.py`
+(verifica + ricerca pairs), `tsmom_research.py` (TSM01), `combine_v2.py` (master
+esteso); `scripts/strategies/PR01_pairs_reversion.py` (artefatto pairs).

docs/diary/2026-05-29.md

@@ -0,0 +1,64 @@
+# Diario di ricerca — 2026-05-29
+
+## Combinare le strategie migliora i risultati?
+
+**Domanda:** usare insieme le due famiglie di strategie presenti sul repo migliora
+il profilo rischio/rendimento rispetto a usarle separatamente?
+
+- **FADE** (mie): reversione intraday 1h, long/short, BTC/ETH — MR01 Bollinger,
+  MR02 Donchian, MR03 Keltner, MR07 Return-reversal (tutte col filtro trend 3.0 ATR).
+- **HONEST** (altra sessione): long-only multi-regime multi-crypto — DIP01 (dip-buy
+  1h BTC), TR01 (EMA-trend 4h basket), ROT02 (dual-momentum rotation 1d).
+
+**Metodo** (`scripts/analysis/combine_portfolio.py`): per ogni sleeve si costruisce
+l'equity **giornaliera** normalizzata su un indice comune (2021-01-01 → 2026-05-26),
+si passa ai rendimenti giornalieri, si misura la correlazione cross-famiglia e si
+confrontano i portafogli equal-weight (ribilanciati ogni giorno), 50/50 fra famiglie
+e inverse-vol. Metriche FULL e OOS (ultimo 30% della finestra comune, da 2024-10-12):
+ritorno, CAGR, max DD, Sharpe annualizzato. Le curve honest sono riusate da
+`honest_improve2.py`; quelle fade da `risk_management.build_trades`.
+
+**Correlazione:** cross-famiglia **+0.05** (quasi indipendenti). Intra-fade +0.18,
+intra-honest +0.05. L'unica coppia un po' correlata è MR01_BTC↔DIP01_BTC (+0.43),
+entrambe mean-reversion su BTC. Famiglie scorrelate ⇒ diversificazione quasi ideale.
+
+**Risultati (FULL | OOS):**
+
+| Portafoglio | Ret% | CAGR | DD% | Sharpe | oDD% | oSharpe |
+|---|---|---|---|---|---|---|
+| FADE only (8) | +549 | 41 | 8.6 | 3.75 | 5.4 | 4.14 |
+| HONEST only (3) | +642 | 45 | 12.0 | 1.90 | 6.5 | 2.23 |
+| **ALL equal-weight (11)** | +589 | 43 | 6.1 | **3.95** | 4.6 | **4.46** |
+| **ALL 50/50 famiglie** | +615 | 44 | **5.5** | 3.18 | **4.0** | 3.87 |
+| ALL inverse-vol | +483 | 39 | 5.8 | 3.97 | 4.6 | 4.02 |
+
+**Conclusione: sì, combinare conviene.**
+- DD crolla: combinato 5.5–6.1% full / 4.0–4.6% OOS, contro 8.6% (fade) e **12%**
+  (honest) da sole → drawdown ridotto del 35–50%.
+- Sharpe sale: combinato OOS **4.46** vs honest 2.23 (raddoppia) e batte pure fade (4.14).
+- CAGR resta ~43–44% (≈ media delle due famiglie) ma con metà del rischio: è il
+  "free lunch" della diversificazione fra sorgenti di edge scorrelate.
+- Best Sharpe = equal-weight degli 11 sleeve; best DD = 50/50 fra le due famiglie.
+
+**Caveat onesti:** la finestra comune è 2021–2026 (5.4 anni), OOS ~2024-10→oggi
+(1.6 anni) — pochi regimi. CAGR e Sharpe sono backtest a leva 3x; il 2024 cripto
+favorevole pesa. Il target €50/giorno resta vincolato dal capitale: 43% CAGR su
+€1000 non fa €50/giorno a breve, serve compounding pluriennale o più capitale.
+Prossimo passo: confermare il portafoglio combinato nel paper trader live.
+
+**File:** `scripts/analysis/combine_portfolio.py` (nuovo).
+
+## Pulizia roster + miglioria ROT02
+
+- **Waste delle peggiori:** MR03 Keltner (fade più debole, Sharpe 1.22, ridondante
+  con MR01 — rimuoverla *migliora* il portafoglio fade: DD 8.6→8.2, ret +549→+666)
+  e ROT01 (dominata da ROT02). Spostate in `scripts/waste/`.
+- **Portafogli pronti:** `PORT02_fade_master` (6 sleeve fade) e `PORT03_all_master`
+  (9 sleeve fade+honest, varianti equal/5050).
+- **ROT02 DD alto → migliorato:** la rotazione concentrava il book su 2 asset
+  (DD 40%). Sweep su `rot_improved`: `top_k=3` dimezza quasi il DD (40%→26%) e
+  *alza* il ritorno full (+1095→+1303%, ret/DD 27→50). Il vol-target abbassa il DD
+  ma sacrifica ritorno (de-leverage) → tenuto top_k=3 senza VT. Caveat onesto:
+  l'OOS di ROT02 cala un po' (+98→+68%, DD 12→14%), ma il MASTER (config deployata)
+  migliora lo Sharpe full 3.95→4.23. Applicato a `ROT02_dual_momentum.py` e
+  `_rot_daily_equity`. Sweep in `honest_improve.rot_improved`.

docs/specs/2026-05-27-multi-strategy-paper-trader.md

@@ -0,0 +1,174 @@
+# Multi-Strategy Paper Trader — Design Spec
+
+## Obiettivo
+
+Eseguire N strategie di trading in parallelo su Deribit testnet (paper trading locale), ognuna con capitale virtuale indipendente di €1000 USDC. Lo storico trade di ogni strategia persiste tra restart. Nuove strategie aggiungibili in corso d'opera via config YAML senza perdere lo storico delle esistenti.
+
+## Architettura
+
+Un singolo container Docker esegue un orchestratore (`MultiStrategyRunner`) che gestisce N `StrategyWorker`. Ogni worker è indipendente: proprio capital, propri trade, proprio stato.
+
+```
+Docker Container
+├── MultiStrategyRunner (orchestratore, loop principale)
+│   ├── StrategyWorker[SQ02_BTC_15m] → paper trade → JSONL
+│   ├── StrategyWorker[ML01_ETH_15m] → paper trade → JSONL
+│   └── ...altri worker da YAML
+├── CerberoClient (condiviso, fetch prezzi)
+└── TelegramNotifier (condiviso)
+```
+
+## Componenti
+
+### 1. `strategies.yml` — Configurazione
+
+```yaml
+defaults:
+  capital: 1000
+  position_size: 0.15
+  leverage: 3
+  hold_bars: 3
+  poll_seconds: 60
+  retrain_hours: 24
+
+strategies:
+  - name: SQ02_antifake_vol
+    asset: BTC
+    tf: 15m
+    enabled: true
+
+  - name: SQ02_antifake_vol
+    asset: ETH
+    tf: 15m
+    enabled: true
+
+  - name: ML01_squeeze_gbm
+    asset: ETH
+    tf: 15m
+    enabled: true
+    position_size: 0.20
+    params:
+      ml_threshold: 0.70
+      bb_window: 14
+      sq_threshold: 0.8
+```
+
+Ogni entry eredita `defaults`. Override per-strategia possibile su tutti i campi. Il campo `params` passa kwargs a `generate_signals()` o al backtest ML.
+
+### 2. `StrategyWorker` — Worker per singola strategia
+
+Responsabilità:
+- Importa la classe Strategy corrispondente da `scripts/strategies/`
+- Mantiene stato: capital, posizione aperta, equity
+- Al startup: ricarica `status.json` se esiste (resume), altrimenti inizia da zero
+- Ad ogni tick: riceve DataFrame candele, genera segnali, paper-trade
+- Logga ogni evento in `trades.jsonl` (append-only)
+- Aggiorna `status.json` ad ogni tick
+
+Stato persistente (`status.json`):
+```json
+{
+  "capital": 1023.45,
+  "in_position": true,
+  "direction": "long",
+  "entry_price": 2534.20,
+  "entry_time": "2026-05-27T14:30:00Z",
+  "bars_held": 1,
+  "total_trades": 15,
+  "total_wins": 12,
+  "started_at": "2026-05-27T10:00:00Z"
+}
+```
+
+Trade log (`trades.jsonl`), append-only:
+```json
+{"ts": "2026-05-27T14:30:00Z", "event": "OPEN", "direction": "long", "price": 2534.20, "size": 0.18, "capital": 1023.45}
+{"ts": "2026-05-27T15:15:00Z", "event": "CLOSE", "reason": "hold_limit", "entry": 2534.20, "exit": 2560.10, "pnl": 3.45, "fee": 0.92, "net_pnl": 2.53, "capital": 1025.98}
+```
+
+### 3. `MultiStrategyRunner` — Orchestratore
+
+Loop principale:
+1. Carica `strategies.yml`
+2. Per ogni entry, crea `StrategyWorker` (o riprende se già esiste)
+3. Ogni 60s:
+   a. Fetch candele live da Cerbero (una volta per asset/tf unico)
+   b. Passa DataFrame a ogni worker
+   c. Ogni worker valuta segnali e gestisce posizione
+   d. Worker ML: retrain ogni 24h
+4. Notifica Telegram per ogni trade
+
+Ottimizzazione: fetch candele raggruppato per (asset, tf). Se 3 strategie usano BTC 15m, fetch una volta sola.
+
+### 4. Persistenza
+
+```
+data/paper_trades/
+  SQ02_antifake_vol__BTC__15m/
+    trades.jsonl
+    status.json
+  SQ02_antifake_vol__ETH__15m/
+    trades.jsonl
+    status.json
+  ML01_squeeze_gbm__ETH__15m/
+    trades.jsonl
+    status.json
+```
+
+Directory naming: `{strategy_name}__{asset}__{tf}` con double underscore separatore.
+
+Volume Docker: `./data:/app/data` — persiste tra restart.
+
+### 5. Aggiunta strategia in corso
+
+1. Aggiungi entry in `strategies.yml`
+2. `docker compose restart`
+3. Runner carica YAML, trova nuova entry senza `status.json` → parte da €1000
+4. Strategie esistenti riprendono da `status.json` → storico intatto
+
+### 6. Docker
+
+`Dockerfile` — invariato, aggiunge `strategies.yml` alla COPY.
+
+`docker-compose.yml`:
+```yaml
+services:
+  paper-trader:
+    build: .
+    container_name: pythagoras-multi
+    restart: unless-stopped
+    volumes:
+      - ./data:/app/data
+      - ./strategies.yml:/app/strategies.yml:ro
+    env_file:
+      - .env
+    environment:
+      - PYTHONUNBUFFERED=1
+```
+
+`CMD` cambia a: `uv run python -m src.live.multi_runner`
+
+### 7. Strategia-specifica: ML01
+
+ML01 richiede training del modello GBM. Il worker ML01:
+- Al primo avvio: train su storico (365 giorni via Cerbero)
+- Ogni `retrain_hours`: retrain
+- Usa `SignalEngine` esistente per check_signal()
+- Le strategie SQ* non hanno training — solo regole deterministiche
+
+### 8. File da creare/modificare
+
+Nuovi:
+- `src/live/multi_runner.py` — orchestratore
+- `src/live/strategy_worker.py` — worker per singola strategia
+- `strategies.yml` — config
+- `src/live/strategy_loader.py` — import dinamico classi Strategy
+
+Modifiche:
+- `docker-compose.yml` — nuovo CMD, volume strategies.yml
+- `Dockerfile` — COPY strategies.yml
+
+Invariati:
+- `src/live/cerbero_client.py`
+- `src/live/telegram_notifier.py`
+- `src/live/signal_engine.py` (usato da ML01 worker)

pyproject.toml

@@ -14,6 +14,7 @@ dependencies = [
     "torch>=2.0",
     "matplotlib>=3.7",
     "tqdm>=4.65",
+    "pyyaml>=6.0",
 ]
 
 [project.optional-dependencies]

scripts/12_final_report.py

@@ -1,298 +0,0 @@
-"""Report finale: TOP 5 metodi + simulazione crescita capitale €1000 → €50/giorno."""
-from __future__ import annotations
-import sys
-sys.path.insert(0, ".")
-
-import numpy as np
-from src.data.downloader import load_data
-
-print("=" * 70)
-print("  REPORT FINALE — TOP 5 METODI")
-print("  Target: accuracy >80%, ROI annuo >30%, €50/giorno da €1000")
-print("=" * 70)
-
-# Metodo 1: Squeeze Breakout ETH 1h (BBw=20, sqThr=0.8, volume confirmed)
-# Metodo 2: Squeeze Breakout ETH 1h (BBw=30, sqThr=0.9, senza vol filter)
-# Metodo 3: Squeeze Breakout BTC+ETH combinato
-# Metodo 4: Squeeze Breakout 15m (alta frequenza)
-# Metodo 5: GBM Structural + Squeeze filter (ibrido ML + strutturale)
-
-FEE = 0.001
-LEVERAGE = 3
-INITIAL = 1000
-
-
-def bollinger_bandwidth(close, window=20):
-    n = len(close)
-    result = np.full(n, np.nan)
-    for i in range(window, n):
-        w = close[i-window:i]
-        ma = np.mean(w)
-        std = np.std(w)
-        if ma > 0:
-            result[i] = (2 * 2 * std) / ma
-    return result
-
-
-def keltner_ratio(close, high, low, window=20):
-    n = len(close)
-    result = np.full(n, np.nan)
-    for i in range(window, n):
-        wc = close[i-window:i]
-        wh = high[i-window:i]
-        wl = low[i-window:i]
-        ma = np.mean(wc)
-        bb_std = np.std(wc)
-        tr = np.maximum(wh - wl, np.maximum(np.abs(wh - np.roll(wc,1)), np.abs(wl - np.roll(wc,1))))
-        atr = np.mean(tr[1:])
-        kc_r = (ma + 1.5*atr) - (ma - 1.5*atr)
-        bb_r = (ma + 2*bb_std) - (ma - 2*bb_std)
-        if kc_r > 0:
-            result[i] = bb_r / kc_r
-    return result
-
-
-def run_squeeze_backtest(close, high, low, volume, bb_w, sq_thr, brk_bars, vol_filter, split_pct=0.7, leverage=3, pos_pct=0.2):
-    n = len(close)
-    split = int(n * split_pct)
-    kcr = keltner_ratio(close, high, low, bb_w)
-
-    in_sq = False
-    sq_start = 0
-    capital = float(INITIAL)
-    equity = [capital]
-    trades = []
-
-    for i in range(bb_w + 1, n):
-        if np.isnan(kcr[i]):
-            continue
-        is_sq = kcr[i] < sq_thr
-        if is_sq and not in_sq:
-            in_sq = True
-            sq_start = i
-        elif not is_sq and in_sq:
-            in_sq = False
-            duration = i - sq_start
-            if duration < 5 or i < split or i + brk_bars >= n:
-                continue
-
-            # Volume check
-            if vol_filter:
-                avg_v = np.mean(volume[sq_start:i])
-                brk_v = np.mean(volume[i:i+brk_bars])
-                if avg_v > 0 and brk_v < avg_v * 1.3:
-                    continue
-
-            first_ret = (close[i] - close[i-1]) / close[i-1]
-            if abs(first_ret) < 0.001:
-                continue
-
-            direction = 1 if first_ret > 0 else -1
-            actual = (close[i+brk_bars-1] - close[i-1]) / close[i-1]
-            is_correct = (direction == 1 and actual > 0) or (direction == -1 and actual < 0)
-
-            trade_ret = actual * direction
-            net = trade_ret * leverage - FEE * 2 * leverage
-            pnl = capital * pos_pct * net
-            capital += pnl
-            capital = max(capital, 0)
-            equity.append(capital)
-
-            trades.append({
-                "correct": is_correct,
-                "actual_ret": actual,
-                "net_pnl": pnl,
-                "capital_after": capital,
-            })
-
-    if not trades:
-        return None
-
-    correct = sum(1 for t in trades if t["correct"])
-    acc = correct / len(trades) * 100
-    total_ret = (capital - INITIAL) / INITIAL * 100
-    test_candles = n - split
-    test_days = test_candles / 24
-    test_years = test_days / 365.25
-    ann = ((capital / INITIAL) ** (1/test_years) - 1) * 100 if test_years > 0 and capital > 0 else -100
-    daily_pnl = (capital - INITIAL) / test_days if test_days > 0 else 0
-
-    peak = equity[0]
-    max_dd = 0
-    for v in equity:
-        if v > peak: peak = v
-        dd = (peak - v) / peak if peak > 0 else 0
-        max_dd = max(max_dd, dd)
-
-    return {
-        "trades": len(trades),
-        "accuracy": acc,
-        "total_return": total_ret,
-        "annualized": ann,
-        "max_drawdown": max_dd * 100,
-        "final_capital": capital,
-        "daily_pnl": daily_pnl,
-        "trades_per_year": len(trades) / test_years if test_years > 0 else 0,
-    }
-
-
-methods = []
-
-# --- Metodo 1: ETH 1h, BBw=20, sqThr=0.8, vol confirmed ---
-df_eth = load_data("ETH", "1h")
-r1 = run_squeeze_backtest(df_eth["close"].values, df_eth["high"].values, df_eth["low"].values, df_eth["volume"].values,
-                           bb_w=20, sq_thr=0.8, brk_bars=3, vol_filter=True)
-methods.append(("M1: ETH 1h Squeeze+Vol (BBw=20,sq=0.8)", r1))
-
-# --- Metodo 2: ETH 1h, BBw=30, sqThr=0.9, no vol ---
-r2 = run_squeeze_backtest(df_eth["close"].values, df_eth["high"].values, df_eth["low"].values, df_eth["volume"].values,
-                           bb_w=30, sq_thr=0.9, brk_bars=3, vol_filter=False)
-methods.append(("M2: ETH 1h Squeeze (BBw=30,sq=0.9)", r2))
-
-# --- Metodo 3: BTC+ETH combinato ---
-df_btc = load_data("BTC", "1h")
-r3a = run_squeeze_backtest(df_btc["close"].values, df_btc["high"].values, df_btc["low"].values, df_btc["volume"].values,
-                            bb_w=14, sq_thr=0.8, brk_bars=3, vol_filter=False, pos_pct=0.1)
-r3b = run_squeeze_backtest(df_eth["close"].values, df_eth["high"].values, df_eth["low"].values, df_eth["volume"].values,
-                            bb_w=20, sq_thr=0.8, brk_bars=3, vol_filter=False, pos_pct=0.1)
-
-if r3a and r3b:
-    combined_trades = r3a["trades"] + r3b["trades"]
-    combined_correct = int(r3a["accuracy"]/100 * r3a["trades"]) + int(r3b["accuracy"]/100 * r3b["trades"])
-    combined_acc = combined_correct / combined_trades * 100 if combined_trades > 0 else 0
-
-    # Simulate portfolio
-    cap = float(INITIAL)
-    # Rough estimate: alternate between assets
-    for r in [r3a, r3b]:
-        ret_per_trade = r["total_return"] / 100 / r["trades"] if r["trades"] > 0 else 0
-        for _ in range(r["trades"]):
-            cap *= (1 + ret_per_trade * 0.5)
-
-    r3 = {
-        "trades": combined_trades,
-        "accuracy": combined_acc,
-        "total_return": (cap - INITIAL) / INITIAL * 100,
-        "annualized": r3a["annualized"] * 0.5 + r3b["annualized"] * 0.5,
-        "max_drawdown": max(r3a["max_drawdown"], r3b["max_drawdown"]),
-        "final_capital": cap,
-        "daily_pnl": r3a["daily_pnl"] + r3b["daily_pnl"],
-        "trades_per_year": r3a["trades_per_year"] + r3b["trades_per_year"],
-    }
-    methods.append(("M3: BTC+ETH 1h Portafoglio Squeeze", r3))
-
-# --- Metodo 4: BTC 15m alta frequenza ---
-df_btc_15 = load_data("BTC", "15m")
-r4 = run_squeeze_backtest(df_btc_15["close"].values, df_btc_15["high"].values, df_btc_15["low"].values, df_btc_15["volume"].values,
-                           bb_w=14, sq_thr=0.9, brk_bars=3, vol_filter=True)
-methods.append(("M4: BTC 15m Squeeze+Vol alta freq", r4))
-
-# --- Metodo 5: ETH 1h squeeze aggressivo ---
-r5 = run_squeeze_backtest(df_eth["close"].values, df_eth["high"].values, df_eth["low"].values, df_eth["volume"].values,
-                           bb_w=20, sq_thr=0.8, brk_bars=3, vol_filter=False, leverage=3)
-methods.append(("M5: ETH 1h Squeeze aggressivo (no vol)", r5))
-
-# --- Print results ---
-print("\n")
-for i, (name, r) in enumerate(methods, 1):
-    if r is None:
-        print(f"  {name}: NO TRADES")
-        continue
-    print(f"  {'='*65}")
-    print(f"  #{i} — {name}")
-    print(f"  {'='*65}")
-    print(f"    Trades:           {r['trades']}")
-    print(f"    Accuracy:         {r['accuracy']:.1f}%  {'✅' if r['accuracy'] >= 80 else '⚠️' if r['accuracy'] >= 70 else '❌'}")
-    print(f"    Return totale:    {r['total_return']:+.1f}%")
-    print(f"    Return annuo:     {r['annualized']:+.1f}%  {'✅' if r['annualized'] >= 30 else '⚠️' if r['annualized'] >= 15 else '❌'}")
-    print(f"    Max Drawdown:     {r['max_drawdown']:.1f}%")
-    print(f"    Capitale finale:  €{r['final_capital']:.0f}")
-    print(f"    €/giorno media:   €{r['daily_pnl']:.2f}")
-    print(f"    Trades/anno:      {r['trades_per_year']:.0f}")
-    print()
-
-
-# --- Simulazione crescita 6 mesi ---
-print("\n" + "=" * 70)
-print("  SIMULAZIONE CRESCITA CAPITALE — 6 MESI")
-print("  Metodo: M1 (ETH 1h Squeeze+Vol) — il più preciso (83.9%)")
-print("=" * 70)
-
-# M1 params: ~87 trades in ~2.5 anni test = ~35 trades/anno = ~3 al mese
-# Accuracy: 83.9%, average return per trade with 3x leverage
-
-# Simulo con dati reali: prendo i trade dal test period
-close = df_eth["close"].values
-high = df_eth["high"].values
-low = df_eth["low"].values
-volume = df_eth["volume"].values
-n = len(close)
-split = int(n * 0.7)
-
-kcr = keltner_ratio(close, high, low, 20)
-in_sq = False
-sq_start = 0
-all_trade_rets = []
-
-for i in range(21, n):
-    if np.isnan(kcr[i]):
-        continue
-    is_sq = kcr[i] < 0.8
-    if is_sq and not in_sq:
-        in_sq = True
-        sq_start = i
-    elif not is_sq and in_sq:
-        in_sq = False
-        if i - sq_start < 5 or i < split or i + 3 >= n:
-            continue
-        avg_v = np.mean(volume[sq_start:i])
-        brk_v = np.mean(volume[i:i+3])
-        if avg_v > 0 and brk_v < avg_v * 1.3:
-            continue
-        first_ret = (close[i] - close[i-1]) / close[i-1]
-        if abs(first_ret) < 0.001:
-            continue
-        direction = 1 if first_ret > 0 else -1
-        actual = (close[i+2] - close[i-1]) / close[i-1]
-        trade_ret = actual * direction
-        all_trade_rets.append(trade_ret)
-
-avg_win = np.mean([r for r in all_trade_rets if r > 0]) if any(r > 0 for r in all_trade_rets) else 0
-avg_loss = np.mean([r for r in all_trade_rets if r <= 0]) if any(r <= 0 for r in all_trade_rets) else 0
-win_rate = sum(1 for r in all_trade_rets if r > 0) / len(all_trade_rets)
-
-print(f"\n  Statistiche trade:")
-print(f"    Win rate: {win_rate*100:.1f}%")
-print(f"    Avg win:  {avg_win*100:.2f}%")
-print(f"    Avg loss: {avg_loss*100:.2f}%")
-print(f"    Trades totali nel test: {len(all_trade_rets)}")
-print(f"    Trades/mese stimati: ~{len(all_trade_rets) / 30:.0f}")
-
-print(f"\n  Crescita simulata mese per mese (€1000 iniziali, leva 3x, 20% per trade):")
-capital = 1000.0
-monthly_trades = max(len(all_trade_rets) // 30, 3)
-
-# Shuffle trades to simulate different sequences
-np.random.seed(42)
-for month in range(1, 7):
-    n_trades = monthly_trades
-    month_rets = np.random.choice(all_trade_rets, size=n_trades, replace=True)
-
-    for ret in month_rets:
-        net = ret * LEVERAGE - FEE * 2 * LEVERAGE
-        capital += capital * 0.2 * net
-        capital = max(capital, 10)
-
-    daily_pnl = capital * 0.003  # stima conservativa 0.3% daily basata su performance
-    print(f"    Mese {month}: capitale €{capital:.0f}, €/giorno stima: €{daily_pnl:.1f}")
-
-print(f"\n  Capitale dopo 6 mesi: €{capital:.0f}")
-print(f"  €/giorno necessari: €50")
-print(f"  €/giorno ottenibili (0.5% daily su capitale): €{capital * 0.005:.1f}")
-
-if capital * 0.005 >= 50:
-    print(f"\n  ✅ TARGET RAGGIUNGIBILE: con €{capital:.0f} di capitale, 0.5% daily = €{capital*0.005:.0f}/giorno")
-else:
-    needed = 50 / 0.005
-    print(f"\n  ⚠️ Servono €{needed:.0f} di capitale per €50/giorno al 0.5% daily")
-    print(f"  Raggiungibile estendendo il periodo di crescita a ~{int(np.log(needed/1000) / np.log(1 + 0.15) + 0.5)} mesi")

scripts/analysis/combine_portfolio.py

@@ -0,0 +1,156 @@
+"""Studio: combinare TUTTE le strategie (fade + honest) migliora i risultati?
+
+Due famiglie con meccanismi e orizzonti diversi:
+  FADE (intraday 1h, long/short, BTC/ETH): MR01 boll, MR02 donchian, MR07
+       return-reversal — tutte col filtro trend 3.0 ATR. (MR03 keltner -> waste.)
+  HONEST (long-only, multi-regime, multi-crypto): DIP01 (dip-buy 1h BTC),
+       TR01 (EMA trend 4h basket), ROT02 (dual-momentum rotation 1d).
+
+Metodo: per ogni sleeve si costruisce l'equity GIORNALIERA normalizzata su un
+indice comune (2021-01-01 -> 2026-05-26), si passa ai rendimenti giornalieri,
+si misura la correlazione cross-famiglia e si confrontano i portafogli
+equal-weight (ribilanciati ogni giorno) e inverse-vol. Metriche FULL e OOS
+(ultimo 30% della finestra comune): ritorno, CAGR, max DD, Sharpe annualizzato.
+
+Tutto NETTO (fee gia' incluse nelle sleeve), leva 3x, pos 15% per sleeve.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from src.data.downloader import load_data
+from scripts.analysis.risk_management import strats_for, build_trades, INIT
+# curve daily honest gia' pronte nell'altra famiglia
+from scripts.analysis.honest_improve2 import (
+    _daily_equity, _norm, dip_market_gated, _tr_basket_daily, _rot_daily_equity,
+)
+
+IDX = pd.date_range("2021-01-01", "2026-05-26", freq="1D", tz="UTC")
+OOS_FRAC = 0.30
+SPLIT = int(len(IDX) * (1 - OOS_FRAC))          # confine OOS sulla finestra comune
+OOS_DATE = IDX[SPLIT].date()
+ANN = 365.0                                      # giorni/anno per annualizzare
+
+
+# ---------------- equity giornaliere ----------------
+def fade_daily_equity(asset: str, fn, params) -> pd.Series:
+    """Equity giornaliera di uno sleeve fade: trade 1h (filtro trend 3.0) -> equity -> daily."""
+    df = load_data(asset, "1h")
+    ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+    trades = build_trades(fn(df, **params), df, trend_max=3.0)
+    n = len(df); eq = np.full(n, INIT, dtype=float); cap = INIT
+    for i, j, ret in sorted(trades, key=lambda t: t[1]):
+        cap = max(cap + cap * 0.15 * ret, 10.0)
+        eq[j:] = cap
+    s = pd.Series(eq, index=ts).resample("1D").last().reindex(IDX).ffill().bfill()
+    return _norm(s)
+
+
+def build_all_sleeves() -> dict[str, pd.Series]:
+    sleeves: dict[str, pd.Series] = {}
+    # --- FADE: 8 sleeve ---
+    for asset in ["BTC", "ETH"]:
+        for nm, (fn, params) in strats_for(asset).items():
+            sleeves[f"{nm}_{asset}"] = fade_daily_equity(asset, fn, params)
+    # --- HONEST: 3 sleeve (riuso le funzioni dell'altra famiglia) ---
+    d = dip_market_gated("BTC", market_n=0, return_equity=True)
+    sleeves["DIP01_BTC"] = _norm(_daily_equity(d["eq_ts"], d["eq_v"], IDX))
+    sleeves["TR01_basket"] = _norm(_tr_basket_daily(["BNB", "BTC", "DOGE", "SOL", "XRP"], IDX))
+    sleeves["ROT02_rot"] = _norm(_rot_daily_equity(IDX))
+    return sleeves
+
+
+# ---------------- metriche ----------------
+def metrics(daily_ret: pd.Series, lo: int = 0, hi: int | None = None) -> dict:
+    r = daily_ret.iloc[lo:hi]
+    eq = (1 + r).cumprod()
+    peak = eq.cummax(); dd = float(((peak - eq) / peak).max() * 100)
+    yrs = len(r) / ANN
+    tot = (eq.iloc[-1] - 1) * 100
+    cagr = ((eq.iloc[-1]) ** (1 / yrs) - 1) * 100 if yrs > 0 else 0.0
+    sharpe = float(r.mean() / r.std() * np.sqrt(ANN)) if r.std() > 0 else 0.0
+    return dict(ret=tot, cagr=cagr, dd=dd, sharpe=sharpe)
+
+
+def yearly_returns(daily_ret: pd.Series) -> dict[int, float]:
+    """Rendimento % netto per anno solare dai rendimenti giornalieri composti."""
+    g = daily_ret.groupby(daily_ret.index.year).apply(lambda x: ((1 + x).prod() - 1) * 100)
+    return {int(y): float(v) for y, v in g.items()}
+
+
+def port_returns(members: dict[str, pd.Series], weights: dict[str, float] | None = None) -> pd.Series:
+    """Rendimenti giornalieri di un portafoglio ribilanciato ogni giorno ai pesi dati."""
+    dr = pd.DataFrame({k: v.pct_change().fillna(0.0) for k, v in members.items()})
+    if weights is None:
+        return dr.mean(axis=1)
+    w = pd.Series(weights); w = w / w.sum()
+    return (dr * w).sum(axis=1)
+
+
+def inv_vol_weights(members: dict[str, pd.Series], lo=0, hi=None) -> dict[str, float]:
+    """Pesi inversamente proporzionali alla volatilita' (stimata sulla finestra train)."""
+    vol = {k: v.pct_change().iloc[lo:hi].std() for k, v in members.items()}
+    inv = {k: (1.0 / s if s and s > 0 else 0.0) for k, s in vol.items()}
+    tot = sum(inv.values())
+    return {k: x / tot for k, x in inv.items()}
+
+
+# ---------------- report ----------------
+def row(label, dr):
+    f = metrics(dr); o = metrics(dr, lo=SPLIT)
+    print(f"  {label:<26s}{f['ret']:>+9.0f}{f['cagr']:>7.0f}{f['dd']:>7.1f}{f['sharpe']:>7.2f}"
+          f"   | {o['ret']:>+9.0f}{o['cagr']:>7.0f}{o['dd']:>7.1f}{o['sharpe']:>7.2f}")
+
+
+def main():
+    print("Costruzione equity giornaliere (puo' richiedere ~1 min)...")
+    S = build_all_sleeves()
+    fade = {k: v for k, v in S.items() if k.startswith("MR")}
+    honest = {k: v for k, v in S.items() if not k.startswith("MR")}
+
+    # --- correlazione cross-famiglia ---
+    dr = pd.DataFrame({k: v.pct_change().fillna(0.0) for k, v in S.items()})
+    corr = dr.corr()
+    fade_k, hon_k = list(fade), list(honest)
+    cross = corr.loc[fade_k, hon_k]
+    print("\n" + "=" * 92)
+    print(f"  CORRELAZIONE rendimenti giornalieri — FADE (righe) vs HONEST (colonne) | {IDX[0].date()}->{IDX[-1].date()}")
+    print("=" * 92)
+    print(f"  {'':<12s}" + "".join(f"{c:>13s}" for c in hon_k))
+    for f in fade_k:
+        print(f"  {f:<12s}" + "".join(f"{cross.loc[f,c]:>13.2f}" for c in hon_k))
+    intra_fade = corr.loc[fade_k, fade_k].values[np.triu_indices(len(fade_k), 1)].mean()
+    intra_hon = corr.loc[hon_k, hon_k].values[np.triu_indices(len(hon_k), 1)].mean()
+    print(f"\n  Corr media intra-FADE {intra_fade:+.2f} | intra-HONEST {intra_hon:+.2f} | "
+          f"cross-famiglia {cross.values.mean():+.2f}  (piu' bassa = piu' diversificazione)")
+
+    # --- confronto portafogli ---
+    print("\n" + "=" * 92)
+    print(f"  PORTAFOGLI equal-weight (ribil. giornaliero) | OOS da {OOS_DATE} | leva3x pos15%/sleeve")
+    print("=" * 92)
+    print(f"  {'portafoglio':<26s}{'Ret%':>9s}{'CAGR':>7s}{'DD%':>7s}{'Shrp':>7s}"
+          f"   | {'oRet%':>9s}{'oCAGR':>7s}{'oDD%':>7s}{'oShrp':>7s}")
+    print("  " + "-" * 88)
+    row("FADE only (8 sleeve)", port_returns(fade))
+    row("HONEST only (3 sleeve)", port_returns(honest))
+    row("ALL equal-weight (11)", port_returns(S))
+    # 50/50 fra le due famiglie (ogni famiglia equipesata al suo interno)
+    fr, hr = port_returns(fade), port_returns(honest)
+    row("ALL 50/50 famiglie", (fr + hr) / 2)
+    # inverse-vol sul train, applicato a tutti gli 11 sleeve
+    w = inv_vol_weights(S, lo=0, hi=SPLIT)
+    row("ALL inverse-vol", port_returns(S, w))
+    print("  " + "-" * 88)
+    print("  Sharpe annualizzato sui rendimenti giornalieri. Confronta DD e Sharpe:")
+    print("  se il combinato ha DD piu' basso e Sharpe piu' alto delle singole famiglie, combinare conviene.")
+
+
+if __name__ == "__main__":
+    main()

scripts/analysis/combine_v2.py

@@ -0,0 +1,87 @@
+"""Combina i NUOVI edge (pairs + TSM01) col MASTER esistente: migliora il portafoglio?
+
+Aggiunge al MASTER a 9 sleeve (6 fade + 3 honest) due nuove fonti scoperte
+nell'esplorazione, poco correlate:
+  - PAIRS market-neutral (ETH/BTC, LTC/ETH, ADA/ETH)  -> corr ~0 col mercato
+  - TSM01 (TSMOM multi-orizzonte + risk-off)          -> corr ~0.53 con ROT02
+
+Misura correlazione delle nuove sleeve vs esistenti e confronta MASTER-9 vs
+MASTER-esteso su Ret/CAGR/DD/Sharpe, FULL e OOS (finestra comune 2021-2026).
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.combine_portfolio import (
+    build_all_sleeves, port_returns, metrics, yearly_returns, SPLIT, OOS_DATE, IDX,
+)
+from scripts.analysis.honest_improve2 import _daily_equity, _norm
+from scripts.analysis.pairs_research import pairs_sim
+from scripts.analysis.tsmom_research import tsmom_sim
+
+
+def daily_from(eq_ts, eq_v):
+    return _norm(_daily_equity(eq_ts, eq_v, IDX))
+
+
+def main():
+    print("Costruzione equity (puo' richiedere ~1-2 min)...\n")
+    S = build_all_sleeves()                       # 9 sleeve esistenti
+
+    # nuove sleeve: i 6 pairs robusti di PR01 + TSM01
+    from scripts.strategies.PR01_pairs_reversion import PAIRS
+    new = {}
+    for a, b, p in PAIRS:
+        r = pairs_sim(a, b, **p)
+        new[f"PR_{a}{b}"] = daily_from(r["eq_ts"], r["eq_v"])
+    t = tsmom_sim()
+    new["TSM01"] = daily_from(t["eq_ts"], t["eq_v"])
+
+    allS = {**S, **new}
+
+    # --- correlazione nuove vs esistenti ---
+    dr = pd.DataFrame({k: v.pct_change().fillna(0.0) for k, v in allS.items()})
+    corr = dr.corr()
+    old_k = list(S); new_k = list(new)
+    print("=" * 88)
+    print("  CORRELAZIONE rendimenti giornalieri — NUOVE (righe) vs media esistenti")
+    print("=" * 88)
+    for nk in new_k:
+        avg = corr.loc[nk, old_k].mean()
+        mx = corr.loc[nk, old_k].abs().max()
+        print(f"  {nk:<12s} corr media col MASTER-9 = {avg:+.2f}   |max| = {mx:.2f}")
+
+    # --- confronto portafogli ---
+    def line(label, members):
+        pr = port_returns(members)
+        f, o = metrics(pr), metrics(pr, lo=SPLIT)
+        print(f"  {label:<26s}{f['ret']:>+9.0f}{f['cagr']:>7.0f}{f['dd']:>7.1f}{f['sharpe']:>7.2f}"
+              f"   | {o['ret']:>+9.0f}{o['dd']:>7.1f}{o['sharpe']:>7.2f}")
+        return pr
+
+    print("\n" + "=" * 96)
+    print(f"  MASTER-9 vs MASTER-ESTESO (con pairs+TSM01) | OOS da {OOS_DATE} | equal-weight daily")
+    print("=" * 96)
+    print(f"  {'portafoglio':<26s}{'Ret%':>9s}{'CAGR':>7s}{'DD%':>7s}{'Shrp':>7s}"
+          f"   | {'oRet%':>9s}{'oDD%':>7s}{'oShrp':>7s}")
+    print("  " + "-" * 92)
+    pairs_only = {k: v for k, v in new.items() if k.startswith('PR_')}
+    line(f"MASTER-9 (base)", S)
+    line(f"MASTER +pairs ({len(S)+len(pairs_only)})", {**S, **pairs_only})
+    line(f"MASTER +TSM01 ({len(S)+1})", {**S, "TSM01": new["TSM01"]})
+    pr_all = line(f"MASTER-esteso ({len(allS)})", allS)
+    print("  " + "-" * 92)
+    pa = yearly_returns(pr_all)
+    print("  MASTER-esteso per-anno: " + "  ".join(f"{y}:{v:+.0f}%" for y, v in pa.items()))
+    print("\n  Se il MASTER-esteso ha DD piu' basso e/o Sharpe piu' alto del MASTER-9, le nuove")
+    print("  famiglie aggiungono valore (diversificazione da fonti scorrelate).")
+
+
+if __name__ == "__main__":
+    main()

scripts/analysis/explore_lab.py

@@ -0,0 +1,171 @@
+"""Harness ONESTO condiviso per esplorare nuove famiglie di strategie.
+
+Regole NON negoziabili (per non ripetere l'errore squeeze look-ahead):
+  - direzione e prezzo decisi con dati FINO a close[i] incluso, mai con la barra i
+    usata per scegliere la direzione e poi entrare a i-1;
+  - ingresso ESEGUIBILE a close[i];
+  - exit: take-profit / stop-loss intrabar (high/low) e/o time-limit max_bars;
+    tp/sl possono essere None -> exit solo a tempo (utile per stagionalita');
+  - una posizione per volta (non-overlap), capitale composto;
+  - NETTO dopo fee round-trip (default 0.10% RT reale Deribit) e leva;
+  - validazione OOS (held-out, ultimo 30%) + sweep fee 0.00-0.20% RT.
+
+Le strategie ad alta frequenza muoiono di fee: ogni entry costa fee_rt*lev sul
+notional. Tienine conto: meno operazioni e edge > costi.
+
+Asset disponibili: ADA BNB BTC DOGE ETH LTC SOL XRP (1h, 15m; BTC/ETH anche 5m).
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from src.data.downloader import load_data
+
+FEE_RT = 0.001       # Deribit perp realistico: taker ~0.05%/lato = 0.10% RT
+LEV = 3.0
+POS = 0.15
+OOS_FRAC = 0.30
+ASSETS = ["ADA", "BNB", "BTC", "DOGE", "ETH", "LTC", "SOL", "XRP"]
+BARS_PER_YEAR = {"5m": 105120, "15m": 35040, "1h": 8760, "4h": 2190, "1d": 365}
+
+
+# --------------------------- dati ---------------------------
+def get_df(asset: str, tf: str) -> pd.DataFrame:
+    """OHLCV con colonna dt (UTC). tf nativo (5m,15m,1h) o resample da 1h (4h,1d).
+    timestamp resta ms-epoch reale anche dopo il resample (no placeholder)."""
+    if tf in ("5m", "15m", "1h"):
+        df = load_data(asset, tf).reset_index(drop=True)
+    else:
+        base = load_data(asset, "1h").copy()
+        base["dt"] = pd.to_datetime(base["timestamp"], unit="ms", utc=True)
+        base = base.set_index("dt")
+        rule = {"4h": "4h", "1d": "1D"}[tf]
+        agg = base.resample(rule).agg(
+            {"open": "first", "high": "max", "low": "min", "close": "last", "volume": "sum"}
+        ).dropna()
+        epoch = pd.Timestamp("1970-01-01", tz="UTC")          # ms-epoch portabile (qualsiasi risoluzione)
+        agg["timestamp"] = ((agg.index - epoch) // pd.Timedelta(milliseconds=1)).astype("int64")
+        df = agg.reset_index(drop=True)
+    df["dt"] = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+    return df
+
+
+def _dt(df: pd.DataFrame) -> pd.DatetimeIndex:
+    return pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+
+
+# --------------------------- indicatori ---------------------------
+def atr(df: pd.DataFrame, n: int = 14) -> np.ndarray:
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    pc = np.roll(c, 1); pc[0] = c[0]
+    tr = np.maximum(h - l, np.maximum(np.abs(h - pc), np.abs(l - pc)))
+    return pd.Series(tr).rolling(n).mean().values
+
+
+def ema(x: np.ndarray, n: int) -> np.ndarray:
+    return pd.Series(x).ewm(span=n, adjust=False).mean().values
+
+
+def rsi(close: np.ndarray, n: int = 14) -> np.ndarray:
+    d = np.diff(close, prepend=close[0])
+    up = pd.Series(np.where(d > 0, d, 0.0)).ewm(alpha=1/n, adjust=False).mean()
+    dn = pd.Series(np.where(d < 0, -d, 0.0)).ewm(alpha=1/n, adjust=False).mean()
+    rs = up / dn.replace(0, np.nan)
+    return (100 - 100 / (1 + rs)).values
+
+
+# --------------------------- engine ---------------------------
+def simulate(entries: list[dict], df: pd.DataFrame, fee_rt: float = FEE_RT,
+             lev: float = LEV, pos: float = POS, split: int = -1) -> dict:
+    """entries: dict con i(idx), d(+1/-1), max_bars; tp/sl opzionali (None=solo tempo).
+    split: se >0, conta solo entries con i>=split (finestra OOS)."""
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    n = len(c)
+    ts = _dt(df)
+    cap = peak = 1000.0
+    max_dd = 0.0
+    fee = fee_rt * lev
+    trades = wins = 0
+    last_exit = -1
+    bars_in = 0
+    yearly: dict[int, float] = {}
+    rets: list[float] = []
+
+    for e in entries:
+        i, d = e["i"], e["d"]
+        if i <= last_exit or i + 1 >= n or i < split:
+            continue
+        entry = c[i]
+        tp, sl, mb = e.get("tp"), e.get("sl"), e["max_bars"]
+        exit_p = c[min(i + mb, n - 1)]
+        j = min(i + mb, n - 1)
+        for k in range(1, mb + 1):
+            j = i + k
+            if j >= n:
+                exit_p = c[n - 1]; break
+            hit_sl = sl is not None and ((d == 1 and l[j] <= sl) or (d == -1 and h[j] >= sl))
+            hit_tp = tp is not None and ((d == 1 and h[j] >= tp) or (d == -1 and l[j] <= tp))
+            if hit_sl:
+                exit_p = sl; break
+            if hit_tp:
+                exit_p = tp; break
+            if k == mb:
+                exit_p = c[j]
+        ret = (exit_p - entry) / entry * d * lev - fee
+        cb = cap
+        cap = max(cb + cb * pos * ret, 10.0)
+        peak = max(peak, cap); max_dd = max(max_dd, (peak - cap) / peak)
+        trades += 1; wins += ret > 0; bars_in += (j - i)
+        last_exit = j
+        rets.append(ret * pos)
+        yearly[ts.iloc[i].year] = yearly.get(ts.iloc[i].year, 0.0) + ret * 100
+
+    sharpe = float(np.mean(rets) / np.std(rets) * np.sqrt(len(rets))) if len(rets) > 1 and np.std(rets) > 0 else 0.0
+    return {
+        "trades": trades,
+        "win": wins / trades * 100 if trades else 0.0,
+        "ret": (cap / 1000 - 1) * 100,
+        "dd": max_dd * 100,
+        "sharpe": sharpe,
+        "yearly": yearly,
+        "exposure": bars_in / n * 100 if n else 0.0,
+    }
+
+
+def evaluate(name: str, entries: list[dict], df: pd.DataFrame,
+             fees=(0.0, 0.0005, 0.001, 0.002)) -> dict:
+    """Valuta una lista di entries: FULL, OOS e sweep fee. Stampa una riga sintetica."""
+    split = int(len(df) * (1 - OOS_FRAC))
+    full = simulate(entries, df)
+    oos = simulate(entries, df, split=split)
+    sweep = {f: simulate(entries, df, fee_rt=f)["ret"] for f in fees}
+    sweep_oos = {f: simulate(entries, df, fee_rt=f, split=split)["ret"] for f in fees}
+    yrs = full["yearly"]; pos_yrs = sum(1 for v in yrs.values() if v > 0)
+    print(f"  {name:<24s} trd={full['trades']:>5d} win={full['win']:>4.1f}% "
+          f"FULL={full['ret']:>+7.0f}% OOS={oos['ret']:>+7.0f}% DD={full['dd']:>4.0f}% "
+          f"oDD={oos['dd']:>4.0f}% Shrp={full['sharpe']:>4.2f} exp={full['exposure']:>4.1f}% "
+          f"anniPos={pos_yrs}/{len(yrs)} | fee0.2%: FULL={sweep[0.002]:>+6.0f} OOS={sweep_oos[0.002]:>+6.0f}")
+    return {"full": full, "oos": oos, "sweep": sweep, "sweep_oos": sweep_oos, "pos_yrs": pos_yrs, "n_yrs": len(yrs)}
+
+
+def robust(res: dict) -> bool:
+    """Verdetto onesto: positivo FULL e OOS, regge a fee 0.20% RT, quasi tutti gli anni positivi."""
+    return (res["full"]["ret"] > 0 and res["oos"]["ret"] > 0
+            and res["sweep"][0.002] > 0 and res["sweep_oos"][0.002] > 0
+            and res["pos_yrs"] >= max(res["n_yrs"] - 1, 1))
+
+
+if __name__ == "__main__":
+    # smoke test: una stagionalita' banale (hour-of-day) su BTC 1h
+    df = get_df("BTC", "1h"); ts = _dt(df)
+    ents = [{"i": i, "d": 1, "max_bars": 6, "tp": None, "sl": None}
+            for i in range(len(df) - 7) if ts.iloc[i].hour == 0]
+    print("smoke test — BTC long ad ogni 00:00 UTC, hold 6h:")
+    evaluate("seasonality_h0", ents, df)

scripts/analysis/honest_final.py

@@ -0,0 +1,103 @@
+"""Validazione FINALE delle 3 strategie oneste selezionate.
+Per ciascuna: per-asset FULL/OOS/DD/anni-positivi + sweep fee (0/0.05/0.10/0.20% RT).
+Tutto NETTO, ingresso eseguibile, OOS = ultimo 30%, leva 3x.
+
+S1 DIP   — long-only dip-buy z-score reversion (1h)         [regime: reversione]
+S2 TREND — long-only EMA 20/100 trend-following (4h)        [regime: momentum singolo]
+S3 ROT   — rotazione cross-sectional momentum sul paniere (1d) [regime: forza relativa]
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.honest_lab import atr, ema, get_df, simulate, oos_split, available_assets
+from scripts.analysis.honest_trend import simulate_position, ema_dual_signal, oos as trend_oos
+from scripts.analysis.honest_rotation import build_panel, simulate_rotation
+
+FEES = [0.0, 0.0005, 0.001, 0.002]
+
+
+# ---- S1 DIP ----
+def dip_entries(df, n=50, z_in=2.5, sl_atr=2.5, max_bars=24):
+    c = df["close"].values
+    ma = pd.Series(c).rolling(n).mean().values
+    sd = pd.Series(c).rolling(n).std().values
+    a = atr(df, 14)
+    z = (c - ma) / np.where(sd == 0, np.nan, sd)
+    ents = []
+    for i in range(n + 14, len(c)):
+        if np.isnan(z[i]) or np.isnan(a[i]):
+            continue
+        if z[i] <= -z_in and z[i - 1] > -z_in:
+            ents.append({"i": i, "d": 1, "tp": ma[i], "sl": c[i] - sl_atr * a[i], "max_bars": max_bars})
+    return ents
+
+
+def validate_dip(assets):
+    print("\n" + "=" * 100)
+    print("  S1 DIP — long-only dip-buy z-score reversion | 1h | n=50 z=2.5 sl=2.5ATR mb=24")
+    print("=" * 100)
+    print(f"  {'Asset':<6s}{'Trd':>6s}{'Win%':>7s}{'FULL%':>9s}{'OOS%':>9s}{'DD%':>6s}{'Exp%':>6s}{'AnniP':>8s}"
+          f"{'  fee-sweep OOS% (0/0.05/0.10/0.20)':<40s}")
+    ok = 0
+    for a in assets:
+        df = get_df(a, "1h"); ents = dip_entries(df)
+        if len(ents) < 30:
+            continue
+        full = simulate(ents, df); _, oe = oos_split(ents, df); oos = simulate(oe, df)
+        sweep = "  ".join(f"{simulate(oe, df, fee_rt=f).ret:+.0f}" for f in FEES)
+        good = full.ret > 0 and oos.ret > 0
+        ok += good
+        print(f"  {a:<6s}{full.trades:>6d}{full.win:>7.1f}{full.ret:>+9.0f}{oos.ret:>+9.0f}"
+              f"{full.dd:>6.0f}{full.exposure:>6.0f}{f'{full.pos_years}/{full.n_years}':>8s}   [{sweep}]"
+              f"{'  OK' if good else ''}")
+    print(f"  -> robusto (FULL+OOS>0) su {ok}/{len(assets)} asset")
+
+
+def validate_trend(assets):
+    print("\n" + "=" * 100)
+    print("  S2 TREND — long-only EMA 20/100 trend | 4h")
+    print("=" * 100)
+    print(f"  {'Asset':<6s}{'Flip':>6s}{'FULL%':>9s}{'OOS%':>9s}{'DD%':>6s}{'Exp%':>6s}{'AnniP':>8s}")
+    ok = 0
+    for a in assets:
+        df = get_df(a, "4h"); sig = ema_dual_signal(df, 20, 100, long_only=True)
+        full = simulate_position(sig, df); oos = trend_oos(sig, df)
+        good = full["ret"] > 0 and oos["ret"] > 0
+        ok += good
+        print(f"  {a:<6s}{full['flips']:>6d}{full['ret']:>+9.0f}{oos['ret']:>+9.0f}"
+              f"{full['dd']:>6.0f}{full['exposure']:>6.0f}{(str(full['pos_years'])+'/'+str(full['n_years'])):>8s}"
+              f"{'  OK' if good else ''}")
+    print(f"  -> robusto su {ok}/{len(assets)} asset")
+
+
+def validate_rot(assets):
+    print("\n" + "=" * 100)
+    print("  S3 ROT — rotazione cross-sectional momentum | 1d | lb=60 top2 su tutto il paniere")
+    print("=" * 100)
+    panel = build_panel(assets, "1d")
+    print(f"  Paniere {list(panel.columns)}  {panel.shape[0]} barre {panel.index[0].date()}->{panel.index[-1].date()}")
+    print(f"  {'fee RT':<10s}{'FULL%':>9s}{'OOS%':>9s}{'DD%':>6s}{'AnniP':>8s}")
+    for f in FEES:
+        full = simulate_rotation(panel, lookback=60, top_k=2, fee_rt=f)
+        oos = simulate_rotation(panel, lookback=60, top_k=2, fee_rt=f, oos_frac=0.30)
+        anni = str(full['pos_years']) + '/' + str(full['n_years'])
+        print(f"  {f*100:>5.2f}%RT  {full['ret']:>+9.0f}{oos['ret']:>+9.0f}{full['dd']:>6.0f}{anni:>8s}")
+    # per-anno alla fee reale
+    full = simulate_rotation(panel, lookback=60, top_k=2, fee_rt=0.001)
+    print("  per-anno (fee 0.10%): " + "  ".join(f"{y}:{v:+.0f}%" for y, v in sorted(full["yearly"].items())))
+
+
+if __name__ == "__main__":
+    assets = available_assets()
+    print(f"VALIDAZIONE FINALE — asset disponibili: {assets}")
+    validate_dip(assets)
+    validate_trend(assets)
+    validate_rot(assets)

scripts/analysis/honest_improve.py

@@ -0,0 +1,175 @@
+"""Miglioramenti ONESTI: alzare Acc, ridurre DD, migliorare PnL senza overfitting.
+
+Leve usate (tutte robuste e documentate, niente tuning sui singoli anni):
+  1. ABSOLUTE-MOMENTUM overlay (dual momentum): vai in CASH quando il "mercato"
+     (BTC) e' sotto la sua media di lungo periodo -> taglia i bear (2022/2026).
+  2. VOL-TARGETING: scala l'esposizione per puntare a una volatilita' costante
+     -> riduce il DD e liscia la PnL.
+  3. TRAILING STOP ad ATR per il trend (TR01) -> blocca i profitti.
+Confronto base vs migliorata su FULL + OOS + DD pieno + per-anno.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.honest_lab import atr, ema, get_df, available_assets, FEE_RT
+from scripts.analysis.honest_rotation import build_panel
+
+LEV, POS = 3.0, 0.15
+
+
+def _dd(eq: np.ndarray) -> float:
+    peak = eq[0]; mx = 0.0
+    for v in eq:
+        peak = max(peak, v); mx = max(mx, (peak - v) / peak if peak > 0 else 0.0)
+    return mx * 100
+
+
+# ============================================================================
+# ROT01 migliorata: dual-momentum (cash se BTC < SMA) + vol-target
+# ============================================================================
+def rot_improved(lookback=60, top_k=2, gross=0.45, regime_n=100,
+                 target_vol=0.0, vol_n=20, fee_rt=FEE_RT, oos_frac=0.0):
+    panel = build_panel(available_assets(), "1d")
+    cols = list(panel.columns)
+    P = panel.values; T, N = P.shape
+    rets = np.zeros_like(P); rets[1:] = P[1:] / P[:-1] - 1
+    years = panel.index.year.values
+    btc = P[:, cols.index("BTC")]
+    use_regime = regime_n and regime_n > 1
+    btc_ma = pd.Series(btc).rolling(max(regime_n, 2)).mean().values
+    # vol realizzata del portafoglio equal-weight come proxy di scala
+    mkt_ret = rets.mean(axis=1)
+    rv = pd.Series(mkt_ret).rolling(vol_n).std().values * np.sqrt(365)
+    start = max(lookback + 1, (regime_n + 1) if use_regime else 0, int(T * (1 - oos_frac)) if oos_frac else 0)
+    cap = 1000.0; w = np.zeros(N)
+    eq = [cap]; yearly: dict[int, float] = {}; pos_days = {}; days = {}; reb = {}
+    for i in range(start, T - 1):
+        if use_regime:
+            risk_on = btc[i] > btc_ma[i] if not np.isnan(btc_ma[i]) else False
+        else:
+            risk_on = True
+        mom = P[i] / P[i - lookback] - 1
+        order = np.argsort(mom)[::-1]
+        chosen = [j for j in order if mom[j] > 0][:top_k] if risk_on else []
+        g = gross
+        if target_vol > 0 and not np.isnan(rv[i]) and rv[i] > 0:
+            g = min(gross, gross * target_vol / rv[i])     # solo riduzione (no leva extra)
+        new_w = np.zeros(N)
+        for j in chosen:
+            new_w[j] = g / len(chosen)
+        turnover = np.abs(new_w - w).sum()
+        if turnover > 1e-9:
+            cap -= cap * turnover * (fee_rt / 2)
+        w = new_w
+        pr = float(np.dot(w, rets[i + 1]))
+        cap = max(cap * (1 + pr), 10.0)
+        eq.append(cap)
+        y = int(years[i])
+        yearly[y] = yearly.get(y, 0.0) + pr * 100
+        pos_days[y] = pos_days.get(y, 0) + (pr > 0); days[y] = days.get(y, 0) + 1
+        reb[y] = reb.get(y, 0) + (turnover > 1e-9)
+    return {"ret": (cap / 1000 - 1) * 100, "dd": _dd(np.array(eq)), "yearly": yearly,
+            "pos_years": sum(1 for v in yearly.values() if v > 0), "n_years": len(yearly),
+            "pos_days": pos_days, "days": days, "reb": reb}
+
+
+# ============================================================================
+# DIP01 migliorata: filtro regime (no dip in bear forte) + vol-target sizing
+# ============================================================================
+def dip_improved(asset, tf="1h", n=50, z_in=2.5, sl_atr=2.5, max_bars=24,
+                 regime_n=200, vol_target=0.0, fee_rt=FEE_RT, oos_frac=0.0):
+    df = get_df(asset, tf)
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    N = len(c); ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+    ma = pd.Series(c).rolling(n).mean().values
+    sd = pd.Series(c).rolling(n).std().values
+    a = atr(df, 14)
+    z = (c - ma) / np.where(sd == 0, np.nan, sd)
+    sma_r = pd.Series(c).rolling(regime_n).mean().values
+    atr_pct = a / c           # volatilita' relativa
+    base_vol = np.nanmedian(atr_pct[regime_n:regime_n * 2]) if N > regime_n * 2 else np.nanmedian(atr_pct)
+    fee = fee_rt * LEV
+    cap = 1000.0; last_exit = -1
+    eq = [cap]; yt: dict[int, list] = {}
+    start = max(n + 14, regime_n + 1) if regime_n else n + 14
+    split = int(N * (1 - oos_frac)) if oos_frac else 0
+    for i in range(start, N):
+        if i < split or np.isnan(z[i]) or np.isnan(a[i]):
+            continue
+        if not (z[i] <= -z_in and z[i - 1] > -z_in):
+            continue
+        # filtro regime: salta i dip in bear forte (prezzo molto sotto SMA lunga)
+        if regime_n and not np.isnan(sma_r[i]) and c[i] < sma_r[i] * 0.90:
+            continue
+        if i <= last_exit or i + 1 >= N:
+            continue
+        # vol-target: riduci posizione se ATR% > base (no leva extra)
+        psize = POS
+        if vol_target > 0 and not np.isnan(atr_pct[i]) and atr_pct[i] > 0:
+            psize = POS * min(1.0, base_vol / atr_pct[i])
+        entry = c[i]; tp, sl, mb = ma[i], c[i] - sl_atr * a[i], max_bars
+        exit_p = c[min(i + mb, N - 1)]; j = min(i + mb, N - 1)
+        for k in range(1, mb + 1):
+            j = i + k
+            if j >= N:
+                j = N - 1; exit_p = c[j]; break
+            if l[j] <= sl:
+                exit_p = sl; break
+            if h[j] >= tp:
+                exit_p = tp; break
+            if k == mb:
+                exit_p = c[j]
+        ret = (exit_p - entry) / entry * LEV - fee
+        cap = max(cap + cap * psize * ret, 10.0)
+        last_exit = j
+        y = ts.iloc[i].year
+        rec = yt.setdefault(y, [0, 0]); rec[0] += 1; rec[1] += ret > 0
+        eq.append(cap)
+    t = sum(v[0] for v in yt.values()); w = sum(v[1] for v in yt.values())
+    return {"ret": (cap / 1000 - 1) * 100, "dd": _dd(np.array(eq)),
+            "trades": t, "acc": w / t * 100 if t else 0.0,
+            "yt": yt, "pos_years": sum(1 for v in yt.values() if v[1] / max(v[0],1) and v[1]>v[0]*0 and (v[1]>0)), "n_years": len(yt)}
+
+
+def dip_acc_pnl(asset, **kw):
+    """ritorna anche FULL e OOS."""
+    full = dip_improved(asset, **kw)
+    oos = dip_improved(asset, oos_frac=0.30, **kw)
+    return full, oos
+
+
+if __name__ == "__main__":
+    print("=" * 92)
+    print("  ROT01 — BASE vs MIGLIORATA (dual-momentum cash + vol-target)")
+    print("=" * 92)
+    print(f"  {'config':<40s}{'FULL%':>9s}{'OOS%':>9s}{'DD%pieno':>10s}{'AnniP':>8s}")
+    b = rot_improved(regime_n=0); bo = rot_improved(regime_n=0, oos_frac=0.30)
+    print(f"  {'BASE (no overlay)':<40s}{b['ret']:>+9.0f}{bo['ret']:>+9.0f}{b['dd']:>10.0f}"
+          f"{str(b['pos_years'])+'/'+str(b['n_years']):>8s}")
+    for rn in [100, 150, 200]:
+        f = rot_improved(regime_n=rn); o = rot_improved(regime_n=rn, oos_frac=0.30)
+        print(f"  {'+ dual-mom cash (BTC<SMA'+str(rn)+')':<40s}{f['ret']:>+9.0f}{o['ret']:>+9.0f}"
+              f"{f['dd']:>10.0f}{str(f['pos_years'])+'/'+str(f['n_years']):>8s}")
+    for tv in [0.6, 0.8]:
+        f = rot_improved(regime_n=150, target_vol=tv); o = rot_improved(regime_n=150, target_vol=tv, oos_frac=0.30)
+        print(f"  {'+ dual-mom150 + volTarget'+str(tv):<40s}{f['ret']:>+9.0f}{o['ret']:>+9.0f}"
+              f"{f['dd']:>10.0f}{str(f['pos_years'])+'/'+str(f['n_years']):>8s}")
+
+    print("\n" + "=" * 92)
+    print("  DIP01 — BASE vs MIGLIORATA (filtro regime + vol-target)")
+    print("=" * 92)
+    print(f"  {'asset / config':<34s}{'Trd':>6s}{'Acc%':>7s}{'FULL%':>9s}{'OOS%':>9s}{'DD%pieno':>10s}")
+    for a in ["BTC", "ETH", "SOL"]:
+        for label, kw in [("base", dict(regime_n=0, vol_target=0)),
+                          ("+regime+volTgt", dict(regime_n=200, vol_target=0.5))]:
+            f, o = dip_acc_pnl(a, **kw)
+            print(f"  {a+' '+label:<34s}{f['trades']:>6d}{f['acc']:>7.1f}{f['ret']:>+9.0f}"
+                  f"{o['ret']:>+9.0f}{f['dd']:>10.0f}")

scripts/analysis/honest_improve2.py

@@ -0,0 +1,184 @@
+"""Miglioramenti v2: market-regime gate su DIP01 + PORTAFOGLIO combinato.
+
+- DIP01 con gate di mercato: compra i dip solo quando BTC e' risk-on (BTC>SMA),
+  cosi' si evitano le capitolazioni dei bear (2018/2022) che peggiorano Acc/DD/PnL.
+- Portafoglio: equal-weight giornaliero delle 3 strategie migliorate -> la
+  diversificazione taglia il DD mantenendo la PnL (migliora il risk-adjusted).
+Tutto NETTO, con DD pieno e per-anno.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.honest_lab import atr, ema, get_df, available_assets, FEE_RT
+from scripts.analysis.honest_improve import rot_improved, _dd
+
+LEV, POS = 3.0, 0.15
+
+
+def _daily_equity(ts_list, cap_list, idx):
+    """serie di equity giornaliera (ffill) su un DatetimeIndex comune."""
+    s = pd.Series(cap_list, index=pd.to_datetime(ts_list, utc=True))
+    s = s[~s.index.duplicated(keep="last")].sort_index()
+    daily = s.resample("1D").last().reindex(idx).ffill().bfill()
+    return daily
+
+
+# ---------- DIP01 con market-regime gate ----------
+def dip_market_gated(asset, n=50, z_in=2.5, sl_atr=2.5, max_bars=24,
+                     market_n=100, fee_rt=FEE_RT, oos_frac=0.0, return_equity=False):
+    df = get_df(asset, "1h")
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    N = len(c); ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+    ma = pd.Series(c).rolling(n).mean().values
+    sd = pd.Series(c).rolling(n).std().values
+    a = atr(df, 14)
+    z = (c - ma) / np.where(sd == 0, np.nan, sd)
+    # regime di mercato: BTC 1h > SMA(market_n in giorni -> *24 barre)
+    btc = get_df("BTC", "1h")
+    bser = pd.Series(btc["close"].values,
+                     index=pd.to_datetime(btc["timestamp"], unit="ms", utc=True))
+    bser = bser[~bser.index.duplicated()]
+    bma = bser.rolling(market_n * 24).mean()
+    risk_on = (bser > bma).reindex(ts, method="ffill").fillna(False).values
+    fee = fee_rt * LEV
+    cap = 1000.0; last_exit = -1
+    eq_ts, eq_v = [], []
+    yt: dict[int, list] = {}; ypnl: dict[int, float] = {}
+    split = int(N * (1 - oos_frac)) if oos_frac else 0
+    for i in range(n + 14, N):
+        if i < split or np.isnan(z[i]) or np.isnan(a[i]):
+            continue
+        if not (z[i] <= -z_in and z[i - 1] > -z_in):
+            continue
+        if market_n and not risk_on[i]:
+            continue
+        if i <= last_exit or i + 1 >= N:
+            continue
+        entry = c[i]; tp, sl, mb = ma[i], c[i] - sl_atr * a[i], max_bars
+        exit_p = c[min(i + mb, N - 1)]; j = min(i + mb, N - 1)
+        for k in range(1, mb + 1):
+            j = i + k
+            if j >= N:
+                j = N - 1; exit_p = c[j]; break
+            if l[j] <= sl:
+                exit_p = sl; break
+            if h[j] >= tp:
+                exit_p = tp; break
+            if k == mb:
+                exit_p = c[j]
+        ret = (exit_p - entry) / entry * LEV - fee
+        cap = max(cap + cap * POS * ret, 10.0)
+        last_exit = j
+        y = ts.iloc[i].year
+        rec = yt.setdefault(y, [0, 0]); rec[0] += 1; rec[1] += ret > 0
+        ypnl[y] = ypnl.get(y, 0.0) + ret * 100
+        eq_ts.append(ts.iloc[j]); eq_v.append(cap)
+    t = sum(v[0] for v in yt.values()); w = sum(v[1] for v in yt.values())
+    out = {"ret": (cap / 1000 - 1) * 100, "dd": _dd(np.array(eq_v)) if eq_v else 0.0,
+           "trades": t, "acc": w / t * 100 if t else 0.0, "yt": yt, "ypnl": ypnl,
+           "pos_years": sum(1 for v in ypnl.values() if v > 0), "n_years": len(ypnl)}
+    if return_equity:
+        out["eq_ts"], out["eq_v"] = eq_ts, eq_v
+    return out
+
+
+def main():
+    print("=" * 96)
+    print("  DIP01 — base vs MARKET-GATE (compra dip solo se BTC>SMA100)")
+    print("=" * 96)
+    print(f"  {'asset / config':<30s}{'Trd':>6s}{'Acc%':>7s}{'FULL%':>9s}{'OOS%':>9s}{'DD%':>7s}{'AnniP':>8s}")
+    for a in ["BTC", "ETH", "SOL"]:
+        b = dip_market_gated(a, market_n=0); bo = dip_market_gated(a, market_n=0, oos_frac=0.30)
+        g = dip_market_gated(a, market_n=100); go = dip_market_gated(a, market_n=100, oos_frac=0.30)
+        print(f"  {a+' base':<30s}{b['trades']:>6d}{b['acc']:>7.1f}{b['ret']:>+9.0f}{bo['ret']:>+9.0f}"
+              f"{b['dd']:>7.0f}{str(b['pos_years'])+'/'+str(b['n_years']):>8s}")
+        print(f"  {a+' +gate100':<30s}{g['trades']:>6d}{g['acc']:>7.1f}{g['ret']:>+9.0f}{go['ret']:>+9.0f}"
+              f"{g['dd']:>7.0f}{str(g['pos_years'])+'/'+str(g['n_years']):>8s}")
+
+    # ---------- PORTAFOGLIO combinato (3 sleeve diversificate) ----------
+    print("\n" + "=" * 96)
+    print("  PORTAFOGLIO equal-weight giornaliero (ribilanciato): DIP01 + TR01-basket + ROT02")
+    print("=" * 96)
+    idx = pd.date_range("2021-01-01", "2026-05-26", freq="1D", tz="UTC")
+    # sleeve 1: DIP01 base su BTC (la migliore)
+    d = dip_market_gated("BTC", market_n=0, return_equity=True)
+    eq_dip = _norm(_daily_equity(d["eq_ts"], d["eq_v"], idx))
+    # sleeve 2: TR01 equal-weight su {BNB,BTC,DOGE,SOL,XRP}
+    eq_tr = _norm(_tr_basket_daily(["BNB", "BTC", "DOGE", "SOL", "XRP"], idx))
+    # sleeve 3: ROT02 dual-momentum
+    eq_rot = _norm(_rot_daily_equity(idx))
+    members = {"DIP01_BTC": eq_dip, "TR01_basket": eq_tr, "ROT02_dualmom": eq_rot}
+    # ribilanciamento giornaliero equal-weight: media dei rendimenti giornalieri
+    drets = pd.DataFrame({k: v.pct_change().fillna(0) for k, v in members.items()})
+    port_ret = drets.mean(axis=1)
+    combo = (1 + port_ret).cumprod()
+    print(f"  Periodo {idx[0].date()} -> {idx[-1].date()}  (leva/pos gia' incluse nelle sleeve)")
+    print(f"  {'sleeve':<16s}{'ret%':>9s}{'DD%':>7s}{'CAGR%':>8s}")
+    yrs = (idx[-1] - idx[0]).days / 365.25
+    for name, s in members.items():
+        r = (s.iloc[-1] / s.iloc[0] - 1) * 100
+        cagr = ((s.iloc[-1] / s.iloc[0]) ** (1 / yrs) - 1) * 100
+        print(f"  {name:<16s}{r:>+9.0f}{_dd(s.values):>7.0f}{cagr:>8.0f}")
+    r = (combo.iloc[-1] / combo.iloc[0] - 1) * 100
+    cagr = ((combo.iloc[-1] / combo.iloc[0]) ** (1 / yrs) - 1) * 100
+    print(f"  {'PORTAFOGLIO':<16s}{r:>+9.0f}{_dd(combo.values):>7.0f}{cagr:>8.0f}  <-- DD molto piu' basso, CAGR solida")
+    # per-anno del portafoglio
+    pa = (port_ret.groupby(port_ret.index.year).apply(lambda x: ((1 + x).prod() - 1) * 100))
+    print("  Portafoglio per-anno: " + "  ".join(f"{y}:{v:+.0f}%" for y, v in pa.items()))
+
+
+def _norm(s):
+    return s / s.iloc[0]
+
+
+def _tr_basket_daily(assets, idx):
+    """equity giornaliera media di TR01 (EMA20/100 long-only, 4h) sul paniere."""
+    eqs = []
+    for a in assets:
+        df = get_df(a, "4h"); c = df["close"].values; n = len(c)
+        ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+        ef, es = ema(c, 20), ema(c, 100)
+        sig = np.where(ef > es, 1.0, 0.0); sig[:100] = 0.0
+        cap = 1000.0; cur = 0.0; fee = FEE_RT / 2 * LEV
+        tl, cl = [], []
+        for i in range(n - 1):
+            s = sig[i]
+            if s != cur:
+                cap -= cap * POS * fee * abs(s - cur); cur = s
+            cap = max(cap * (1 + POS * LEV * (c[i + 1] - c[i]) / c[i] * cur), 10.0)
+            tl.append(ts.iloc[i]); cl.append(cap)
+        eqs.append(_norm(_daily_equity(tl, cl, idx)))
+    return _norm(pd.concat(eqs, axis=1).mean(axis=1))
+
+
+def _rot_daily_equity(idx):
+    """equity giornaliera della ROT01 dual-momentum (ricostruita bar-by-bar)."""
+    from scripts.analysis.honest_rotation import build_panel
+    panel = build_panel(available_assets(), "1d")
+    cols = list(panel.columns); P = panel.values; T, N = P.shape
+    rets = np.zeros_like(P); rets[1:] = P[1:] / P[:-1] - 1
+    btc = P[:, cols.index("BTC")]; bma = pd.Series(btc).rolling(100).mean().values
+    cap = 1000.0; w = np.zeros(N); ts_list = []; cap_list = []
+    for i in range(101, T - 1):
+        risk_on = btc[i] > bma[i] if not np.isnan(bma[i]) else False
+        mom = P[i] / P[i - 60] - 1; order = np.argsort(mom)[::-1]
+        chosen = [j for j in order if mom[j] > 0][:3] if risk_on else []   # top_k=3 (era 2): DD piu' basso
+        nw = np.zeros(N)
+        for j in chosen:
+            nw[j] = 0.45 / len(chosen)
+        cap -= cap * np.abs(nw - w).sum() * (FEE_RT / 2); w = nw
+        cap = max(cap * (1 + float(np.dot(w, rets[i + 1]))), 10.0)
+        ts_list.append(panel.index[i]); cap_list.append(cap)
+    s = _daily_equity(ts_list, cap_list, idx); return s / s.iloc[0]
+
+
+if __name__ == "__main__":
+    main()

scripts/analysis/honest_lab.py

@@ -0,0 +1,192 @@
+"""honest_lab — laboratorio di ricerca strategie ONESTO e fee-aware.
+
+Principi (per non ripetere l'errore look-ahead della famiglia squeeze):
+  1. Ogni segnale a barra i usa SOLO dati fino a close[i]. Ingresso a close[i]
+     (eseguibile dal vivo: il worker vede la candela chiusa ed entra). Opzione
+     di robustezza: ingresso a open[i+1] (ancora piu' conservativo).
+  2. Uscita TP/SL valutata intrabar su high/low, conservativa: SL prima del TP
+     nello stesso bar. Time-limit max_bars. Una posizione per volta (non-overlap).
+  3. Tutto NETTO dopo fee round-trip realistiche (0.10% Deribit) * leva.
+  4. Validazione: FULL + OOS (held-out ultimo 30%) + per-anno + sweep fee
+     + griglia parametri + su PIU' asset. Niente di tutto cio' -> scartata.
+
+Engine condiviso riusabile da tutte le strategie candidate.
+"""
+from __future__ import annotations
+
+import sys
+from dataclasses import dataclass
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from src.data.downloader import load_data  # noqa: E402
+
+FEE_RT = 0.001          # Deribit perp realistico: taker ~0.05%/lato = 0.10% RT
+LEV = 3.0
+POS = 0.15
+OOS_FRAC = 0.30
+DATA_DIR = PROJECT_ROOT / "data" / "raw"
+
+# ----------------------------------------------------------------------------
+# dati
+# ----------------------------------------------------------------------------
+_CACHE: dict[tuple[str, str], pd.DataFrame] = {}
+
+
+def available_assets() -> list[str]:
+    out = []
+    for p in sorted(DATA_DIR.glob("*_1h.parquet")):
+        name = p.stem.replace("_1h", "").upper()
+        if name not in ("BTC_DVOL", "ETH_DVOL"):
+            out.append(name)
+    return out
+
+
+def get_df(asset: str, tf: str) -> pd.DataFrame:
+    """tf nativo (15m,1h) o resample da 1h (2h,4h,6h,12h,1d)."""
+    key = (asset, tf)
+    if key in _CACHE:
+        return _CACHE[key]
+    if tf in ("15m", "1h"):
+        df = load_data(asset, tf).reset_index(drop=True)
+    else:
+        base = load_data(asset, "1h").copy()
+        base["dt"] = pd.to_datetime(base["timestamp"], unit="ms", utc=True)
+        base = base.set_index("dt")
+        rule = {"2h": "2h", "4h": "4h", "6h": "6h", "12h": "12h", "1d": "1D"}[tf]
+        agg = base.resample(rule).agg(
+            {"open": "first", "high": "max", "low": "min", "close": "last", "volume": "sum"}
+        ).dropna()
+        # l'indice puo' essere datetime64[ms] o [ns]: forza ms in modo robusto
+        agg["timestamp"] = agg.index.values.astype("datetime64[ms]").astype("int64")
+        df = agg.reset_index(drop=True)
+    df = df[["timestamp", "open", "high", "low", "close", "volume"]].copy()
+    _CACHE[key] = df
+    return df
+
+
+# ----------------------------------------------------------------------------
+# indicatori
+# ----------------------------------------------------------------------------
+def atr(df: pd.DataFrame, n: int = 14) -> np.ndarray:
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    pc = np.roll(c, 1); pc[0] = c[0]
+    tr = np.maximum(h - l, np.maximum(np.abs(h - pc), np.abs(l - pc)))
+    return pd.Series(tr).rolling(n).mean().values
+
+
+def rsi(close: np.ndarray, n: int = 14) -> np.ndarray:
+    d = np.diff(close, prepend=close[0])
+    up = pd.Series(np.where(d > 0, d, 0.0)).ewm(alpha=1 / n, adjust=False).mean()
+    dn = pd.Series(np.where(d < 0, -d, 0.0)).ewm(alpha=1 / n, adjust=False).mean()
+    rs = up / dn.replace(0, np.nan)
+    return (100 - 100 / (1 + rs)).values
+
+
+def ema(close: np.ndarray, n: int) -> np.ndarray:
+    return pd.Series(close).ewm(span=n, adjust=False).mean().values
+
+
+# ----------------------------------------------------------------------------
+# engine
+# ----------------------------------------------------------------------------
+@dataclass
+class SimResult:
+    trades: int
+    win: float
+    ret: float          # ritorno % netto composto su 1000
+    dd: float
+    exposure: float
+    yearly: dict[int, float]
+
+    @property
+    def pos_years(self) -> int:
+        return sum(1 for v in self.yearly.values() if v > 0)
+
+    @property
+    def n_years(self) -> int:
+        return len(self.yearly)
+
+
+def simulate(entries: list[dict], df: pd.DataFrame, fee_rt: float = FEE_RT,
+             lev: float = LEV, pos: float = POS, entry_on_open: bool = False) -> SimResult:
+    """entries: dict {i, d(+1/-1), tp, sl, max_bars}.
+
+    entry_on_open=True -> ingresso a open[i+1] invece di close[i] (robustezza).
+    """
+    o, h, l, c = (df["open"].values, df["high"].values,
+                  df["low"].values, df["close"].values)
+    n = len(c)
+    cap = peak = 1000.0
+    max_dd = 0.0
+    fee = fee_rt * lev
+    trades = wins = 0
+    last_exit = -1
+    bars_in = 0
+    ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+    yearly: dict[int, float] = {}
+
+    for e in entries:
+        i, d = e["i"], e["d"]
+        ei = i + 1 if entry_on_open else i        # barra di ingresso
+        if ei <= last_exit or ei + 1 >= n:
+            continue
+        entry = o[ei] if entry_on_open else c[i]
+        tp, sl, mb = e["tp"], e["sl"], e["max_bars"]
+        exit_p = c[min(ei + mb, n - 1)]
+        j = min(ei + mb, n - 1)
+        for k in range(1, mb + 1):
+            j = ei + k
+            if j >= n:
+                j = n - 1; exit_p = c[j]; break
+            hit_sl = (d == 1 and l[j] <= sl) or (d == -1 and h[j] >= sl)
+            hit_tp = (d == 1 and h[j] >= tp) or (d == -1 and l[j] <= tp)
+            if hit_sl:                      # conservativo: SL prima del TP nello stesso bar
+                exit_p = sl; break
+            if hit_tp:
+                exit_p = tp; break
+            if k == mb:
+                exit_p = c[j]
+        ret = (exit_p - entry) / entry * d * lev - fee
+        cap = max(cap + cap * pos * ret, 10.0)
+        peak = max(peak, cap); max_dd = max(max_dd, (peak - cap) / peak)
+        trades += 1; wins += ret > 0; bars_in += (j - ei)
+        last_exit = j
+        yr = ts.iloc[i].year
+        yearly[yr] = yearly.get(yr, 0.0) + ret * 100
+    return SimResult(
+        trades=trades,
+        win=wins / trades * 100 if trades else 0.0,
+        ret=(cap / 1000 - 1) * 100,
+        dd=max_dd * 100,
+        exposure=bars_in / n * 100,
+        yearly=yearly,
+    )
+
+
+def oos_split(entries: list[dict], df: pd.DataFrame, frac: float = OOS_FRAC):
+    split = int(len(df) * (1 - frac))
+    ins = [e for e in entries if e["i"] < split]
+    oos = [e for e in entries if e["i"] >= split]
+    return ins, oos
+
+
+# ----------------------------------------------------------------------------
+# criterio di accettazione
+# ----------------------------------------------------------------------------
+def verdict(full: SimResult, oos: SimResult) -> bool:
+    """Strategia attendibile su un singolo asset/tf."""
+    if full.trades < 30:
+        return False
+    if full.ret <= 0 or oos.ret <= 0:
+        return False
+    if full.pos_years < max(full.n_years - 1, 1):
+        return False
+    if full.dd > 45:
+        return False
+    return True

scripts/analysis/honest_matrix.py

@@ -0,0 +1,80 @@
+"""Tabella unica consolidata: PnL% NETTO per anno, tutte le strategie a confronto.
+Colonne: DIP01(BTC) · TR01(basket) · ROT01(base) · ROT02(dual-mom) · PORTAFOGLIO.
+Ultima riga: TOT e DD full-period.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.honest_lab import available_assets, FEE_RT
+from scripts.analysis.honest_improve import _dd
+from scripts.analysis.honest_improve2 import (
+    dip_market_gated, _daily_equity, _norm, _tr_basket_daily,
+)
+from scripts.analysis.honest_rotation import build_panel
+
+LEV, POS = 3.0, 0.15
+
+
+def rot_daily(idx, regime_n=0, lookback=60, top_k=2, gross=0.45):
+    """equity giornaliera della rotazione, con/senza overlay dual-momentum."""
+    panel = build_panel(available_assets(), "1d")
+    cols = list(panel.columns); P = panel.values; T, N = P.shape
+    rets = np.zeros_like(P); rets[1:] = P[1:] / P[:-1] - 1
+    btc = P[:, cols.index("BTC")]
+    bma = pd.Series(btc).rolling(max(regime_n, 2)).mean().values
+    use_reg = regime_n and regime_n > 1
+    cap = 1000.0; w = np.zeros(N); tl, cl = [], []
+    start = max(lookback + 1, regime_n + 1 if use_reg else 0)
+    for i in range(start, T - 1):
+        risk_on = (btc[i] > bma[i]) if (use_reg and not np.isnan(bma[i])) else True
+        mom = P[i] / P[i - lookback] - 1; order = np.argsort(mom)[::-1]
+        chosen = [j for j in order if mom[j] > 0][:top_k] if risk_on else []
+        nw = np.zeros(N)
+        for j in chosen:
+            nw[j] = gross / len(chosen)
+        cap -= cap * np.abs(nw - w).sum() * (FEE_RT / 2); w = nw
+        cap = max(cap * (1 + float(np.dot(w, rets[i + 1]))), 10.0)
+        tl.append(panel.index[i]); cl.append(cap)
+    return _norm(_daily_equity(tl, cl, idx))
+
+
+def year_pnl(eq):
+    return {int(y): (g.iloc[-1] / g.iloc[0] - 1) * 100 for y, g in _norm(eq).groupby(eq.index.year)}
+
+
+if __name__ == "__main__":
+    idx = pd.date_range("2021-01-01", "2026-05-26", freq="1D", tz="UTC")
+    d = dip_market_gated("BTC", market_n=0, return_equity=True)
+    cols = {
+        "DIP01(BTC)": _norm(_daily_equity(d["eq_ts"], d["eq_v"], idx)),
+        "TR01(bskt)": _norm(_tr_basket_daily(["BNB", "BTC", "DOGE", "SOL", "XRP"], idx)),
+        "ROT01": rot_daily(idx, regime_n=0),
+        "ROT02": rot_daily(idx, regime_n=100),
+    }
+    drets = pd.DataFrame({k: v.pct_change().fillna(0) for k, v in {
+        "DIP01(BTC)": cols["DIP01(BTC)"], "TR01(bskt)": cols["TR01(bskt)"], "ROT02": cols["ROT02"]
+    }.items()})
+    cols["PORTAF."] = (1 + drets.mean(axis=1)).cumprod()
+
+    names = list(cols)
+    py = {n: year_pnl(cols[n]) for n in names}
+    years = sorted({y for n in names for y in py[n]})
+
+    print("=" * 78)
+    print("  PnL% NETTO PER ANNO — confronto strategie (leva 3x, fee 0.10% RT)")
+    print("=" * 78)
+    print(f"  {'Anno':>6s}" + "".join(f"{n:>12s}" for n in names))
+    print("  " + "-" * 72)
+    for y in years:
+        print(f"  {y:>6d}" + "".join(f"{py[n].get(y, float('nan')):>+12.0f}" if y in py[n] else f"{'-':>12s}" for n in names))
+    print("  " + "-" * 72)
+    print(f"  {'TOT%':>6s}" + "".join(f"{(cols[n].iloc[-1]/cols[n].iloc[0]-1)*100:>+12.0f}" for n in names))
+    print(f"  {'DDfull':>6s}" + "".join(f"{_dd(cols[n].values):>12.0f}" for n in names))

scripts/analysis/honest_rotation.py

@@ -0,0 +1,96 @@
+"""Strategia #3 candidata: ROTAZIONE cross-sectional momentum (multi-crypto).
+Una sola strategia che usa l'INTERO paniere: ad ogni ribilanciamento alloca il
+capitale agli asset con momentum migliore (long-only). Cattura la dispersione tra
+crypto (gli alt forti corrono molto piu' di BTC nei bull) senza shortare nulla.
+
+Onesto: i pesi a close[i] usano solo rendimenti passati; il rendimento del bar
+i->i+1 e' realizzato con quei pesi. Fee sul turnover. Allineamento per timestamp.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.honest_lab import get_df, available_assets, FEE_RT  # noqa: E402
+
+LEV = 3.0
+GROSS = 0.45        # esposizione lorda = LEV*POS del singolo (0.15*3) per confronto equo
+
+
+def build_panel(assets: list[str], tf: str) -> pd.DataFrame:
+    """Matrice close allineata per timestamp (inner join)."""
+    closes = {}
+    for a in assets:
+        df = get_df(a, tf)
+        s = pd.Series(df["close"].values,
+                      index=pd.to_datetime(df["timestamp"], unit="ms", utc=True))
+        closes[a] = s[~s.index.duplicated()]
+    panel = pd.DataFrame(closes).dropna()
+    return panel
+
+
+def simulate_rotation(panel: pd.DataFrame, lookback=30, top_k=2,
+                      fee_rt=FEE_RT, gross=GROSS, abs_filter=True,
+                      oos_frac=0.0) -> dict:
+    """Ad ogni barra: ranking per rendimento passato `lookback`; pesi uguali sui
+    top_k con momentum>0 (se abs_filter); altrimenti cash. gross = esposizione tot.
+    oos_frac>0: parte a investire solo dall'ultimo frac del campione."""
+    P = panel.values
+    T, N = P.shape
+    rets = np.zeros_like(P)
+    rets[1:] = P[1:] / P[:-1] - 1
+    years = panel.index.year.values
+    start = max(lookback + 1, int(T * (1 - oos_frac)) if oos_frac else lookback + 1)
+    cap = peak = 1000.0
+    max_dd = 0.0
+    w = np.zeros(N)
+    yearly: dict[int, float] = {}
+    turn_total = 0.0
+    for i in range(start, T - 1):
+        mom = P[i] / P[i - lookback] - 1
+        order = np.argsort(mom)[::-1]
+        new_w = np.zeros(N)
+        chosen = [j for j in order if (mom[j] > 0 or not abs_filter)][:top_k]
+        if chosen:
+            for j in chosen:
+                new_w[j] = gross / len(chosen)
+        # fee sul turnover (one-way = fee_rt/2 su ogni variazione di peso)
+        turnover = np.abs(new_w - w).sum()
+        cap -= cap * turnover * (fee_rt / 2)
+        turn_total += turnover
+        w = new_w
+        port_ret = float(np.dot(w, rets[i + 1]))      # rendimento bar i->i+1
+        cap = max(cap * (1 + port_ret), 10.0)
+        peak = max(peak, cap); max_dd = max(max_dd, (peak - cap) / peak)
+        yearly[years[i]] = yearly.get(years[i], 0.0) + port_ret * 100
+    return {
+        "ret": (cap / 1000 - 1) * 100,
+        "dd": max_dd * 100,
+        "turnover": turn_total,
+        "yearly": yearly,
+        "pos_years": sum(1 for v in yearly.values() if v > 0),
+        "n_years": len(yearly),
+    }
+
+
+if __name__ == "__main__":
+    assets = available_assets()
+    print(f"ROTATION cross-sectional momentum — fee {FEE_RT*100:.2f}% RT, gross {GROSS} | OOS 30%")
+    print(f"  Paniere: {assets}")
+    for tf in ["1d", "4h"]:
+        panel = build_panel(assets, tf)
+        print(f"\n  === {tf} === panel {panel.shape[0]} barre, {panel.index[0].date()} -> {panel.index[-1].date()}")
+        print(f"  {'config':<22s}{'FULL%':>9s}{'OOS%':>9s}{'DD%':>6s}{'Turn':>7s}{'AnniP':>8s}")
+        for lb in [20, 30, 60, 90]:
+            for k in [1, 2, 3]:
+                full = simulate_rotation(panel, lookback=lb, top_k=k)
+                oos = simulate_rotation(panel, lookback=lb, top_k=k, oos_frac=0.30)
+                anni = f"{full['pos_years']}/{full['n_years']}"
+                print(f"  lb{lb:<3d} top{k:<14d}{full['ret']:>+9.0f}{oos['ret']:>+9.0f}"
+                      f"{full['dd']:>6.0f}{full['turnover']:>7.0f}{anni:>8s}")

scripts/analysis/honest_trend.py

@@ -0,0 +1,109 @@
+"""Strategia #3 candidata: time-series momentum / trend (TSMOM).
+Posizione continua decisa a close[i] dai dati passati; fee SOLO sui cambi di
+posizione (poche operazioni su TF alto = fee non letali). Niente look-ahead:
+il rendimento del bar i->i+1 usa la direzione decisa a close[i].
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.honest_lab import ema, get_df, available_assets, FEE_RT  # noqa: E402
+
+LEV = 3.0
+POS = 0.15
+
+
+def simulate_position(sig: np.ndarray, df: pd.DataFrame, fee_rt: float = FEE_RT,
+                      lev: float = LEV, pos: float = POS) -> dict:
+    """sig[i] in {-1,0,1} = direzione tenuta nel bar i->i+1, decisa a close[i].
+    Fee one-way = fee_rt/2 su ogni unita' di variazione posizione."""
+    c = df["close"].values
+    n = len(c)
+    ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+    cap = peak = 1000.0
+    max_dd = 0.0
+    cur = 0.0
+    flips = 0
+    bars_in = 0
+    yearly: dict[int, float] = {}
+    for i in range(n - 1):
+        s = sig[i]
+        if not np.isfinite(s):
+            s = 0.0
+        if s != cur:
+            cap -= cap * pos * (fee_rt / 2) * lev * abs(s - cur)
+            flips += abs(s - cur) > 0
+            cur = s
+        pr = (c[i + 1] - c[i]) / c[i]
+        bar_ret = pos * lev * pr * cur
+        cap = max(cap * (1 + bar_ret), 10.0)
+        peak = max(peak, cap); max_dd = max(max_dd, (peak - cap) / peak)
+        if cur != 0:
+            bars_in += 1
+        yr = ts.iloc[i].year
+        yearly[yr] = yearly.get(yr, 0.0) + bar_ret * 100
+    return {
+        "ret": (cap / 1000 - 1) * 100,
+        "dd": max_dd * 100,
+        "flips": flips,
+        "exposure": bars_in / n * 100,
+        "yearly": yearly,
+        "pos_years": sum(1 for v in yearly.values() if v > 0),
+        "n_years": len(yearly),
+    }
+
+
+def tsmom_signal(df, lookback=30, long_only=False):
+    """+1 se close>close[-lookback], -1 (o 0 se long_only) altrimenti."""
+    c = df["close"].values
+    sig = np.zeros(len(c))
+    for i in range(lookback, len(c)):
+        up = c[i] > c[i - lookback]
+        sig[i] = 1.0 if up else (0.0 if long_only else -1.0)
+    return sig
+
+
+def ema_dual_signal(df, fast=20, slow=100, long_only=False):
+    """+1 se EMA_fast>EMA_slow."""
+    c = df["close"].values
+    ef, es = ema(c, fast), ema(c, slow)
+    sig = np.where(ef > es, 1.0, 0.0 if long_only else -1.0)
+    sig[:slow] = 0.0
+    return sig
+
+
+def oos(sig, df, frac=0.30):
+    split = int(len(df) * (1 - frac))
+    s2 = sig.copy(); s2[:split] = 0.0
+    return simulate_position(s2, df)
+
+
+def show(label, df, sig):
+    full = simulate_position(sig, df)
+    o = oos(sig, df)
+    anni = f"{full['pos_years']}/{full['n_years']}"
+    print(f"  {label:<26s}{full['flips']:>6d}{full['ret']:>+9.0f}{o['ret']:>+9.0f}"
+          f"{full['dd']:>6.0f}{full['exposure']:>6.0f}{anni:>8s}")
+    return full, o
+
+
+if __name__ == "__main__":
+    assets = available_assets()
+    print(f"TSMOM / trend — fee {FEE_RT*100:.2f}% RT, leva3x pos15% | OOS30%")
+    for tf in ["1d", "4h"]:
+        print(f"\n  ###### TF {tf} ######")
+        for a in assets:
+            df = get_df(a, tf)
+            print(f"\n  === {a} {tf} ===  {'Flip':>5s}{'FULL%':>8s}{'OOS%':>8s}{'DD%':>6s}{'Exp%':>6s}{'AnniP':>8s}")
+            show("TSMOM lb30 long/short", df, tsmom_signal(df, 30))
+            show("TSMOM lb30 long-only", df, tsmom_signal(df, 30, long_only=True))
+            show("TSMOM lb90 long/short", df, tsmom_signal(df, 90))
+            show("EMA 20/100 long/short", df, ema_dual_signal(df, 20, 100))
+            show("EMA 20/100 long-only", df, ema_dual_signal(df, 20, 100, long_only=True))

scripts/analysis/intrabar_test.py

@@ -0,0 +1,188 @@
+"""Test ingresso intra-barra: rottura banda squeeze rilevata sul 5m vs close 15m.
+
+Domanda: entrando sul 5m appena il prezzo rompe la banda di Bollinger dello
+squeeze (bande dall'ultima barra 15m CHIUSA -> nessun look-ahead), si recupera
+parte del movimento che l'ingresso al close della barra 15m si perde?
+
+Confronto a parita' di EXIT (stesso wall-clock): l'unica differenza e' il prezzo
+d'ingresso (5m anticipato vs close 15m ritardato). La differenza di rendimento e'
+esattamente lo "scatto" del breakout catturato in piu'.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from src.data.downloader import load_data
+from src.live.signal_engine import keltner_ratio
+
+OOS_START = "2023-11-20"
+BB_W = 14
+SQ_THR = 0.8
+MIN_DUR = 5
+LEV = 3.0
+POS = 0.15
+M15 = 15 * 60 * 1000
+M5 = 5 * 60 * 1000
+
+
+def build_15m_levels(df15: pd.DataFrame) -> pd.DataFrame:
+    c = df15["close"].values
+    h = df15["high"].values
+    l = df15["low"].values
+    n = len(c)
+    kcr = keltner_ratio(c, h, l, BB_W)
+    ma = np.full(n, np.nan)
+    sd = np.full(n, np.nan)
+    for t in range(BB_W, n):
+        w = c[t - BB_W + 1 : t + 1]
+        ma[t] = w.mean()
+        sd[t] = w.std()
+    upper = ma + 2 * sd
+    lower = ma - 2 * sd
+
+    # durata squeeze consecutiva e maturita'
+    dur = np.zeros(n, dtype=int)
+    run = 0
+    for t in range(n):
+        if not np.isnan(kcr[t]) and kcr[t] < SQ_THR:
+            run += 1
+        else:
+            run = 0
+        dur[t] = run
+    mature = dur >= MIN_DUR
+
+    return pd.DataFrame({
+        "ts15": df15["timestamp"].values,
+        "close_time15": df15["timestamp"].values + M15,
+        "close15": c,
+        "upper": upper,
+        "lower": lower,
+        "mature": mature,
+    })
+
+
+def run_asset(asset: str, hold_min: int, fee_rt: float) -> dict:
+    df5 = load_data(asset, "5m").reset_index(drop=True)
+    df15 = load_data(asset, "15m").reset_index(drop=True)
+    lvl = build_15m_levels(df15)
+
+    d5 = pd.DataFrame({
+        "ts5": df5["timestamp"].values,
+        "close_time5": df5["timestamp"].values + M5,
+        "close5": df5["close"].values,
+    })
+
+    # banda armata: ultima barra 15m CHIUSA prima della chiusura del bar 5m
+    armed = pd.merge_asof(
+        d5.sort_values("close_time5"),
+        lvl[["close_time15", "upper", "lower", "mature"]].sort_values("close_time15"),
+        left_on="close_time5", right_on="close_time15", direction="backward",
+    )
+    # barra 15m CONTENENTE il bar 5m (per l'ingresso ritardato a close 15m)
+    cont = pd.merge_asof(
+        d5.sort_values("ts5"),
+        lvl[["ts15", "close15", "close_time15"]].rename(
+            columns={"close_time15": "cont_close_time"}).sort_values("ts15"),
+        left_on="ts5", right_on="ts15", direction="backward",
+    )
+
+    m = armed.copy()
+    m["cont_close"] = cont["close15"].values
+    m["cont_close_time"] = cont["cont_close_time"].values
+
+    oos_ms = int(pd.Timestamp(OOS_START, tz="UTC").timestamp() * 1000)
+    close5 = m["close5"].values
+    ct5 = m["close_time5"].values
+    upper = m["upper"].values
+    lower = m["lower"].values
+    mature = m["mature"].values
+    cont_close = m["cont_close"].values
+    cont_ct = m["cont_close_time"].values
+    n = len(m)
+
+    cap_e = cap_l = 1000.0   # equity ingresso early(5m) e late(15m)
+    peak_e = peak_l = 1000.0
+    dd_e = dd_l = 0.0
+    trades = win_e = win_l = 0
+    thrust_sum = 0.0
+    fee = fee_rt * LEV
+    busy_until = -1
+
+    for i in range(n):
+        if ct5[i] < oos_ms or ct5[i] <= busy_until:
+            continue
+        if not mature[i] or np.isnan(upper[i]):
+            continue
+        if close5[i] > upper[i]:
+            d = 1
+        elif close5[i] < lower[i]:
+            d = -1
+        else:
+            continue
+
+        entry_e = close5[i]
+        entry_l = cont_close[i]
+        exit_time = cont_ct[i] + hold_min * 60 * 1000
+        # primo close 5m al/oltre exit_time
+        j = np.searchsorted(ct5, exit_time, side="left")
+        if j >= n:
+            break
+        exit_p = close5[j]
+
+        ret_e = ((exit_p - entry_e) / entry_e) * d * LEV - fee
+        ret_l = ((exit_p - entry_l) / entry_l) * d * LEV - fee
+        thrust_sum += (entry_l - entry_e) / entry_e * d * 100  # scatto % (no leva)
+
+        cb_e, cb_l = cap_e, cap_l
+        cap_e = max(cb_e + cb_e * POS * ret_e, 10.0)
+        cap_l = max(cb_l + cb_l * POS * ret_l, 10.0)
+        peak_e = max(peak_e, cap_e); dd_e = max(dd_e, (peak_e - cap_e) / peak_e)
+        peak_l = max(peak_l, cap_l); dd_l = max(dd_l, (peak_l - cap_l) / peak_l)
+        trades += 1
+        win_e += ret_e > 0
+        win_l += ret_l > 0
+        busy_until = exit_time
+
+    return {
+        "trades": trades,
+        "avg_thrust": thrust_sum / trades if trades else 0.0,
+        "early_win": win_e / trades * 100 if trades else 0.0,
+        "late_win": win_l / trades * 100 if trades else 0.0,
+        "early_ret": (cap_e / 1000 - 1) * 100,
+        "late_ret": (cap_l / 1000 - 1) * 100,
+        "early_dd": dd_e * 100,
+        "late_dd": dd_l * 100,
+    }
+
+
+def main():
+    for fee_rt in (0.002, 0.001):
+        print("=" * 104)
+        print(f"  INGRESSO INTRA-BARRA 5m vs CLOSE 15m — OOS da {OOS_START} | leva={LEV:.0f}x "
+              f"| fee={fee_rt*100:.2f}% RT")
+        print("  EARLY = entra al close 5m che rompe la banda | LATE = entra al close della barra 15m | stesso exit")
+        print("=" * 104)
+        print(f"  {'Asset':>5s}{'Hold':>6s}{'Trd':>6s}{'Scatto%':>9s}"
+              f"{'EARLY win%':>12s}{'EARLY ret%':>12s}{'LATE win%':>11s}{'LATE ret%':>11s}{'Δret%':>9s}")
+        print("  " + "-" * 100)
+        for asset in ["BTC", "ETH"]:
+            for hold_min in (15, 30, 45):
+                r = run_asset(asset, hold_min, fee_rt)
+                print(f"  {asset:>5s}{hold_min:>5d}m{r['trades']:>6d}{r['avg_thrust']:>+9.3f}"
+                      f"{r['early_win']:>12.1f}{r['early_ret']:>+12.1f}"
+                      f"{r['late_win']:>11.1f}{r['late_ret']:>+11.1f}"
+                      f"{r['early_ret']-r['late_ret']:>+9.1f}")
+            print("  " + "-" * 100)
+        print("  Scatto% = movimento medio (no leva) catturato tra rottura 5m e close 15m, nella direzione.")
+        print("  Δret% = vantaggio dell'ingresso anticipato. Se ~0 o negativo, il 5m non aiuta.\n")
+
+
+if __name__ == "__main__":
+    main()

scripts/analysis/oos_validation.py

@@ -0,0 +1,259 @@
+"""Validazione out-of-sample fee-aware di tutte le strategie live.
+
+Per ognuna delle 6 config in strategies.yml:
+  - split temporale held-out (train = primi (1-test_frac), test = ultimo test_frac)
+  - ML01 (SignalEngine): allena sul train, predice sul test (come il worker live)
+  - rule-based: i segnali sono causali, si valutano quelli nella finestra test
+  - simulazione fedele al worker live: una posizione per volta (non-overlap),
+    uscita a `hold` barre o stop a -2%, fee round-trip e leva inclusi
+
+Stampa, per ogni config: numero trade nel test, win% lordo e netto, return netto,
+costo commissioni, e confronto lordo-vs-netto per isolare l'impatto delle fee.
+Usa i parquet locali (data/raw), nessuna chiamata di rete.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+import yaml
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from src.data.downloader import load_data
+from src.live.strategy_loader import load_strategy
+from src.live.signal_engine import SignalEngine, keltner_ratio, build_features
+
+TEST_FRAC = 0.30
+STOP_PCT = -0.02
+
+
+def simulate(entries: list[tuple[int, int]], close: np.ndarray, hold: int,
+             fee_rt: float, lev: float, pos: float,
+             initial: float = 1000.0, entry_offset: int = 0) -> dict:
+    """FSM fedele al worker live: non-overlap, hold N barre o stop -2%.
+
+    entry_offset: 0 = ingresso a close[i] (worker live); 1 = close[i-1]
+    (convenzione del backtest storico, che conosce la direzione di barra i).
+    """
+    n = len(close)
+    capital = peak = initial
+    max_dd = 0.0
+    fees_eur = gross_eur = 0.0
+    wins_gross = wins_net = n_trades = 0
+    last_exit = -1
+
+    for i, d in entries:
+        e = i - entry_offset
+        if e <= last_exit or e < 0 or e + 1 >= n:
+            continue
+        entry = close[e]
+        exit_price = close[min(e + hold, n - 1)]
+        for k in range(1, hold + 1):
+            j = e + k
+            if j >= n:
+                exit_price = close[n - 1]
+                break
+            if k < hold and (close[j] - entry) / entry * d <= STOP_PCT:
+                exit_price = close[j]
+                break
+            if k == hold:
+                exit_price = close[j]
+
+        actual = (exit_price - entry) / entry * d   # movimento prezzo * direzione (no leva)
+        gross = actual * lev
+        fee = fee_rt * lev
+        net = gross - fee
+
+        cap_before = capital
+        capital = max(cap_before + cap_before * pos * net, 10.0)
+        gross_eur += cap_before * pos * gross
+        fees_eur += cap_before * pos * fee
+        peak = max(peak, capital)
+        max_dd = max(max_dd, (peak - capital) / peak)
+
+        n_trades += 1
+        wins_gross += actual > 0
+        wins_net += net > 0
+        last_exit = e + hold
+
+    return {
+        "trades": n_trades,
+        "win_gross": wins_gross / n_trades * 100 if n_trades else 0.0,
+        "win_net": wins_net / n_trades * 100 if n_trades else 0.0,
+        "net_return_pct": (capital / initial - 1) * 100,
+        "net_eur": capital - initial,
+        "gross_eur": gross_eur,
+        "fees_eur": fees_eur,
+        "final_capital": capital,
+        "max_dd": max_dd * 100,
+    }
+
+
+def rule_entries(name: str, df: pd.DataFrame, params: dict, split: int) -> list[tuple[int, int]]:
+    strat = load_strategy(name)
+    ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+    sigs = strat.generate_signals(df, ts, **params)
+    return [(s.idx, s.direction) for s in sigs if s.idx >= split]
+
+
+def ml_entries(df: pd.DataFrame, params: dict, split: int, hold: int) -> tuple[list[tuple[int, int]], dict]:
+    bb_w = params.get("bb_window", 14)
+    sq_thr = params.get("sq_threshold", 0.8)
+    ml_thr = params.get("ml_threshold", 0.70)
+
+    eng = SignalEngine(bb_w=bb_w, sq_thr=sq_thr, ml_thr=ml_thr)
+    train_res = eng.train(df.iloc[:split].reset_index(drop=True), lookahead=hold)
+    if not eng.trained:
+        return [], train_res
+
+    close = df["close"].values
+    high = df["high"].values
+    low = df["low"].values
+    volume = df["volume"].values
+    n = len(df)
+    kcr = keltner_ratio(close, high, low, bb_w)
+    up_idx = list(eng.model.classes_).index(1)
+
+    entries: list[tuple[int, int]] = []
+    in_sq = False
+    sq_start = 0
+    for i in range(bb_w + 1, n):
+        if np.isnan(kcr[i]):
+            continue
+        is_sq = kcr[i] < sq_thr
+        if is_sq and not in_sq:
+            in_sq, sq_start = True, i
+        elif not is_sq and in_sq:
+            in_sq = False
+            dur = i - sq_start
+            if dur < eng.min_squeeze_bars or i < split or i + hold >= n:
+                continue
+            avg_vol = float(np.mean(volume[sq_start:i]))
+            feats = build_features(df, i, dur, avg_vol, kcr[i])
+            if feats is None:
+                continue
+            p_up = eng.model.predict_proba(eng.scaler.transform(feats.reshape(1, -1)))[0][up_idx]
+            if p_up >= ml_thr:
+                entries.append((i, 1))
+            elif p_up <= (1 - ml_thr):
+                entries.append((i, -1))
+    return entries, train_res
+
+
+def squeeze_releases(df: pd.DataFrame, bb_w: int, sq_thr: float, min_dur: int,
+                     split: int) -> list[int]:
+    """Indici delle barre di rilascio squeeze nella finestra test (idx >= split)."""
+    close = df["close"].values
+    high = df["high"].values
+    low = df["low"].values
+    kcr = keltner_ratio(close, high, low, bb_w)
+    rels: list[int] = []
+    in_sq = False
+    sq_start = 0
+    for i in range(bb_w + 1, len(df)):
+        if np.isnan(kcr[i]):
+            continue
+        is_sq = kcr[i] < sq_thr
+        if is_sq and not in_sq:
+            in_sq, sq_start = True, i
+        elif not is_sq and in_sq:
+            in_sq = False
+            if i - sq_start >= min_dur and i >= split:
+                rels.append(i)
+    return rels
+
+
+def honest_entries(df: pd.DataFrame, rels: list[int], rule: str, mom: int = 4) -> list[tuple[int, int]]:
+    """Direzione da regole honest (solo dati <= i-1) o baseline breakout.
+
+    breakout: sign(close[i]-close[i-1])  -> conoscibile solo a close[i] (= live attuale)
+    premom:   sign(close[i-1]-close[i-1-mom]) -> trend pre-release, 100% honest
+    fade:     -sign(close[i]-close[i-1]) -> mean-reversion del breakout
+    """
+    close = df["close"].values
+    out: list[tuple[int, int]] = []
+    for i in rels:
+        if i - 1 - mom < 0:
+            continue
+        if rule == "premom":
+            d = np.sign(close[i - 1] - close[i - 1 - mom])
+        elif rule == "fade":
+            d = -np.sign(close[i] - close[i - 1])
+        else:  # breakout
+            d = np.sign(close[i] - close[i - 1])
+        if d != 0:
+            out.append((i, int(d)))
+    return out
+
+
+def main():
+    cfg = yaml.safe_load((PROJECT_ROOT / "strategies.yml").read_text())
+    defaults = cfg.get("defaults", {})
+    hold = defaults.get("hold_bars", 3)
+    lev = defaults.get("leverage", 3)
+    fee_rt = 0.002
+
+    fee_grid = [0.0, 0.0005, 0.001, 0.0015, 0.002]
+
+    # ---- (b) SENSIBILITA' ALLE FEE (config live, ingresso close[i]) ----
+    print("=" * 104)
+    print(f"  (b) SENSIBILITA' ALLE FEE — config live, ingresso close[i] | OOS {int(TEST_FRAC*100)}% | hold={hold} leva={lev}x")
+    print("=" * 104)
+    print(f"  {'Strategia':<26s}{'Asset':>5s}{'Trd':>5s}{'Lordo€':>9s}"
+          + "".join(f"{f'{f*100:.2f}%':>10s}" for f in fee_grid))
+    print("  " + "-" * 100)
+
+    for entry in cfg.get("strategies", []):
+        if not entry.get("enabled", True):
+            continue
+        name, asset, tf = entry["name"], entry["asset"], entry["tf"]
+        pos = entry.get("position_size", defaults.get("position_size", 0.15))
+        params = dict(entry.get("params", {}))
+        params["asset"], params["tf"] = asset, tf
+
+        df = load_data(asset, tf).reset_index(drop=True)
+        split = int(len(df) * (1 - TEST_FRAC))
+        close = df["close"].values
+        entries = (ml_entries(df, params, split, hold)[0] if name.startswith("ML01")
+                   else rule_entries(name, df, params, split))
+
+        gross = simulate(entries, close, hold, 0.0, lev, pos)["net_eur"]
+        rets = [simulate(entries, close, hold, f, lev, pos)["net_return_pct"] for f in fee_grid]
+        print(f"  {name:<26s}{asset:>5s}{len(entries):>5d}{gross:>+9.0f}"
+              + "".join(f"{r:>+10.1f}" for r in rets))
+
+    print("  " + "-" * 100)
+    print("  Colonne = Ret% netto al variare della fee RT. 0.00% isola l'edge puro (senza costi).")
+    print("  Deribit perp reale: taker ~0.10% RT, maker ~0%. Il modello live usa 0.20% RT.")
+
+    # ---- (a) HONEST-ENTRY squeeze: direzione decisa <= i-1, ingresso close[i] ----
+    print("\n" + "=" * 104)
+    print(f"  (a) HONEST-ENTRY squeeze (bb14 sq0.8 dur>=5) — ingresso close[i], fee={fee_rt*100:.1f}% RT")
+    print("=" * 104)
+    print(f"  {'Asset':>5s}{'Regola direzione':>20s}{'Trd':>6s}{'Win%g':>8s}{'Win%n':>8s}{'Netto€':>9s}{'Ret%':>9s}{'DD%':>7s}")
+    print("  " + "-" * 100)
+
+    rules = [("breakout (=live)", "breakout"), ("pre-trend mom4", "premom"),
+             ("pre-trend mom8", "premom8"), ("fade breakout", "fade")]
+    for asset in ["BTC", "ETH"]:
+        df = load_data(asset, "15m").reset_index(drop=True)
+        split = int(len(df) * (1 - TEST_FRAC))
+        close = df["close"].values
+        rels = squeeze_releases(df, 14, 0.8, 5, split)
+        for label, rule in rules:
+            mom = 8 if rule == "premom8" else 4
+            ents = honest_entries(df, rels, "premom" if rule == "premom8" else rule, mom=mom)
+            r = simulate(ents, close, hold, fee_rt, lev, 0.15)
+            print(f"  {asset:>5s}{label:>20s}{r['trades']:>6d}{r['win_gross']:>8.1f}"
+                  f"{r['win_net']:>8.1f}{r['net_eur']:>+9.0f}{r['net_return_pct']:>+9.1f}{r['max_dd']:>7.1f}")
+        print("  " + "-" * 100)
+    print("  pre-trend = direzione dal trend PRIMA del rilascio (solo dati <= i-1): 100% honest.")
+    print("  Se nessuna regola honest batte ~breakeven, non esiste edge direzionale tradeable.")
+
+
+if __name__ == "__main__":
+    main()

scripts/analysis/pairs_research.py

@@ -0,0 +1,133 @@
+"""Verifica indipendente + ricerca PAIRS / SPREAD MEAN-REVERSION fra cripto.
+
+Famiglia nuova market-neutral (distinta da tutto l'esistente, single-asset).
+Idea: il log-ratio di due cripto oscilla attorno alla media; z-score estremo -> rientra.
+
+Engine ONESTO (no look-ahead, verificato):
+  - r[i] = log(closeA[i]/closeB[i]); ma/sd = rolling(n) su r -> usano solo r[<=i].
+  - z[i] = (r[i]-ma[i])/sd[i]. ENTRY a close[i] (eseguibile):
+      z<=-z_in -> LONG ratio (long A / short B);  z>=+z_in -> SHORT ratio.
+  - EXIT quando |z[j]| <= z_exit (rientro) o time-limit max_bars, a close[j].
+  - pairs = 2 GAMBE -> fee = 2*fee_rt*lev (0.20% RT/coppia a fee_rt=0.001), il doppio
+    del single-asset. Rendimento neutral = retA*d - retB*d (notional uguale per gamba).
+  - non-overlap, capitale composto. Filtro candele sporche: salta salti |dr|>jump_max.
+  - Ritorno riportato come CAGR e Sharpe ANNUALIZZATO sul tempo reale (no sqrt(n_trade)).
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from src.data.downloader import load_data
+
+FEE_RT, LEV, POS, OOS_FRAC = 0.001, 3.0, 0.15, 0.30
+BARS_YEAR = 8760  # 1h
+
+
+def aligned(a: str, b: str, tf: str = "1h"):
+    da = load_data(a, tf)[["timestamp", "open", "high", "low", "close"]].rename(columns=lambda x: x + "_a" if x != "timestamp" else x)
+    db = load_data(b, tf)[["timestamp", "close"]].rename(columns={"close": "close_b"})
+    m = da.merge(db, on="timestamp", how="inner").reset_index(drop=True)
+    m["dt"] = pd.to_datetime(m["timestamp"], unit="ms", utc=True)
+    return m
+
+
+def pairs_sim(a, b, tf="1h", n=50, z_in=2.0, z_exit=0.5, max_bars=72,
+              jump_max=0.08, fee_rt=FEE_RT, lev=LEV, pos=POS, split_frac=0.0):
+    m = aligned(a, b, tf)
+    ca, cb = m["close_a"].values, m["close_b"].values
+    r = np.log(ca / cb)
+    dr = np.abs(np.diff(r, prepend=r[0]))               # salto 1-bar del log-ratio
+    ma = pd.Series(r).rolling(n).mean().values
+    sd = pd.Series(r).rolling(n).std().values
+    z = (r - ma) / np.where(sd == 0, np.nan, sd)        # causale: usa r[<=i]
+    ts = m["dt"]; N = len(r)
+    split = int(N * split_frac)
+    fee = 2 * fee_rt * lev                               # 2 gambe
+    cap = peak = 1000.0; dd = 0.0; last = -1
+    trades = wins = 0; rets = []; yearly = {}
+    eq_ts: list = []; eq_v: list = []
+    for i in range(n + 1, N - 1):
+        if i < split or np.isnan(z[i]) or dr[i] > jump_max:
+            continue
+        if i <= last:
+            continue
+        if z[i] <= -z_in:
+            d = 1
+        elif z[i] >= z_in:
+            d = -1
+        else:
+            continue
+        # exit: |z|<=z_exit o max_bars
+        j = min(i + max_bars, N - 1)
+        for k in range(1, max_bars + 1):
+            jj = i + k
+            if jj >= N:
+                j = N - 1; break
+            if abs(z[jj]) <= z_exit:
+                j = jj; break
+            j = jj
+        retA = (ca[j] - ca[i]) / ca[i]
+        retB = (cb[j] - cb[i]) / cb[i]
+        ret = (retA - retB) * d * lev - fee             # long A / short B (o viceversa)
+        cap = max(cap + cap * pos * ret, 10.0)
+        peak = max(peak, cap); dd = max(dd, (peak - cap) / peak)
+        trades += 1; wins += ret > 0; rets.append(ret * pos); last = j
+        eq_ts.append(ts.iloc[j]); eq_v.append(cap)
+        yearly[ts.iloc[i].year] = yearly.get(ts.iloc[i].year, 0.0) + ret * 100
+    yrs_span = (ts.iloc[-1] - ts.iloc[max(split, 0)]).days / 365.25 or 1
+    sharpe = float(np.mean(rets) / np.std(rets) * np.sqrt(BARS_YEAR / np.mean([max_bars])) ) if len(rets) > 1 and np.std(rets) > 0 else 0.0
+    # Sharpe annualizzato sul tempo reale: usa rendimenti per-trade scalati alla frequenza media
+    if len(rets) > 1 and np.std(rets) > 0:
+        trades_per_year = trades / yrs_span
+        sharpe = float(np.mean(rets) / np.std(rets) * np.sqrt(trades_per_year))
+    ret_tot = (cap / 1000 - 1) * 100
+    cagr = ((cap / 1000) ** (1 / yrs_span) - 1) * 100 if cap > 0 else -100
+    return dict(trades=trades, win=wins / trades * 100 if trades else 0, ret=ret_tot, cagr=cagr,
+                dd=dd * 100, sharpe=sharpe, yearly=yearly, eq_ts=eq_ts, eq_v=eq_v)
+
+
+def check_no_lookahead():
+    """Perturba il FUTURO del ratio e verifica che z[i] non cambi (causalita')."""
+    m = aligned("ETH", "BTC")
+    r = np.log(m["close_a"].values / m["close_b"].values)
+    n = 50; i = 1000
+    z_i = (r[i] - pd.Series(r).rolling(n).mean().values[i]) / pd.Series(r).rolling(n).std().values[i]
+    r2 = r.copy(); r2[i + 1:] += 0.5      # stravolge il futuro
+    z_i2 = (r2[i] - pd.Series(r2).rolling(n).mean().values[i]) / pd.Series(r2).rolling(n).std().values[i]
+    print(f"  no-look-ahead: z[i]={z_i:.6f} vs z[i] con futuro perturbato={z_i2:.6f} -> "
+          f"{'OK (identico)' if abs(z_i - z_i2) < 1e-9 else 'VIOLAZIONE!'}")
+
+
+def main():
+    print("=" * 104)
+    print(f"  PAIRS spread reversion — NETTO fee 0.20% RT/coppia (2 gambe), leva {LEV:.0f}x, OOS ultimo {int(OOS_FRAC*100)}%")
+    print("=" * 104)
+    check_no_lookahead()
+    pairs = [("ETH", "BTC"), ("LTC", "ETH"), ("ADA", "ETH"), ("SOL", "ETH"),
+             ("BNB", "BTC"), ("XRP", "BTC"), ("SOL", "BTC"), ("DOGE", "BTC")]
+    print(f"\n  {'coppia':<10s}{'trd':>5s}{'win%':>6s}{'FULL%':>8s}{'OOS%':>8s}{'CAGR%':>7s}"
+          f"{'DD%':>6s}{'oDD%':>6s}{'Shrp':>6s}{'anni+':>7s}{'fee0.4%RT':>11s}")
+    print("  " + "-" * 96)
+    for a, b in pairs:
+        f = pairs_sim(a, b, n=50, z_in=2.0, z_exit=0.5, max_bars=72)
+        o = pairs_sim(a, b, n=50, z_in=2.0, z_exit=0.5, max_bars=72, split_frac=1 - OOS_FRAC)
+        hi = pairs_sim(a, b, n=50, z_in=2.0, z_exit=0.5, max_bars=72, fee_rt=0.002)  # 0.4% RT/coppia
+        yrs = f["yearly"]; pos_y = sum(1 for v in yrs.values() if v > 0)
+        print(f"  {a+'/'+b:<10s}{f['trades']:>5d}{f['win']:>6.1f}{f['ret']:>+8.0f}{o['ret']:>+8.0f}"
+              f"{f['cagr']:>7.0f}{f['dd']:>6.0f}{o['dd']:>6.0f}{f['sharpe']:>6.2f}{f'{pos_y}/{len(yrs)}':>7s}"
+              f"{hi['ret']:>+11.0f}")
+    # correlazione con BTC daily (market-neutrality) sulla coppia migliore
+    print("\n  Verifica market-neutrality ETH/BTC: per-anno")
+    f = pairs_sim("ETH", "BTC", n=50, z_in=2.0, z_exit=0.5, max_bars=72)
+    print("  " + "  ".join(f"{y}:{v:+.0f}%" for y, v in sorted(f["yearly"].items())))
+
+
+if __name__ == "__main__":
+    main()

scripts/analysis/report.py

@@ -0,0 +1,136 @@
+"""Report aggiornato: risultati per anno + numero trade per anno, tutte le strategie.
+
+Sezioni:
+  (A) RET% NETTO per anno — ogni strategia singola + i portafogli (FADE / HONEST /
+      MASTER equal / MASTER 50-50). Ret% dai rendimenti giornalieri composti.
+  (B) NUMERO TRADE per anno — per ogni strategia singola. Per le fade e DIP01 è il
+      numero di ingressi; per TR01 e ROT02 (posizione continua) è il numero di
+      ribilanciamenti/cambi di stato nell'anno.
+  (C) RIEPILOGO — TOT%, CAGR, DD, Sharpe (FULL e OOS) dei portafogli.
+
+Tutto NETTO fee 0.10% RT, leva 3x, pos 15%/sleeve. Finestra comune 2021-2026,
+OOS = ultimo 30%. Config = quella deployata (MR03/ROT01 in waste; ROT02 top_k=3).
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from src.data.downloader import load_data
+from scripts.analysis.combine_portfolio import (
+    build_all_sleeves, port_returns, yearly_returns, metrics, SPLIT, OOS_DATE, IDX,
+)
+from scripts.analysis.risk_management import strats_for, build_trades
+from scripts.analysis.honest_lab import get_df, ema, FEE_RT, LEV, POS
+from scripts.analysis.honest_improve import rot_improved
+from scripts.analysis.honest_improve2 import dip_market_gated
+
+YEARS = sorted(set(IDX.year))
+
+
+# ---------------- trade per anno, per tipo di strategia ----------------
+def fade_trades_year(asset, fn, params) -> dict[int, int]:
+    df = load_data(asset, "1h")
+    ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+    out: dict[int, int] = {}
+    for i, j, ret in build_trades(fn(df, **params), df, trend_max=3.0):
+        y = ts.iloc[i].year
+        out[y] = out.get(y, 0) + 1
+    return out
+
+
+def dip_trades_year() -> dict[int, int]:
+    d = dip_market_gated("BTC", market_n=0)
+    # yt[anno] = lista dei trade dell'anno -> il conteggio e' la lunghezza
+    return {int(y): (len(v) if isinstance(v, (list, tuple)) else int(v)) for y, v in d["yt"].items()}
+
+
+def tr_rebalances_year(assets) -> dict[int, int]:
+    """Cambi di stato (entra/esce dal trend) per anno, sommati sul paniere TR01."""
+    out: dict[int, int] = {}
+    for a in assets:
+        df = get_df(a, "4h"); c = df["close"].values
+        ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+        ef, es = ema(c, 20), ema(c, 100)
+        sig = np.where(ef > es, 1.0, 0.0); sig[:100] = 0.0
+        for i in range(1, len(c)):
+            if sig[i] != sig[i - 1]:
+                y = ts.iloc[i].year
+                out[y] = out.get(y, 0) + 1
+    return out
+
+
+def rot_rebalances_year() -> dict[int, int]:
+    r = rot_improved(lookback=60, top_k=3, regime_n=100)
+    return {int(y): int(n) for y, n in r["reb"].items()}
+
+
+def main():
+    print("Costruzione equity e conteggi (puo' richiedere ~1 min)...\n")
+    S = build_all_sleeves()
+    fade = {k: v for k, v in S.items() if k.startswith("MR")}
+    honest = {k: v for k, v in S.items() if not k.startswith("MR")}
+
+    # rendimenti giornalieri per Ret%/anno
+    sleeve_ret = {k: v.pct_change().fillna(0.0) for k, v in S.items()}
+    ports = {
+        "FADE": port_returns(fade),
+        "HONEST": port_returns(honest),
+        "MASTEReq": port_returns(S),
+        "MAST5050": (port_returns(fade) + port_returns(honest)) / 2,
+    }
+
+    # ---- (A) RET% per anno ----
+    cols_A = list(S) + list(ports)
+    rety = {**{k: yearly_returns(v) for k, v in sleeve_ret.items()},
+            **{k: yearly_returns(v) for k, v in ports.items()}}
+    print("=" * 132)
+    print("  (A) RET% NETTO PER ANNO — strategie singole e portafogli | leva 3x pos 15% fee 0.10% RT")
+    print("=" * 132)
+    print(f"  {'Anno':>5s}" + "".join(f"{c.replace('_',''):>11s}" for c in cols_A))
+    print("  " + "-" * 126)
+    for y in YEARS:
+        print(f"  {y:>5d}" + "".join(f"{rety[c].get(y, 0):>+11.0f}" for c in cols_A))
+
+    # ---- (B) NUMERO TRADE per anno ----
+    tcounts = {}
+    for asset in ["BTC", "ETH"]:
+        for nm, (fn, params) in strats_for(asset).items():
+            tcounts[f"{nm}_{asset}"] = fade_trades_year(asset, fn, params)
+    tcounts["DIP01_BTC"] = dip_trades_year()
+    tcounts["TR01_basket*"] = tr_rebalances_year(["BNB", "BTC", "DOGE", "SOL", "XRP"])
+    tcounts["ROT02_rot*"] = rot_rebalances_year()
+    cols_B = list(tcounts)
+    print("\n" + "=" * 132)
+    print("  (B) NUMERO TRADE PER ANNO — fade/DIP01 = ingressi; TR01/ROT02 (*) = ribilanciamenti")
+    print("=" * 132)
+    print(f"  {'Anno':>5s}" + "".join(f"{c.replace('_',''):>13s}" for c in cols_B))
+    print("  " + "-" * 126)
+    for y in YEARS:
+        print(f"  {y:>5d}" + "".join(f"{tcounts[c].get(y, 0):>13d}" for c in cols_B))
+    print("  " + "-" * 126)
+    print(f"  {'TOT':>5s}" + "".join(f"{sum(tcounts[c].values()):>13d}" for c in cols_B))
+
+    # ---- (C) riepilogo portafogli ----
+    print("\n" + "=" * 92)
+    print(f"  (C) RIEPILOGO PORTAFOGLI | OOS da {OOS_DATE}")
+    print("=" * 92)
+    print(f"  {'portafoglio':<14s}{'Ret%':>9s}{'CAGR':>7s}{'DD%':>7s}{'Shrp':>7s}"
+          f"   | {'oRet%':>9s}{'oDD%':>7s}{'oShrp':>7s}")
+    print("  " + "-" * 74)
+    for name, pr in ports.items():
+        f, o = metrics(pr), metrics(pr, lo=SPLIT)
+        print(f"  {name:<14s}{f['ret']:>+9.0f}{f['cagr']:>7.0f}{f['dd']:>7.1f}{f['sharpe']:>7.2f}"
+              f"   | {o['ret']:>+9.0f}{o['dd']:>7.1f}{o['sharpe']:>7.2f}")
+    print("\n  MASTEReq (9 sleeve) = configurazione consigliata. (*) TR01/ROT02 = posizione")
+    print("  continua: il conteggio e' il numero di ribilanciamenti/cambi di stato, non di trade discreti.")
+
+
+if __name__ == "__main__":
+    main()

scripts/analysis/report_families.py

@@ -0,0 +1,136 @@
+"""Report riassuntivo: tutte le strategie/famiglie per anno + analisi di integrazione.
+
+Consolida in un solo posto:
+  (A) RET% NETTO per anno per FAMIGLIA (FADE / HONEST / PAIRS / TSM01) e per i portafogli.
+  (B) RET% NETTO per anno per ogni STRATEGIA singola (tutti gli sleeve).
+  (C) INTEGRAZIONE: cosa succede al MASTER aggiungendo le nuove famiglie (pairs, TSM01).
+  (D) Numeri SOBRI (worst-case) e raccomandazione operativa.
+
+Famiglie:
+  FADE   (reversione intraday 1h, long/short, BTC/ETH): MR01, MR02, MR07
+  HONEST (long-only multi-regime multi-crypto): DIP01, TR01, ROT02
+  PAIRS  (market-neutral spread reversion, config universale): 5 coppie
+  TSM01  (TSMOM multi-orizzonte, diversificatore)
+
+Tutto NETTO fee, leva 3x (vedi nota sobria leva 2x), finestra comune 2021-2026, OOS=ultimo 30%.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.combine_portfolio import (
+    build_all_sleeves, port_returns, metrics, yearly_returns, SPLIT, OOS_DATE, IDX,
+)
+from scripts.analysis.honest_improve2 import _daily_equity, _norm
+from scripts.analysis.pairs_research import pairs_sim
+from scripts.analysis.tsmom_research import tsmom_sim
+from scripts.strategies.PR01_pairs_reversion import PAIRS
+
+YEARS = sorted(set(IDX.year))
+
+
+def daily_from(eq_ts, eq_v):
+    return _norm(_daily_equity(eq_ts, eq_v, IDX))
+
+
+def build_everything():
+    S = build_all_sleeves()                       # 9 sleeve (FADE 6 + HONEST 3)
+    pairs = {}
+    for a, b, p in PAIRS:
+        r = pairs_sim(a, b, **p)
+        pairs[f"PR_{a}{b}"] = daily_from(r["eq_ts"], r["eq_v"])
+    t = tsmom_sim()
+    tsm = {"TSM01": daily_from(t["eq_ts"], t["eq_v"])}
+    return S, pairs, tsm
+
+
+def yrow(label, dr):
+    yr = yearly_returns(dr)
+    return f"  {label:<14s}" + "".join(f"{yr.get(y, 0):>+9.0f}" for y in YEARS)
+
+
+def metric_block(label, dr):
+    f, o = metrics(dr), metrics(dr, lo=SPLIT)
+    return (f"  {label:<16s}{f['ret']:>+9.0f}{f['cagr']:>7.0f}{f['dd']:>7.1f}{f['sharpe']:>7.2f}"
+            f"   | {o['ret']:>+9.0f}{o['dd']:>7.1f}{o['sharpe']:>7.2f}")
+
+
+def main():
+    print("Costruzione (puo' richiedere ~1-2 min)...\n")
+    S, pairs, tsm = build_everything()
+    fade = {k: v for k, v in S.items() if k.startswith("MR")}
+    honest = {k: v for k, v in S.items() if not k.startswith("MR")}
+
+    fam = {
+        "FADE": port_returns(fade),
+        "HONEST": port_returns(honest),
+        "PAIRS": port_returns(pairs),
+        "TSM01": tsm["TSM01"].pct_change().fillna(0.0),
+    }
+    master9 = port_returns(S)
+    master_p = port_returns({**S, **pairs})
+    master_x = port_returns({**S, **pairs, **tsm})
+
+    # ---------- (A) per anno, per FAMIGLIA + portafogli ----------
+    print("=" * 110)
+    print("  (A) RET% NETTO PER ANNO — per FAMIGLIA e per PORTAFOGLIO | leva 3x, fee netta")
+    print("=" * 110)
+    print(f"  {'':<14s}" + "".join(f"{y:>9d}" for y in YEARS))
+    print("  " + "-" * 104)
+    for k, dr in fam.items():
+        print(yrow(k, dr))
+    print("  " + "-" * 104)
+    print(yrow("MASTER-9", master9))
+    print(yrow("MASTER+pairs", master_p))
+    print(yrow("MASTER-esteso", master_x))
+
+    # ---------- (B) per anno, per STRATEGIA singola ----------
+    print("\n" + "=" * 130)
+    print("  (B) RET% NETTO PER ANNO — per STRATEGIA singola (tutti gli sleeve)")
+    print("=" * 130)
+    allsl = {**S, **pairs, **tsm}
+    cols = list(allsl)
+    print(f"  {'Anno':>5s}" + "".join(f"{c.replace('_',''):>11s}" for c in cols))
+    print("  " + "-" * 124)
+    yr_each = {k: yearly_returns(v.pct_change().fillna(0.0)) for k, v in allsl.items()}
+    for y in YEARS:
+        print(f"  {y:>5d}" + "".join(f"{yr_each[c].get(y, 0):>+11.0f}" for c in cols))
+
+    # ---------- (C) integrazione ----------
+    print("\n" + "=" * 96)
+    print(f"  (C) INTEGRAZIONE delle nuove famiglie nel MASTER | OOS da {OOS_DATE} | equal-weight daily")
+    print("=" * 96)
+    print(f"  {'portafoglio':<16s}{'Ret%':>9s}{'CAGR':>7s}{'DD%':>7s}{'Shrp':>7s}"
+          f"   | {'oRet%':>9s}{'oDD%':>7s}{'oShrp':>7s}")
+    print("  " + "-" * 80)
+    print(metric_block("MASTER-9", master9))
+    print(metric_block("+pairs", master_p))
+    print(metric_block("+TSM01", port_returns({**S, **tsm})))
+    print(metric_block("MASTER-esteso", master_x))
+    # correlazione media nuove vs master-9
+    dr_all = pd.DataFrame({k: v.pct_change().fillna(0.0) for k, v in {**S, **pairs, **tsm}.items()})
+    corr = dr_all.corr(); old = list(S)
+    print("  " + "-" * 80)
+    for k in list(pairs) + list(tsm):
+        print(f"  corr {k:<11s} vs MASTER-9 = {corr.loc[k, old].mean():+.2f}")
+
+    # ---------- (D) numeri sobri ----------
+    print("\n" + "=" * 96)
+    print("  (D) NUMERI SOBRI / RACCOMANDAZIONE (anti-overfit)")
+    print("=" * 96)
+    print("  - L'OOS singolo (2024-25) e' regime calmo -> Sharpe/DD OOS ottimistici ~50%.")
+    print("  - Numeri onesti del MASTER-esteso: worst-DD 90g ~6%, Sharpe atteso ~5, ogni anno positivo dal 2021.")
+    print("  - Regge leva 2x + slippage doppio (CAGR ~36%, Sharpe ~5).")
+    print("  - Rischio concentrato sui PAIRS (~57%) -> cap allocazione pairs ~30-35%.")
+    print("  - I pairs sono a 2 gambe (long/short): il worker live va esteso prima del trading reale.")
+    print("  - CONFIG RACCOMANDATA: MASTER-esteso, equal-weight, leva 2x, cap pairs 30-35%.")
+
+
+if __name__ == "__main__":
+    main()

scripts/analysis/risk_management.py

@@ -0,0 +1,261 @@
+"""Gestione del rischio sulle fade (MR01/MR02/MR03/MR07): alzare Acc, ridurre DD.
+
+Due analisi, ognuna misurata FULL e OOS (ultimo 30%) per non illudersi:
+
+  (A) SCREENING LEVE — confronta su ogni strategia le leve di rischio:
+      - vol-target sizing (size ~ 1/distanza-SL)         -> SCARTATA (peggiora)
+      - skip alta volatilita' (ATR% in coda alta)        -> SCARTATA (peggiora)
+      - filtro trend (|close-EMA200|/ATR oltre soglia)   -> ADOTTATA (Acc+ DD-)
+      - combinazione di tutte
+
+  (B) FILTRO TREND + PORTAFOGLIO:
+      - sweep della soglia trend (assoluta in ATR, regola unica = no overfit)
+      - portafoglio equipesato su sotto-conti indipendenti: curve poco correlate
+        -> DD aggregato << DD del singolo sleeve (vera leva anti-drawdown)
+
+Engine fedele: ingresso close[i], exit TP/SL intrabar (high/low) o time-limit,
+non-overlap, capitale composto. Numeri NETTI fee 0.10% RT, leva 3x.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from src.data.downloader import load_data
+from scripts.analysis.strategy_research import bollinger_fade, atr
+from scripts.analysis.strategy_research_v2 import donchian_fade, return_reversal
+
+FEE_RT, LEV, POS, INIT, OOS_FRAC = 0.001, 3.0, 0.15, 1000.0, 0.30
+
+# config base di ogni strategia (come strategies.yml).
+# NB: MR03 keltner_fade spostata in scripts/waste/ (fade piu' debole, ridondante
+# con MR01); la funzione keltner_fade resta in strategy_research_v2 come record.
+STRATS = {
+    "MR01": (bollinger_fade, dict(n=50, k=2.5, sl_atr=2.0, max_bars=24)),
+    "MR02": (donchian_fade,  dict(n=20, sl_atr=2.0, max_bars=24)),
+    "MR07": (return_reversal,dict(n=50, k=3.5, tp_atr=2.0, sl_atr=1.5, max_bars=24)),
+}
+STRATS_ETH = dict(STRATS)
+
+
+def strats_for(asset: str) -> dict:
+    return STRATS_ETH if asset == "ETH" else STRATS
+
+
+# ============================ (A) SCREENING LEVE ============================
+def add_context(ents, df, ema_long=200):
+    """Aggiunge a ogni entry: sl_dist, atr_pct, trend_dist (|close-EMA|/ATR)."""
+    c = df["close"].values
+    a = atr(df, 14)
+    el = pd.Series(c).ewm(span=ema_long, adjust=False).mean().values
+    apct = a / c
+    for e in ents:
+        i = e["i"]
+        e["sl_dist"] = abs(c[i] - e["sl"]) / c[i]
+        e["atr_pct"] = apct[i]
+        e["trend_dist"] = abs(c[i] - el[i]) / a[i] if a[i] else 0.0
+    return ents
+
+
+def simulate(ents, df, fee_rt=FEE_RT, lev=LEV, split=-1,
+             sizer=None, vol_skip=None, trend_skip=None, max_size=0.30):
+    """sizer: funzione(entry)->frazione capitale; default POS fisso.
+    vol_skip: soglia atr_pct sopra cui salto. trend_skip: soglia trend_dist sopra cui salto."""
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    n = len(c)
+    ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+    cap = peak = INIT
+    dd = 0.0; last = -1; trd = wins = 0
+    fee = fee_rt * lev
+    yearly = {}; rets = []
+    for e in ents:
+        i, d = e["i"], e["d"]
+        if i <= last or i + 1 >= n or i < split:
+            continue
+        if vol_skip is not None and e["atr_pct"] > vol_skip:
+            continue
+        if trend_skip is not None and e["trend_dist"] > trend_skip:
+            continue
+        entry = c[i]; tp, sl, mb = e["tp"], e["sl"], e["max_bars"]
+        exit_p = c[min(i + mb, n - 1)]; j = min(i + mb, n - 1)
+        for k in range(1, mb + 1):
+            j = i + k
+            if j >= n:
+                exit_p = c[n - 1]; break
+            hs = (d == 1 and l[j] <= sl) or (d == -1 and h[j] >= sl)
+            ht = (d == 1 and h[j] >= tp) or (d == -1 and l[j] <= tp)
+            if hs: exit_p = sl; break
+            if ht: exit_p = tp; break
+            if k == mb: exit_p = c[j]
+        ret = (exit_p - entry) / entry * d * lev - fee
+        size = POS if sizer is None else min(sizer(e), max_size)
+        cap = max(cap + cap * size * ret, 10.0)
+        peak = max(peak, cap); dd = max(dd, (peak - cap) / peak)
+        trd += 1; wins += ret > 0; last = j; rets.append(ret * size)
+        y = ts.iloc[i].year; yearly[y] = yearly.get(y, 0.0) + ret * size * INIT
+    sharpe = float(np.mean(rets) / np.std(rets) * np.sqrt(len(rets))) if len(rets) > 1 and np.std(rets) > 0 else 0.0
+    return dict(trades=trd, acc=wins / trd * 100 if trd else 0.0,
+                ret=(cap / INIT - 1) * 100, dd=dd * 100, yearly=yearly, sharpe=sharpe)
+
+
+def vol_target_sizer(target=0.015):
+    """size t.c. rischio (size*lev*sl_dist) ~ target; piu' largo lo stop, meno size."""
+    return lambda e: target / (LEV * max(e["sl_dist"], 1e-4))
+
+
+def _line(label, full, oos):
+    print(f"  {label:<28s}{full['trades']:>6d}{full['acc']:>7.1f}{full['ret']:>+10.0f}{full['dd']:>7.1f}{full['sharpe']:>7.2f}"
+          f"  | {oos['trades']:>5d}{oos['acc']:>7.1f}{oos['ret']:>+9.0f}{oos['dd']:>7.1f}{oos['sharpe']:>7.2f}")
+
+
+def screen_levers():
+    print("=" * 110)
+    print("  (A) SCREENING LEVE — vol-target / vol-skip / filtro-trend | NETTO fee 0.10% RT, leva 3x")
+    print("=" * 110)
+    for asset in ["BTC", "ETH"]:
+        df = load_data(asset, "1h")
+        split = int(len(df) * (1 - OOS_FRAC))
+        print(f"\n  {asset} 1h")
+        print(f"  {'config':<28s}{'Trd':>6s}{'Acc%':>7s}{'Ret%':>10s}{'DD%':>7s}{'Shrp':>7s}"
+              f"  | {'oTrd':>5s}{'oAcc':>7s}{'oRet':>9s}{'oDD':>7s}{'oShrp':>7s}")
+        print("  " + "-" * 106)
+        for nm, (fn, params) in strats_for(asset).items():
+            ents = add_context(fn(df, **params), df)
+            p85 = float(np.quantile([e["atr_pct"] for e in ents], 0.85))
+            t90 = float(np.quantile([e["trend_dist"] for e in ents], 0.90))
+            _line(f"{nm} base", simulate(ents, df), simulate(ents, df, split=split))
+            _line(f"{nm} +volTarget", simulate(ents, df, sizer=vol_target_sizer()),
+                  simulate(ents, df, split=split, sizer=vol_target_sizer()))
+            _line(f"{nm} +volSkip(p85)", simulate(ents, df, vol_skip=p85),
+                  simulate(ents, df, split=split, vol_skip=p85))
+            _line(f"{nm} +trendSkip(p90)", simulate(ents, df, trend_skip=t90),
+                  simulate(ents, df, split=split, trend_skip=t90))
+            _line(f"{nm} +ALL", simulate(ents, df, sizer=vol_target_sizer(), vol_skip=p85, trend_skip=t90),
+                  simulate(ents, df, split=split, sizer=vol_target_sizer(), vol_skip=p85, trend_skip=t90))
+            print("  " + "-" * 106)
+    print("\n  Esito: vol-target e vol-skip PEGGIORANO; il filtro trend e' l'unica leva utile.")
+
+
+# ===================== (B) FILTRO TREND + PORTAFOGLIO =====================
+def build_trades(ents, df, lev=LEV, fee_rt=FEE_RT, trend_max=None, ema_long=200):
+    """Lista trade non-overlap: (entry_idx, exit_idx, ret_netto). Filtro trend opzionale."""
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    n = len(c); a = atr(df, 14)
+    el = pd.Series(c).ewm(span=ema_long, adjust=False).mean().values
+    fee = fee_rt * lev
+    out = []; last = -1
+    for e in ents:
+        i, d = e["i"], e["d"]
+        if i <= last or i + 1 >= n:
+            continue
+        if trend_max is not None and a[i] and abs(c[i] - el[i]) / a[i] > trend_max:
+            continue
+        entry = c[i]; tp, sl, mb = e["tp"], e["sl"], e["max_bars"]
+        exit_p = c[min(i + mb, n - 1)]; j = min(i + mb, n - 1)
+        for k in range(1, mb + 1):
+            j = i + k
+            if j >= n:
+                exit_p = c[n - 1]; break
+            hs = (d == 1 and l[j] <= sl) or (d == -1 and h[j] >= sl)
+            ht = (d == 1 and h[j] >= tp) or (d == -1 and l[j] <= tp)
+            if hs: exit_p = sl; break
+            if ht: exit_p = tp; break
+            if k == mb: exit_p = c[j]
+        ret = (exit_p - entry) / entry * d * lev - fee
+        out.append((i, j, ret)); last = j
+    return out
+
+
+def metrics_single(trades, pos=POS, split=-1):
+    cap = peak = INIT; dd = 0.0; trd = wins = 0; rets = []
+    for i, j, ret in trades:
+        if i < split:
+            continue
+        cap = max(cap + cap * pos * ret, 10.0)
+        peak = max(peak, cap); dd = max(dd, (peak - cap) / peak)
+        trd += 1; wins += ret > 0; rets.append(ret * pos)
+    sh = float(np.mean(rets) / np.std(rets) * np.sqrt(len(rets))) if len(rets) > 1 and np.std(rets) > 0 else 0.0
+    return dict(trades=trd, acc=wins / trd * 100 if trd else 0.0,
+                ret=(cap / INIT - 1) * 100, dd=dd * 100, sharpe=sh)
+
+
+def sleeve_equity(trades, n_bars, pos=POS, split=-1):
+    """Equity di uno sleeve su sotto-conto indipendente (capitale INIT, pos fissa)."""
+    eq = np.full(n_bars, INIT, dtype=float)
+    cap = INIT
+    for i, j, ret in sorted(trades, key=lambda t: t[1]):
+        if i < split:
+            continue
+        cap = max(cap + cap * pos * ret, 10.0)
+        eq[j:] = cap
+    return eq
+
+
+def metrics_portfolio(strat_trades, n_bars, pos=POS, split=-1):
+    """Portafoglio equipesato: media di N sotto-conti indipendenti. DD sull'aggregata."""
+    sleeves = [sleeve_equity(tr, n_bars, pos=pos, split=split) for tr in strat_trades.values()]
+    agg = np.mean(sleeves, axis=0)
+    agg = agg[max(split, 0):]
+    peak = np.maximum.accumulate(agg)
+    dd = float(np.max((peak - agg) / peak) * 100)
+    trd = sum(1 for tr in strat_trades.values() for i, _, _ in tr if i >= split)
+    wins = sum(1 for tr in strat_trades.values() for i, _, r in tr if i >= split and r > 0)
+    return dict(trades=trd, acc=wins / trd * 100 if trd else 0.0, ret=(agg[-1] / INIT - 1) * 100, dd=dd)
+
+
+def trend_and_portfolio():
+    # --- sweep soglia trend ---
+    print("\n" + "=" * 104)
+    print("  (B1) FILTRO TREND |close-EMA200|/ATR > soglia -> SALTA | NETTO fee 0.10% RT, leva 3x")
+    print("=" * 104)
+    print(f"  {'Strat/Asset':<14s}{'soglia':>8s}{'Trd':>6s}{'Acc%':>7s}{'Ret%':>9s}{'DD%':>7s}"
+          f"  | {'oAcc':>6s}{'oRet':>9s}{'oDD':>7s}{'oShrp':>7s}")
+    print("  " + "-" * 100)
+    for asset in ["BTC", "ETH"]:
+        df = load_data(asset, "1h"); split = int(len(df) * (1 - OOS_FRAC))
+        for nm, (fn, params) in strats_for(asset).items():
+            ents = fn(df, **params)
+            for thr in [None, 4.0, 3.0, 2.5, 2.0]:
+                tr = build_trades(ents, df, trend_max=thr)
+                f = metrics_single(tr); o = metrics_single(tr, split=split)
+                lab = "base" if thr is None else f"{thr}ATR"
+                print(f"  {nm+' '+asset:<14s}{lab:>8s}{f['trades']:>6d}{f['acc']:>7.1f}{f['ret']:>+9.0f}{f['dd']:>7.1f}"
+                      f"  | {o['acc']:>6.1f}{o['ret']:>+9.0f}{o['dd']:>7.1f}{o['sharpe']:>7.2f}")
+            print("  " + "-" * 100)
+
+    # --- portafoglio equipesato (filtro trend 3.0 ATR) ---
+    print("\n" + "=" * 104)
+    print("  (B2) PORTAFOGLIO equipesato: N sotto-conti indipendenti (pos 0.15, filtro trend 3.0 ATR)")
+    print("=" * 104)
+    print(f"  {'Universo':<26s}{'Trd':>6s}{'Acc%':>7s}{'Ret%':>10s}{'DD%':>7s}"
+          f"  | {'oAcc':>6s}{'oRet':>9s}{'oDD':>7s}")
+    print("  " + "-" * 100)
+    all_trades = {}
+    for asset in ["BTC", "ETH"]:
+        df = load_data(asset, "1h"); split = int(len(df) * (1 - OOS_FRAC)); n = len(df)
+        st = {f"{nm}_{asset}": build_trades(fn(df, **p), df, trend_max=3.0) for nm, (fn, p) in strats_for(asset).items()}
+        all_trades.update(st)
+        f = metrics_portfolio(st, n); o = metrics_portfolio(st, n, split=split)
+        print(f"  {'Portafoglio '+asset+' (4 strat)':<26s}{f['trades']:>6d}{f['acc']:>7.1f}{f['ret']:>+10.0f}{f['dd']:>7.1f}"
+              f"  | {o['acc']:>6.1f}{o['ret']:>+9.0f}{o['dd']:>7.1f}")
+    df0 = load_data("BTC", "1h"); split0 = int(len(df0) * (1 - OOS_FRAC))
+    f = metrics_portfolio(all_trades, len(df0)); o = metrics_portfolio(all_trades, len(df0), split=split0)
+    print("  " + "-" * 100)
+    print(f"  {'GLOBALE BTC+ETH (8 sleeve)':<26s}{f['trades']:>6d}{f['acc']:>7.1f}{f['ret']:>+10.0f}{f['dd']:>7.1f}"
+          f"  | {o['acc']:>6.1f}{o['ret']:>+9.0f}{o['dd']:>7.1f}")
+    print("\n  Curve poco correlate => DD aggregato molto piu' basso del singolo sleeve.")
+
+
+def main():
+    screen_levers()
+    trend_and_portfolio()
+
+
+if __name__ == "__main__":
+    main()

scripts/analysis/strategy_research.py

@@ -0,0 +1,258 @@
+"""Ricerca strategie fee-aware, OOS, oltre la famiglia squeeze.
+
+Lezioni apprese (squeeze breakout = nessun edge):
+  - le FEE sono vincolo di prim'ordine -> default fee realistica Deribit 0.10% RT
+    (taker 0.05%/lato, maker ~0%); poche operazioni meglio di molte
+  - i breakout RIENTRANO -> si esplora mean-reversion, non continuation
+  - ogni numero e' NETTO dopo fee+leva, su finestra held-out + per anno
+
+Engine realistico: ingresso a close[i] (eseguibile), uscita su TP/SL intrabar
+(high/low) o time-limit, una posizione per volta (non-overlap), capitale composto.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from src.data.downloader import load_data
+
+FEE_RT = 0.001          # Deribit perp realistico: taker 0.05%/lato
+LEV = 3.0
+POS = 0.15
+OOS_FRAC = 0.30
+BARS_PER_YEAR = {"15m": 35040, "1h": 8760, "4h": 2190, "1d": 365}
+
+
+# ----------------------------- dati -----------------------------
+def get_df(asset: str, tf: str) -> pd.DataFrame:
+    """tf nativo (15m,1h) o resample da 1h (4h,1d)."""
+    if tf in ("15m", "1h"):
+        return load_data(asset, tf).reset_index(drop=True)
+    base = load_data(asset, "1h").copy()
+    base["dt"] = pd.to_datetime(base["timestamp"], unit="ms", utc=True)
+    base = base.set_index("dt")
+    rule = {"4h": "4h", "1d": "1D"}[tf]
+    agg = base.resample(rule).agg(
+        {"open": "first", "high": "max", "low": "min", "close": "last", "volume": "sum"}
+    ).dropna()
+    agg["timestamp"] = agg.index.asi8 // 10**6
+    return agg.reset_index(drop=True)
+
+
+# --------------------------- indicatori ---------------------------
+def atr(df: pd.DataFrame, n: int = 14) -> np.ndarray:
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    pc = np.roll(c, 1); pc[0] = c[0]
+    tr = np.maximum(h - l, np.maximum(np.abs(h - pc), np.abs(l - pc)))
+    return pd.Series(tr).rolling(n).mean().values
+
+
+def rsi(close: np.ndarray, n: int = 14) -> np.ndarray:
+    d = np.diff(close, prepend=close[0])
+    up = pd.Series(np.where(d > 0, d, 0.0)).ewm(alpha=1/n, adjust=False).mean()
+    dn = pd.Series(np.where(d < 0, -d, 0.0)).ewm(alpha=1/n, adjust=False).mean()
+    rs = up / dn.replace(0, np.nan)
+    return (100 - 100 / (1 + rs)).values
+
+
+# --------------------------- engine ---------------------------
+def simulate(entries: list[dict], df: pd.DataFrame, fee_rt: float = FEE_RT,
+             lev: float = LEV, pos: float = POS) -> dict:
+    """entries: dict con i(idx), d(+1/-1), tp(prezzo), sl(prezzo), max_bars."""
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    n = len(c)
+    cap = peak = 1000.0
+    max_dd = 0.0
+    fee = fee_rt * lev
+    trades = wins = 0
+    last_exit = -1
+    bars_in = 0
+    ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+    yearly: dict[int, float] = {}
+
+    for e in entries:
+        i, d = e["i"], e["d"]
+        if i <= last_exit or i + 1 >= n:
+            continue
+        entry = c[i]
+        tp, sl, mb = e["tp"], e["sl"], e["max_bars"]
+        exit_p = c[min(i + mb, n - 1)]
+        for k in range(1, mb + 1):
+            j = i + k
+            if j >= n:
+                exit_p = c[n - 1]; break
+            hit_sl = (d == 1 and l[j] <= sl) or (d == -1 and h[j] >= sl)
+            hit_tp = (d == 1 and h[j] >= tp) or (d == -1 and l[j] <= tp)
+            if hit_sl:                      # conservativo: SL prima del TP nello stesso bar
+                exit_p = sl; break
+            if hit_tp:
+                exit_p = tp; break
+            if k == mb:
+                exit_p = c[j]
+        ret = (exit_p - entry) / entry * d * lev - fee
+        cb = cap
+        cap = max(cb + cb * pos * ret, 10.0)
+        peak = max(peak, cap); max_dd = max(max_dd, (peak - cap) / peak)
+        trades += 1; wins += ret > 0; bars_in += min(mb, j - i)
+        last_exit = j
+        yearly[ts.iloc[i].year] = yearly.get(ts.iloc[i].year, 0.0) + ret * 100
+    return {
+        "trades": trades,
+        "win": wins / trades * 100 if trades else 0.0,
+        "ret": (cap / 1000 - 1) * 100,
+        "dd": max_dd * 100,
+        "yearly": yearly,
+        "exposure": bars_in / n * 100,
+    }
+
+
+# --------------------------- strategie ---------------------------
+def bollinger_fade(df, n=20, k=2.0, sl_atr=2.0, max_bars=24):
+    """Mean-reversion: fada il close oltre la banda, TP alla media, SL = k_atr*ATR."""
+    c = df["close"].values
+    ma = pd.Series(c).rolling(n).mean().values
+    sd = pd.Series(c).rolling(n).std().values
+    a = atr(df, 14)
+    up, lo = ma + k * sd, ma - k * sd
+    ents = []
+    for i in range(n + 14, len(c)):
+        if np.isnan(up[i]) or np.isnan(a[i]):
+            continue
+        if c[i] < lo[i] and c[i - 1] >= lo[i - 1]:          # appena sotto la banda
+            ents.append({"i": i, "d": 1, "tp": ma[i], "sl": c[i] - sl_atr * a[i], "max_bars": max_bars})
+        elif c[i] > up[i] and c[i - 1] <= up[i - 1]:
+            ents.append({"i": i, "d": -1, "tp": ma[i], "sl": c[i] + sl_atr * a[i], "max_bars": max_bars})
+    return ents
+
+
+def rsi_revert(df, n=14, lo=30, hi=70, sl_atr=2.0, max_bars=24, ma_n=20):
+    """RSI mean-reversion: long su RSI<lo che risale, TP alla media mobile."""
+    c = df["close"].values
+    r = rsi(c, n)
+    ma = pd.Series(c).rolling(ma_n).mean().values
+    a = atr(df, 14)
+    ents = []
+    for i in range(max(n, ma_n) + 1, len(c)):
+        if np.isnan(r[i]) or np.isnan(ma[i]) or np.isnan(a[i]):
+            continue
+        if r[i - 1] < lo <= r[i]:
+            ents.append({"i": i, "d": 1, "tp": ma[i], "sl": c[i] - sl_atr * a[i], "max_bars": max_bars})
+        elif r[i - 1] > hi >= r[i]:
+            ents.append({"i": i, "d": -1, "tp": ma[i], "sl": c[i] + sl_atr * a[i], "max_bars": max_bars})
+    return ents
+
+
+def donchian_trend(df, n=20, sl_atr=2.0, tp_atr=6.0, max_bars=120):
+    """Trend-following: breakout canale Donchian, TP/SL in multipli di ATR."""
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    hh = pd.Series(h).rolling(n).max().shift(1).values
+    ll = pd.Series(l).rolling(n).min().shift(1).values
+    a = atr(df, 14)
+    ents = []
+    for i in range(n + 14, len(c)):
+        if np.isnan(hh[i]) or np.isnan(a[i]):
+            continue
+        if c[i] > hh[i]:
+            ents.append({"i": i, "d": 1, "tp": c[i] + tp_atr * a[i], "sl": c[i] - sl_atr * a[i], "max_bars": max_bars})
+        elif c[i] < ll[i]:
+            ents.append({"i": i, "d": -1, "tp": c[i] - tp_atr * a[i], "sl": c[i] + sl_atr * a[i], "max_bars": max_bars})
+    return ents
+
+
+STRATS = {
+    "BOLL_fade k2 m24":   (bollinger_fade, dict(n=20, k=2.0, sl_atr=2.0, max_bars=24)),
+    "BOLL_fade k2.5 m24": (bollinger_fade, dict(n=20, k=2.5, sl_atr=2.0, max_bars=24)),
+    "RSI_revert 30/70":   (rsi_revert, dict(n=14, lo=30, hi=70, sl_atr=2.0, max_bars=24)),
+    "RSI_revert 25/75":   (rsi_revert, dict(n=14, lo=25, hi=75, sl_atr=2.0, max_bars=24)),
+    "DONCH_trend n20":    (donchian_trend, dict(n=20, sl_atr=2.0, tp_atr=6.0, max_bars=120)),
+    "DONCH_trend n50":    (donchian_trend, dict(n=50, sl_atr=2.0, tp_atr=8.0, max_bars=200)),
+}
+
+
+def deep_dive():
+    print("\n" + "#" * 120)
+    print("  APPROFONDIMENTO BOLLINGER FADE (mean-reversion) — l'unica famiglia con edge netto")
+    print("#" * 120)
+
+    cases = [("BTC", "1h"), ("ETH", "1h"), ("BTC", "4h"), ("ETH", "4h")]
+    base = dict(n=20, k=2.5, sl_atr=2.0, max_bars=24)
+
+    # --- per anno (config base k2.5/n20) ---
+    print(f"\n  [1] PnL NETTO per anno — n=20 k=2.5 sl=2ATR | fee {FEE_RT*100:.2f}% RT")
+    all_years = sorted({y for a, tf in cases for y in simulate(bollinger_fade(get_df(a, tf), **base), get_df(a, tf))["yearly"]})
+    print(f"  {'Asset/TF':<10s}" + "".join(f"{y:>8d}" for y in all_years) + f"{'TOT%':>9s}{'DD%':>6s}")
+    for a, tf in cases:
+        df = get_df(a, tf)
+        r = simulate(bollinger_fade(df, **base), df)
+        row = "".join(f"{r['yearly'].get(y, 0):>+8.0f}" for y in all_years)
+        print(f"  {a+' '+tf:<10s}{row}{r['ret']:>+9.0f}{r['dd']:>6.0f}")
+
+    # --- sensibilita' fee ---
+    print(f"\n  [2] SENSIBILITA' FEE — Ret% FULL / OOS (n=20 k=2.5)")
+    fees = [0.0, 0.0005, 0.001, 0.002]
+    print(f"  {'Asset/TF':<10s}" + "".join(f"{f'{f*100:.2f}%RT':>22s}" for f in fees))
+    print(f"  {'':<10s}" + "".join(f"{'full':>11s}{'oos':>11s}" for _ in fees))
+    for a, tf in cases:
+        df = get_df(a, tf)
+        ents = bollinger_fade(df, **base)
+        split = int(len(df) * (1 - OOS_FRAC))
+        oents = [e for e in ents if e["i"] >= split]
+        cells = ""
+        for f in fees:
+            cells += f"{simulate(ents, df, fee_rt=f)['ret']:>+11.0f}{simulate(oents, df, fee_rt=f)['ret']:>+11.0f}"
+        print(f"  {a+' '+tf:<10s}{cells}")
+
+    # --- griglia parametri (robustezza) su BTC/ETH 1h ---
+    print(f"\n  [3] GRIGLIA PARAMETRI — Ret%OOS (DD%) | fee {FEE_RT*100:.2f}% RT, deve essere stabile")
+    for a in ["BTC", "ETH"]:
+        df = get_df(a, "1h")
+        split = int(len(df) * (1 - OOS_FRAC))
+        print(f"\n  {a} 1h     " + "".join(f"{f'k={k}':>16s}" for k in [2.0, 2.5, 3.0]))
+        for n in [14, 20, 30, 50]:
+            cells = ""
+            for k in [2.0, 2.5, 3.0]:
+                ents = [e for e in bollinger_fade(df, n=n, k=k, sl_atr=2.0, max_bars=24) if e["i"] >= split]
+                r = simulate(ents, df)
+                cell = f"{r['ret']:+.0f}({r['dd']:.0f})"
+                cells += f"{cell:>16s}"
+            print(f"   n={n:<4d}{cells}")
+
+
+def main():
+    print("=" * 120)
+    print(f"  RICERCA STRATEGIE — NETTO dopo fee {FEE_RT*100:.2f}% RT | leva {LEV:.0f}x | pos {POS*100:.0f}% "
+          f"| OOS = ultimo {int(OOS_FRAC*100)}%")
+    print("=" * 120)
+    print(f"  {'Strategia':<20s}{'Asset':>5s}{'TF':>5s}{'Trd':>6s}{'Tr/yr':>7s}{'Win%':>7s}"
+          f"{'Ret%FULL':>10s}{'Ret%OOS':>10s}{'DD%':>7s}{'Exp%':>7s}{'AnniPos':>9s}")
+    print("  " + "-" * 116)
+
+    for label, (fn, params) in STRATS.items():
+        for asset in ["BTC", "ETH"]:
+            for tf in ["1h", "4h"]:
+                df = get_df(asset, tf)
+                ents = fn(df, **params)
+                full = simulate(ents, df)
+                split = int(len(df) * (1 - OOS_FRAC))
+                oos = simulate([e for e in ents if e["i"] >= split], df)
+                yrs = full["yearly"]
+                pos_yrs = sum(1 for v in yrs.values() if v > 0)
+                tr_yr = full["trades"] / max(len(yrs), 1)
+                flag = "  <<<" if oos["ret"] > 0 and full["ret"] > 0 and pos_yrs >= max(len(yrs) - 1, 1) else ""
+                print(f"  {label:<20s}{asset:>5s}{tf:>5s}{full['trades']:>6d}{tr_yr:>7.0f}{full['win']:>7.1f}"
+                      f"{full['ret']:>+10.1f}{oos['ret']:>+10.1f}{full['dd']:>7.1f}{full['exposure']:>7.1f}"
+                      f"{f'{pos_yrs}/{len(yrs)}':>9s}{flag}")
+        print("  " + "-" * 116)
+    print("  Ret%FULL/OOS = ritorno NETTO composto su €1000. AnniPos = anni con PnL netto>0.")
+    print("  <<< = positivo full+OOS e robusto (quasi tutti gli anni positivi).")
+    deep_dive()
+
+
+if __name__ == "__main__":
+    main()

scripts/analysis/strategy_research_v2.py

@@ -0,0 +1,306 @@
+"""Ricerca v2 — nuove strategie oltre MR01, stessa metodologia fee-aware OOS.
+
+Lezioni ereditate (vedi strategy_research.py / oos_validation.py):
+  - mean-reversion ha edge, continuation/trend NO (i breakout rientrano)
+  - fee = vincolo di prim'ordine -> default Deribit 0.10% RT, poche operazioni meglio
+  - ingresso ESEGUIBILE a close[i] (mai look-ahead con direzione da barra i)
+  - ogni numero NETTO dopo fee+leva, su finestra held-out (OOS=ultimo 30%) + per anno
+
+Nuovi candidati (tutti fade/mean-reversion con ingresso onesto):
+  MR02 donchian_fade  - fade rottura canale Donchian (opposto del trend che muore)
+  MR03 keltner_fade   - fade canale Keltner (ATR), TP alla EMA media
+  MR04 zscore_revert  - fade deviazione z-score estrema, TP alla media
+  MR05 boll_fade_adx  - Bollinger fade con filtro regime ADX (solo mercato laterale)
+
+Engine identico a strategy_research.simulate (ingresso close[i], exit TP/SL intrabar
+high/low o time-limit, non-overlap, capitale composto).
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+# riusa engine, dati e indicatori gia' validati
+from scripts.analysis.strategy_research import (
+    FEE_RT, LEV, POS, OOS_FRAC, get_df, atr, rsi, simulate,
+)
+
+
+# --------------------------- indicatori extra ---------------------------
+def ema(x: np.ndarray, n: int) -> np.ndarray:
+    return pd.Series(x).ewm(span=n, adjust=False).mean().values
+
+
+def adx(df: pd.DataFrame, n: int = 14) -> np.ndarray:
+    """Average Directional Index: misura la forza del trend (alto=trend, basso=range)."""
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    up = h - np.roll(h, 1)
+    dn = np.roll(l, 1) - l
+    up[0] = dn[0] = 0.0
+    plus_dm = np.where((up > dn) & (up > 0), up, 0.0)
+    minus_dm = np.where((dn > up) & (dn > 0), dn, 0.0)
+    pc = np.roll(c, 1); pc[0] = c[0]
+    tr = np.maximum(h - l, np.maximum(np.abs(h - pc), np.abs(l - pc)))
+    atr_n = pd.Series(tr).ewm(alpha=1/n, adjust=False).mean().values
+    pdi = 100 * pd.Series(plus_dm).ewm(alpha=1/n, adjust=False).mean().values / np.where(atr_n == 0, np.nan, atr_n)
+    mdi = 100 * pd.Series(minus_dm).ewm(alpha=1/n, adjust=False).mean().values / np.where(atr_n == 0, np.nan, atr_n)
+    dx = 100 * np.abs(pdi - mdi) / np.where((pdi + mdi) == 0, np.nan, pdi + mdi)
+    return pd.Series(dx).ewm(alpha=1/n, adjust=False).mean().values
+
+
+# --------------------------- strategie nuove ---------------------------
+def donchian_fade(df, n=20, sl_atr=2.0, max_bars=24):
+    """MR02 — fade rottura canale Donchian: rompe sopra max-N => short verso il mid.
+
+    Coerente con 'i breakout rientrano': l'opposto di donchian_trend (che fallisce).
+    Ingresso a close[i] sulla barra che chiude oltre il canale precedente.
+    TP al centro del canale, SL = sl_atr*ATR oltre l'estremo.
+    """
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    hh = pd.Series(h).rolling(n).max().shift(1).values
+    ll = pd.Series(l).rolling(n).min().shift(1).values
+    a = atr(df, 14)
+    ents = []
+    for i in range(n + 14, len(c)):
+        if np.isnan(hh[i]) or np.isnan(a[i]):
+            continue
+        mid = (hh[i] + ll[i]) / 2.0
+        if c[i] > hh[i] and c[i - 1] <= hh[i - 1]:          # rottura rialzista => fade short
+            ents.append({"i": i, "d": -1, "tp": mid, "sl": c[i] + sl_atr * a[i], "max_bars": max_bars})
+        elif c[i] < ll[i] and c[i - 1] >= ll[i - 1]:        # rottura ribassista => fade long
+            ents.append({"i": i, "d": 1, "tp": mid, "sl": c[i] - sl_atr * a[i], "max_bars": max_bars})
+    return ents
+
+
+def keltner_fade(df, n=20, k=2.0, sl_atr=2.0, max_bars=24):
+    """MR03 — fade canale Keltner (EMA +/- k*ATR), TP alla EMA media.
+
+    Come Bollinger ma banda basata su ATR (volatilita' di range) invece che std:
+    reagisce diversamente ai gap. Ingresso quando close esce dalla banda.
+    """
+    c = df["close"].values
+    e = ema(c, n)
+    a = atr(df, n)
+    up, lo = e + k * a, e - k * a
+    ents = []
+    for i in range(n + 1, len(c)):
+        if np.isnan(up[i]) or np.isnan(a[i]):
+            continue
+        if c[i] < lo[i] and c[i - 1] >= lo[i - 1]:
+            ents.append({"i": i, "d": 1, "tp": e[i], "sl": c[i] - sl_atr * a[i], "max_bars": max_bars})
+        elif c[i] > up[i] and c[i - 1] <= up[i - 1]:
+            ents.append({"i": i, "d": -1, "tp": e[i], "sl": c[i] + sl_atr * a[i], "max_bars": max_bars})
+    return ents
+
+
+def zscore_revert(df, n=50, z=2.0, sl_atr=2.5, max_bars=24):
+    """MR04 — fade deviazione z-score estrema dalla media, TP alla media.
+
+    z = (close-ma)/std. Entra quando |z| supera la soglia (close fuori); chiude
+    quando torna alla media. Banda di Bollinger riparametrizzata in z (equivalente
+    a k=z) ma con SL piu' largo e finestra lunga: poche operazioni, alta selettivita'.
+    """
+    c = df["close"].values
+    ma = pd.Series(c).rolling(n).mean().values
+    sd = pd.Series(c).rolling(n).std().values
+    a = atr(df, 14)
+    ents = []
+    for i in range(n + 14, len(c)):
+        if np.isnan(ma[i]) or sd[i] == 0 or np.isnan(a[i]):
+            continue
+        zi = (c[i] - ma[i]) / sd[i]
+        zp = (c[i - 1] - ma[i - 1]) / sd[i - 1] if sd[i - 1] else 0.0
+        if zi <= -z and zp > -z:
+            ents.append({"i": i, "d": 1, "tp": ma[i], "sl": c[i] - sl_atr * a[i], "max_bars": max_bars})
+        elif zi >= z and zp < z:
+            ents.append({"i": i, "d": -1, "tp": ma[i], "sl": c[i] + sl_atr * a[i], "max_bars": max_bars})
+    return ents
+
+
+def boll_fade_adx(df, n=50, k=2.5, sl_atr=2.0, max_bars=24, adx_max=25.0):
+    """MR05 — Bollinger fade SOLO in regime laterale (ADX < adx_max).
+
+    Il fade soffre quando c'e' trend forte (il prezzo continua oltre la banda).
+    Filtro ADX: opera solo quando la forza del trend e' bassa -> meno trade, edge piu' pulito.
+    """
+    c = df["close"].values
+    ma = pd.Series(c).rolling(n).mean().values
+    sd = pd.Series(c).rolling(n).std().values
+    a = atr(df, 14)
+    ax = adx(df, 14)
+    up, lo = ma + k * sd, ma - k * sd
+    ents = []
+    for i in range(n + 14, len(c)):
+        if np.isnan(up[i]) or np.isnan(a[i]) or np.isnan(ax[i]):
+            continue
+        if ax[i] >= adx_max:                 # trend forte: niente fade
+            continue
+        if c[i] < lo[i] and c[i - 1] >= lo[i - 1]:
+            ents.append({"i": i, "d": 1, "tp": ma[i], "sl": c[i] - sl_atr * a[i], "max_bars": max_bars})
+        elif c[i] > up[i] and c[i - 1] <= up[i - 1]:
+            ents.append({"i": i, "d": -1, "tp": ma[i], "sl": c[i] + sl_atr * a[i], "max_bars": max_bars})
+    return ents
+
+
+def rsi2_fade(df, rsi_n=2, lo=10, hi=90, ma_n=200, tp_atr=2.0, sl_atr=3.0, max_bars=24):
+    """MR06 — Connors RSI(2) pullback in direzione del trend, TP/SL in ATR.
+
+    Meccanismo distinto da MR01/MR03: non usa bande di prezzo ma l'oscillatore
+    RSI(2), che satura su micro-estremi. Filtro di trend con SMA lunga:
+      - close SOPRA la SMA (uptrend) + RSI(2) < lo (dip) -> long, target rimbalzo
+      - close SOTTO la SMA (downtrend) + RSI(2) > hi (pop) -> short
+    TP = tp_atr*ATR a favore, SL = sl_atr*ATR contro. Compra il ritracciamento
+    nel trend, non il contro-trend.
+    """
+    c = df["close"].values
+    r = rsi(c, rsi_n)
+    ma = pd.Series(c).rolling(ma_n).mean().values
+    a = atr(df, 14)
+    ents = []
+    for i in range(ma_n + 14, len(c)):
+        if np.isnan(r[i]) or np.isnan(ma[i]) or np.isnan(a[i]):
+            continue
+        if r[i] < lo and c[i] > ma[i]:            # dip in uptrend -> long
+            ents.append({"i": i, "d": 1, "tp": c[i] + tp_atr * a[i], "sl": c[i] - sl_atr * a[i], "max_bars": max_bars})
+        elif r[i] > hi and c[i] < ma[i]:          # pop in downtrend -> short
+            ents.append({"i": i, "d": -1, "tp": c[i] - tp_atr * a[i], "sl": c[i] + sl_atr * a[i], "max_bars": max_bars})
+    return ents
+
+
+def return_reversal(df, n=50, k=3.5, tp_atr=2.0, sl_atr=1.5, max_bars=24):
+    """MR07 — fade movimento di barra estremo (return reversal).
+
+    Misura il rendimento dell'ultima barra in unita' di deviazione standard rolling
+    dei rendimenti. Se |ret| > k*sigma, fada nella direzione opposta; TP/SL in ATR.
+    Meccanismo distinto: usa la volatilita' dei RENDIMENTI, non i livelli di prezzo.
+    Config robusta (k=3.5, tp=2ATR, sl=1.5ATR): positivo full+OOS BTC e ETH 1h,
+    DD piu' contenuto (BTC 25% / ETH 46%).
+    """
+    c = df["close"].values
+    ret = np.zeros_like(c)
+    ret[1:] = np.diff(c) / c[:-1]
+    sig = pd.Series(ret).rolling(n).std().values
+    a = atr(df, 14)
+    ents = []
+    for i in range(n + 14, len(c)):
+        if np.isnan(sig[i]) or sig[i] == 0 or np.isnan(a[i]):
+            continue
+        z = ret[i] / sig[i]
+        if z <= -k:                                # crollo di barra -> fade long
+            ents.append({"i": i, "d": 1, "tp": c[i] + tp_atr * a[i], "sl": c[i] - sl_atr * a[i], "max_bars": max_bars})
+        elif z >= k:                               # spike di barra -> fade short
+            ents.append({"i": i, "d": -1, "tp": c[i] - tp_atr * a[i], "sl": c[i] + sl_atr * a[i], "max_bars": max_bars})
+    return ents
+
+
+CANDIDATES = {
+    "MR02 donch_fade n20":  (donchian_fade, dict(n=20, sl_atr=2.0, max_bars=24)),
+    "MR02 donch_fade n50":  (donchian_fade, dict(n=50, sl_atr=2.0, max_bars=24)),
+    "MR03 kelt_fade k2":    (keltner_fade, dict(n=20, k=2.0, sl_atr=2.0, max_bars=24)),
+    "MR03 kelt_fade k2.5":  (keltner_fade, dict(n=20, k=2.5, sl_atr=2.0, max_bars=24)),
+    "MR04 zscore z2 n50":   (zscore_revert, dict(n=50, z=2.0, sl_atr=2.5, max_bars=24)),
+    "MR04 zscore z2.5 n50": (zscore_revert, dict(n=50, z=2.5, sl_atr=2.5, max_bars=24)),
+    "MR05 boll_adx n50":    (boll_fade_adx, dict(n=50, k=2.5, sl_atr=2.0, max_bars=24, adx_max=25)),
+    "MR05 boll_adx n20":    (boll_fade_adx, dict(n=20, k=2.5, sl_atr=2.0, max_bars=24, adx_max=25)),
+    "MR06 rsi2 10/90":      (rsi2_fade, dict(rsi_n=2, lo=10, hi=90, ma_n=200, tp_atr=2.0, sl_atr=3.0, max_bars=24)),
+    "MR06 rsi2 5/95":       (rsi2_fade, dict(rsi_n=2, lo=5, hi=95, ma_n=200, tp_atr=2.0, sl_atr=3.0, max_bars=24)),
+    "MR07 retrev k3.5":     (return_reversal, dict(n=50, k=3.5, tp_atr=2.0, sl_atr=1.5, max_bars=24)),
+    "MR07 retrev k3.0":     (return_reversal, dict(n=50, k=3.0, tp_atr=2.0, sl_atr=1.5, max_bars=24)),
+}
+
+
+def table():
+    print("=" * 122)
+    print(f"  RICERCA v2 — NETTO dopo fee {FEE_RT*100:.2f}% RT | leva {LEV:.0f}x | pos {POS*100:.0f}% "
+          f"| OOS = ultimo {int(OOS_FRAC*100)}%")
+    print("=" * 122)
+    print(f"  {'Strategia':<22s}{'Asset':>5s}{'TF':>5s}{'Trd':>6s}{'Tr/yr':>7s}{'Win%':>7s}"
+          f"{'Ret%FULL':>10s}{'Ret%OOS':>10s}{'DD%':>7s}{'Exp%':>7s}{'AnniPos':>9s}")
+    print("  " + "-" * 118)
+    for label, (fn, params) in CANDIDATES.items():
+        for asset in ["BTC", "ETH"]:
+            for tf in ["1h", "4h"]:
+                df = get_df(asset, tf)
+                ents = fn(df, **params)
+                full = simulate(ents, df)
+                split = int(len(df) * (1 - OOS_FRAC))
+                oos = simulate([e for e in ents if e["i"] >= split], df)
+                yrs = full["yearly"]
+                pos_yrs = sum(1 for v in yrs.values() if v > 0)
+                tr_yr = full["trades"] / max(len(yrs), 1)
+                robust = oos["ret"] > 0 and full["ret"] > 0 and pos_yrs >= max(len(yrs) - 1, 1)
+                flag = "  <<<" if robust else ""
+                print(f"  {label:<22s}{asset:>5s}{tf:>5s}{full['trades']:>6d}{tr_yr:>7.0f}{full['win']:>7.1f}"
+                      f"{full['ret']:>+10.1f}{oos['ret']:>+10.1f}{full['dd']:>7.1f}{full['exposure']:>7.1f}"
+                      f"{f'{pos_yrs}/{len(yrs)}':>9s}{flag}")
+        print("  " + "-" * 118)
+    print("  <<< = positivo full+OOS e robusto (quasi tutti gli anni positivi).")
+
+
+def deep_dive():
+    """Robustezza dei 3 candidati promossi: fee sweep + griglia parametri OOS."""
+    split_of = lambda df: int(len(df) * (1 - OOS_FRAC))
+    fees = [0.0, 0.0005, 0.001, 0.002]
+
+    print("\n" + "#" * 122)
+    print("  APPROFONDIMENTO MR02 / MR03 / MR05 — robustezza fee + griglia (deve restare positivo)")
+    print("#" * 122)
+
+    # --- MR02 Donchian Fade ---
+    print(f"\n  [MR02 donchian_fade] SENSIBILITA' FEE — Ret% FULL/OOS (n=20)")
+    print(f"  {'Asset/TF':<10s}" + "".join(f"{f'{f*100:.2f}%RT':>22s}" for f in fees))
+    print(f"  {'':<10s}" + "".join(f"{'full':>11s}{'oos':>11s}" for _ in fees))
+    for a, tf in [("BTC", "1h"), ("ETH", "1h"), ("BTC", "4h"), ("ETH", "4h")]:
+        df = get_df(a, tf); sp = split_of(df)
+        ents = donchian_fade(df, n=20, sl_atr=2.0, max_bars=24)
+        oents = [e for e in ents if e["i"] >= sp]
+        cells = "".join(f"{simulate(ents, df, fee_rt=f)['ret']:>+11.0f}{simulate(oents, df, fee_rt=f)['ret']:>+11.0f}" for f in fees)
+        print(f"  {a+' '+tf:<10s}{cells}")
+    print(f"\n  [MR02] GRIGLIA n x sl_atr — Ret%OOS(DD%) | fee {FEE_RT*100:.2f}% RT")
+    for a in ["BTC", "ETH"]:
+        df = get_df(a, "1h"); sp = split_of(df)
+        print(f"\n  {a} 1h     " + "".join(f"{f'sl={s}':>16s}" for s in [1.5, 2.0, 3.0]))
+        for n in [10, 20, 30, 50]:
+            cells = ""
+            for s in [1.5, 2.0, 3.0]:
+                r = simulate([e for e in donchian_fade(df, n=n, sl_atr=s, max_bars=24) if e["i"] >= sp], df)
+                cell = "%+.0f(%.0f)" % (r["ret"], r["dd"])
+                cells += f"{cell:>16s}"
+            print(f"   n={n:<4d}{cells}")
+
+    # --- MR03 Keltner Fade ---
+    print(f"\n  [MR03 keltner_fade] GRIGLIA n x k — Ret%OOS(DD%) | fee {FEE_RT*100:.2f}% RT")
+    for a in ["BTC", "ETH"]:
+        df = get_df(a, "1h"); sp = split_of(df)
+        print(f"\n  {a} 1h     " + "".join(f"{f'k={k}':>16s}" for k in [1.5, 2.0, 2.5]))
+        for n in [14, 20, 30, 50]:
+            cells = ""
+            for k in [1.5, 2.0, 2.5]:
+                r = simulate([e for e in keltner_fade(df, n=n, k=k, sl_atr=2.0, max_bars=24) if e["i"] >= sp], df)
+                cell = "%+.0f(%.0f)" % (r["ret"], r["dd"])
+                cells += f"{cell:>16s}"
+            print(f"   n={n:<4d}{cells}")
+
+    # --- MR05 Bollinger Fade + ADX ---
+    print(f"\n  [MR05 boll_fade_adx] GRIGLIA n x adx_max — Ret%OOS(DD%) | fee {FEE_RT*100:.2f}% RT")
+    for a in ["BTC", "ETH"]:
+        df = get_df(a, "1h"); sp = split_of(df)
+        print(f"\n  {a} 1h     " + "".join(f"{f'adx<{x}':>16s}" for x in [20, 25, 30]))
+        for n in [20, 30, 50]:
+            cells = ""
+            for x in [20, 25, 30]:
+                r = simulate([e for e in boll_fade_adx(df, n=n, k=2.5, sl_atr=2.0, max_bars=24, adx_max=x) if e["i"] >= sp], df)
+                cell = "%+.0f(%.0f)" % (r["ret"], r["dd"])
+                cells += f"{cell:>16s}"
+            print(f"   n={n:<4d}{cells}")
+
+
+if __name__ == "__main__":
+    table()
+    deep_dive()

scripts/analysis/tsmom_research.py

@@ -0,0 +1,94 @@
+"""Verifica indipendente + ricerca TSM01 — Time-Series Momentum multi-orizzonte.
+
+Long-only, multi-crypto, bassa frequenza. Per ogni asset il segnale è il CONSENSO
+dei segni del momentum su più orizzonti lunghi (3/6/12 mesi); si tengono equal-weight
+gli asset con consenso pieno positivo. Overlay risk-off: cash se BTC < SMA100.
+
+Distinta da ROT02 (cross-sectional ranking): qui conta la PERSISTENZA assoluta lenta
+di ogni asset, non la classifica relativa. Correlazione con ROT02 ~0.62 -> fattore
+parzialmente indipendente, utile come diversificatore (NON come motore di ritorno:
+rende meno di ROT02 a parita' di OOS). DD basso.
+
+Anti-overfit: edge su ALTOPIANO (36/36 config orizzonti x thr x regime_n restano OOS+),
+walk-forward stabile (4 anni up, 2 piatti per risk-off, mai un anno negativo), regge
+fee 0.40% RT. Gran parte del DD basso viene dall'overlay risk-off SMA100 (condiviso),
+la struttura multi-orizzonte aggiunge ~+38pp OOS e alza lo Sharpe 0.58->1.07.
+Default gross=0.30 (era 0.45): stesso Sharpe ma DD 22%->15% (scelta robusta, non la piu' redditizia).
+
+Engine onesto: pesi a close[i] da soli rendimenti passati, realizzo i->i+1, fee
+one-way fee_rt/2 sul turnover. NETTO, leva implicita gross. OOS = ultimo 30%.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.honest_lab import available_assets, FEE_RT
+from scripts.analysis.honest_rotation import build_panel
+
+GROSS, OOS_FRAC = 0.30, 0.30   # gross 0.30 (anti-overfit): stesso Sharpe di 0.45, DD piu' basso
+
+
+def tsmom_sim(horizons=(63, 126, 252), thr=1.0, regime_n=100, gross=GROSS,
+              fee_rt=FEE_RT, oos_frac=0.0, cheat=False):
+    """horizons in giorni. thr=1.0 -> consenso pieno (tutti i segni positivi)."""
+    panel = build_panel(available_assets(), "1d")
+    cols = list(panel.columns); P = panel.values; T, N = P.shape
+    rets = np.zeros_like(P); rets[1:] = P[1:] / P[:-1] - 1
+    years = panel.index.year.values
+    btc = P[:, cols.index("BTC")]
+    bma = pd.Series(btc).rolling(regime_n).mean().values
+    start = max(max(horizons) + 1, regime_n + 1, int(T * (1 - oos_frac)) if oos_frac else 0)
+    cap = 1000.0; w = np.zeros(N); eq = [cap]; yearly = {}
+    eq_ts: list = []; eq_v: list = []
+    for i in range(start, T - 1):
+        risk_on = btc[i] > bma[i] if not np.isnan(bma[i]) else False
+        wi = i + 1 if cheat else i               # cheat: usa il futuro (test no-look-ahead)
+        score = np.zeros(N)
+        for h in horizons:
+            score += np.sign(P[wi] / P[wi - h] - 1)
+        score /= len(horizons)
+        chosen = [j for j in range(N) if score[j] >= thr] if risk_on else []
+        nw = np.zeros(N)
+        for j in chosen:
+            nw[j] = gross / len(chosen)
+        cap -= cap * np.abs(nw - w).sum() * (fee_rt / 2); w = nw
+        cap = max(cap * (1 + float(np.dot(w, rets[i + 1]))), 10.0)
+        eq.append(cap)
+        eq_ts.append(panel.index[i + 1]); eq_v.append(cap)
+        y = int(years[i]); yearly[y] = yearly.get(y, 0.0) + float(np.dot(w, rets[i + 1])) * 100
+    eq = np.array(eq); peak = np.maximum.accumulate(eq)
+    dd = float(np.max((peak - eq) / peak) * 100)
+    yrs = (panel.index[-1] - panel.index[start]).days / 365.25 or 1
+    rets_d = np.diff(eq) / eq[:-1]
+    sharpe = float(np.mean(rets_d) / np.std(rets_d) * np.sqrt(365)) if np.std(rets_d) > 0 else 0.0
+    return dict(ret=(cap / 1000 - 1) * 100, cagr=((cap / 1000) ** (1 / yrs) - 1) * 100,
+                dd=dd, sharpe=sharpe, yearly=yearly, eq_ts=eq_ts, eq_v=eq_v,
+                pos_years=sum(1 for v in yearly.values() if v > 0), n_years=len(yearly))
+
+
+def main():
+    print("=" * 90)
+    print("  TSM01 — TSMOM multi-orizzonte (3/6/12m consenso pieno) + risk-off SMA100")
+    print("=" * 90)
+    # no-look-ahead: cheat deve esplodere
+    base = tsmom_sim()
+    ch = tsmom_sim(cheat=True)
+    print(f"  no-look-ahead: onesto FULL={base['ret']:+.0f}%  vs  cheat(futuro)={ch['ret']:+.0f}%  -> "
+          f"{'OK (il cheat esplode -> niente leak)' if ch['ret'] > base['ret'] * 2 else 'CONTROLLARE'}")
+    o = tsmom_sim(oos_frac=1 - OOS_FRAC)
+    hi = tsmom_sim(fee_rt=0.002)
+    print(f"\n  FULL {base['ret']:+.0f}%  CAGR {base['cagr']:.0f}%  DD {base['dd']:.0f}%  "
+          f"Sharpe {base['sharpe']:.2f}  anni+ {base['pos_years']}/{base['n_years']}")
+    print(f"  OOS  {o['ret']:+.0f}%  DD {o['dd']:.0f}%   |  fee 0.40% RT: FULL {hi['ret']:+.0f}%")
+    print("  Per-anno: " + "  ".join(f"{y}:{v:+.0f}%" for y, v in sorted(base["yearly"].items())))
+
+
+if __name__ == "__main__":
+    main()

scripts/analysis/validate_worker_mr01.py

@@ -0,0 +1,69 @@
+"""Re-validazione: il StrategyWorker REALE tradi MR01 con edge netto?
+
+Guida il worker vero (generate_signals + nuova logica exit TP/SL/max_bars) su
+finestre mobili di dati 1h storici, simulando il polling live. Conferma che
+sulla finestra OOS l'edge netto (dopo fee 0.10% RT) sopravvive alla meccanica
+del worker (exit su prezzo corrente, piu' conservativa del backtest high/low).
+"""
+from __future__ import annotations
+
+import contextlib
+import os
+import sys
+from pathlib import Path
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from src.data.downloader import load_data
+from src.live.strategy_loader import load_strategy
+from src.live.strategy_worker import StrategyWorker
+
+OOS_FRAC = 0.30
+WIN = 250  # barre per finestra di poll (warmup bb_window=50 + ATR)
+
+
+def replay(asset: str, params: dict):
+    df = load_data(asset, "1h").reset_index(drop=True)
+    n = len(df)
+    split = int(n * (1 - OOS_FRAC))
+    strat = load_strategy("MR01_bollinger_fade")
+    w = StrategyWorker(strat, asset, "1h", capital=1000.0, position_size=0.15,
+                       leverage=3.0, hold_bars=3, params=params,
+                       data_dir=Path(f"/tmp/replay_{asset}"))
+    w._notify = lambda *a, **k: None
+    # stato pulito
+    for attr, val in dict(capital=1000.0, in_position=False, direction=0, entry_price=0,
+                          bars_held=0, total_trades=0, total_wins=0, last_bar_ts=0,
+                          tp=0.0, sl=0.0, max_bars=0).items():
+        setattr(w, attr, val)
+
+    start = max(split, WIN)
+    with contextlib.redirect_stdout(open(os.devnull, "w")):
+        for j in range(start, n):
+            w.tick(df.iloc[j - WIN + 1 : j + 1])
+
+    ret = (w.capital / 1000 - 1) * 100
+    acc = w.total_wins / w.total_trades * 100 if w.total_trades else 0.0
+    import pandas as pd
+    period = (f"{pd.to_datetime(df['timestamp'].iloc[start], unit='ms', utc=True).date()}"
+              f"->{pd.to_datetime(df['timestamp'].iloc[-1], unit='ms', utc=True).date()}")
+    return w.total_trades, acc, ret, w.capital, period
+
+
+def main():
+    print("=" * 90)
+    print("  RE-VALIDAZIONE WORKER REALE su MR01 (OOS, fee 0.10% RT, leva 3x) — finestra poll 250b")
+    print("=" * 90)
+    params = dict(bb_window=50, k=2.5, sl_atr=2.0, max_bars=24)
+    print(f"  {'Asset':>6s}{'Periodo OOS':>26s}{'Trade':>7s}{'Win%':>7s}{'Ret%':>9s}{'Cap€':>9s}")
+    print("  " + "-" * 80)
+    for asset in ["BTC", "ETH"]:
+        t, acc, ret, cap, period = replay(asset, params)
+        print(f"  {asset:>6s}{period:>26s}{t:>7d}{acc:>7.1f}{ret:>+9.1f}{cap:>9.0f}")
+    print("  " + "-" * 80)
+    print("  Atteso: Ret% positivo (l'edge mean-reversion sopravvive alla meccanica del worker).")
+
+
+if __name__ == "__main__":
+    main()

scripts/strategies/DIP01_dip_reversion.py

@@ -0,0 +1,152 @@
+"""DIP01 — Dip-Buy Z-Score Reversion (long-only).
+
+Variante robusta e ONESTA della famiglia mean-reversion: compra SOLO i dip
+(close a z<=-z_in deviazioni sotto la media mobile) e prende profitto al rientro
+verso la media. Niente short: nel campione 2018-2026 shortare cripto perde OOS
+sistematicamente (vedi scripts/analysis/honest_final.py).
+
+Logica:
+    1. z-score = (close - SMA(n)) / STD(n)
+    2. ENTRY long quando z attraversa al ribasso -z_in (capitolazione)
+    3. EXIT: take-profit alla media mobile, stop-loss a sl_atr*ATR sotto l'entry,
+       o time-limit max_bars
+    4. ingresso a close[i] (eseguibile dal vivo, nessun look-ahead)
+
+Validazione (netto, fee 0.10% RT Deribit, leva 3x, OOS = ultimo 30%):
+    BTC 1h: FULL +298% / OOS +59% / DD 23% / 7-9 anni positivi
+    ETH 1h: FULL +190% / OOS +224% / DD 54%
+    SOL 1h: FULL  +50% / OOS +13% / DD 25%
+    Regge lo sweep fee fino a 0.20% RT (BTC OOS +45% anche a 0.20%).
+    Robusto su BTC/ETH/SOL (asset major); sugli alt molto parabolici (DOGE/BNB)
+    non ha edge -> usare solo su BTC/ETH/SOL.
+
+Compatibile con StrategyWorker: ogni Signal porta tp/sl/max_bars in metadata.
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+
+from src.strategies.base import Strategy, Signal, BacktestResult, YearlyStats, TF_MINUTES
+from src.data.downloader import load_data
+
+
+def _atr(df: pd.DataFrame, n: int = 14) -> np.ndarray:
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    pc = np.roll(c, 1); pc[0] = c[0]
+    tr = np.maximum(h - l, np.maximum(np.abs(h - pc), np.abs(l - pc)))
+    return pd.Series(tr).rolling(n).mean().values
+
+
+class DipReversion(Strategy):
+    name = "DIP01_dip_reversion"
+    description = "Long-only dip-buy z-score reversion, TP alla media"
+    default_assets = ["BTC", "ETH", "SOL"]
+    default_timeframes = ["1h"]
+    fee_rt = 0.001
+    leverage = 3.0
+    position_size = 0.15
+    initial_capital = 1000.0
+
+    def generate_signals(self, df: pd.DataFrame, ts: pd.DatetimeIndex,
+                         **params) -> list[Signal]:
+        c = df["close"].values
+        n = params.get("n", 50)
+        z_in = params.get("z_in", 2.5)
+        sl_atr = params.get("sl_atr", 2.5)
+        max_bars = params.get("max_bars", 24)
+
+        ma = pd.Series(c).rolling(n).mean().values
+        sd = pd.Series(c).rolling(n).std().values
+        a = _atr(df, 14)
+        z = (c - ma) / np.where(sd == 0, np.nan, sd)
+
+        signals: list[Signal] = []
+        for i in range(n + 14, len(c)):
+            if np.isnan(z[i]) or np.isnan(a[i]):
+                continue
+            if z[i] <= -z_in and z[i - 1] > -z_in:
+                signals.append(Signal(
+                    idx=i, direction=1, entry_price=c[i],
+                    metadata={"tp": float(ma[i]), "sl": float(c[i] - sl_atr * a[i]),
+                              "max_bars": max_bars},
+                ))
+        return signals
+
+    def backtest(self, asset: str, tf: str = "1h", hold: int = 3,
+                 **params) -> BacktestResult | None:
+        df = load_data(asset, tf)
+        ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+        signals = self.generate_signals(df, ts, **params)
+        if not signals:
+            return None
+        h, l, c = df["high"].values, df["low"].values, df["close"].values
+        n = len(c)
+        fee = self.fee_rt * self.leverage
+        capital = peak = float(self.initial_capital)
+        max_dd = 0.0
+        total_bars = 0
+        last_exit = -1
+        yearly: dict[int, dict] = {}
+
+        for sig in signals:
+            i, d = sig.idx, sig.direction
+            if i <= last_exit or i + 1 >= n:
+                continue
+            entry = c[i]
+            tp, sl, mb = sig.metadata["tp"], sig.metadata["sl"], sig.metadata["max_bars"]
+            exit_p = c[min(i + mb, n - 1)]
+            j = min(i + mb, n - 1)
+            for step in range(1, mb + 1):
+                j = i + step
+                if j >= n:
+                    j = n - 1; exit_p = c[j]; break
+                hit_sl = (d == 1 and l[j] <= sl) or (d == -1 and h[j] >= sl)
+                hit_tp = (d == 1 and h[j] >= tp) or (d == -1 and l[j] <= tp)
+                if hit_sl:
+                    exit_p = sl; break
+                if hit_tp:
+                    exit_p = tp; break
+                if step == mb:
+                    exit_p = c[j]
+            ret = (exit_p - entry) / entry * d * self.leverage - fee
+            capital = max(capital + capital * self.position_size * ret, 10.0)
+            if capital > peak:
+                peak = capital
+            max_dd = max(max_dd, (peak - capital) / peak)
+            total_bars += (j - i)
+            last_exit = j
+            year = ts.iloc[i].year
+            yr = yearly.setdefault(year, {"w": 0, "t": 0, "pnl": 0.0})
+            yr["t"] += 1
+            if ret > 0:
+                yr["w"] += 1
+            yr["pnl"] += ret * self.initial_capital
+
+        all_t = sum(v["t"] for v in yearly.values())
+        all_w = sum(v["w"] for v in yearly.values())
+        if all_t == 0:
+            return None
+        yearly_stats = [YearlyStats(y, v["t"], v["w"], v["pnl"]) for y, v in sorted(yearly.items())]
+        return BacktestResult(
+            strategy_name=self.name, asset=asset, timeframe=tf, params=params,
+            trades=all_t, wins=all_w, pnl=sum(v["pnl"] for v in yearly.values()),
+            capital=capital, initial_capital=self.initial_capital,
+            max_dd=max_dd * 100, time_in_market_pct=total_bars / n * 100,
+            avg_trade_duration_h=total_bars / all_t * TF_MINUTES.get(tf, 60) / 60,
+            years_active=len(yearly), yearly=yearly_stats,
+        )
+
+
+if __name__ == "__main__":
+    strat = DipReversion()
+    print(f"{'=' * 100}")
+    print(f"  DIP01 DIP-BUY REVERSION — netto fee {strat.fee_rt*100:.2f}% RT, leva {strat.leverage:.0f}x")
+    print(f"{'=' * 100}")
+    for asset in ["BTC", "ETH", "SOL"]:
+        r = strat.backtest(asset, "1h", n=50, z_in=2.5, sl_atr=2.5, max_bars=24)
+        if r:
+            r.strategy_name = f"DIP01 {asset} 1h"
+            r.print_summary()

scripts/strategies/MR01_bollinger_fade.py

@@ -0,0 +1,172 @@
+"""MR01 — Bollinger Fade (mean-reversion).
+
+L'UNICA famiglia con edge netto reale dopo l'analisi out-of-sample fee-aware
+(vedi scripts/analysis/strategy_research.py). Contrario della tesi squeeze:
+i breakout RIENTRANO, quindi si fada l'estremo verso la media.
+
+Logica:
+    1. Bollinger Band (window n, k deviazioni) sul close
+    2. ENTRY: close esce sotto la banda inferiore -> long (o sopra la superiore -> short)
+    3. EXIT: take-profit alla media mobile (il rientro atteso),
+       stop-loss a sl_atr*ATR oltre l'estremo, oppure time-limit max_bars
+    4. ingresso a close[i] (eseguibile dal vivo, nessun look-ahead)
+
+Validazione (netto, fee 0.10% RT reale Deribit, leva 3x, OOS = ultimo 30%):
+    BTC 1h n=50 k=2.5: +201% OOS, DD 15%, ~tutti gli anni positivi
+    ETH 1h n=50 k=2.0: +1238% OOS, DD 23%
+    Robusto su TUTTA la griglia n in {14,20,30,50} x k in {2.0,2.5,3.0}
+    e su tutte le fee 0.00-0.20% RT (margine di sicurezza ampio).
+
+NOTA LIVE: usa TP alla media + SL ad ATR + max_bars. Lo StrategyWorker attuale
+esce solo a hold_bars/stop -2% fisso: per tradarla come validata il worker deve
+supportare gli exit TP/SL passati in metadata (vedi metadata di ogni Signal).
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+
+from src.strategies.base import Strategy, Signal, BacktestResult, YearlyStats, TF_MINUTES
+from src.data.downloader import load_data
+
+
+def _atr(df: pd.DataFrame, n: int = 14) -> np.ndarray:
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    pc = np.roll(c, 1); pc[0] = c[0]
+    tr = np.maximum(h - l, np.maximum(np.abs(h - pc), np.abs(l - pc)))
+    return pd.Series(tr).rolling(n).mean().values
+
+
+class BollingerFade(Strategy):
+    name = "MR01_bollinger_fade"
+    description = "Mean-reversion: fada la banda di Bollinger, TP alla media"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["1h"]
+    fee_rt = 0.001          # Deribit perp realistico (taker 0.05%/lato)
+    leverage = 3.0
+    position_size = 0.15
+    initial_capital = 1000.0
+
+    def generate_signals(self, df: pd.DataFrame, ts: pd.DatetimeIndex,
+                         **params) -> list[Signal]:
+        c = df["close"].values
+        n_len = len(c)
+        bb_w = params.get("bb_window", 50)
+        k = params.get("k", 2.5)
+        sl_atr = params.get("sl_atr", 2.0)
+        max_bars = params.get("max_bars", 24)
+        trend_max = params.get("trend_max")          # None = filtro disattivo
+        ema_long = params.get("ema_long", 200)
+
+        ma = pd.Series(c).rolling(bb_w).mean().values
+        sd = pd.Series(c).rolling(bb_w).std().values
+        a = _atr(df, 14)
+        up, lo = ma + k * sd, ma - k * sd
+        el = pd.Series(c).ewm(span=ema_long, adjust=False).mean().values if trend_max is not None else None
+
+        signals: list[Signal] = []
+        for i in range(bb_w + 14, n_len):
+            if np.isnan(up[i]) or np.isnan(a[i]):
+                continue
+            if el is not None and (a[i] == 0 or np.isnan(el[i]) or abs(c[i] - el[i]) / a[i] > trend_max):
+                continue
+            if c[i] < lo[i] and c[i - 1] >= lo[i - 1]:
+                d, sl = 1, c[i] - sl_atr * a[i]
+            elif c[i] > up[i] and c[i - 1] <= up[i - 1]:
+                d, sl = -1, c[i] + sl_atr * a[i]
+            else:
+                continue
+            signals.append(Signal(
+                idx=i, direction=d, entry_price=c[i],
+                metadata={"tp": float(ma[i]), "sl": float(sl), "max_bars": max_bars},
+            ))
+        return signals
+
+    def backtest(self, asset: str, tf: str = "1h", hold: int = 3,
+                 **params) -> BacktestResult | None:
+        """Backtest fedele: TP alla media / SL ad ATR / time-limit, fee+leva nette."""
+        df = load_data(asset, tf)
+        ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+        signals = self.generate_signals(df, ts, **params)
+        if not signals:
+            return None
+
+        h, l, c = df["high"].values, df["low"].values, df["close"].values
+        n = len(c)
+        fee = self.fee_rt * self.leverage
+        capital = peak = float(self.initial_capital)
+        max_dd = 0.0
+        total_bars = 0
+        last_exit = -1
+        yearly: dict[int, dict] = {}
+
+        for sig in signals:
+            i, d = sig.idx, sig.direction
+            if i <= last_exit or i + 1 >= n:
+                continue
+            entry = c[i]
+            tp, sl, mb = sig.metadata["tp"], sig.metadata["sl"], sig.metadata["max_bars"]
+            exit_p = c[min(i + mb, n - 1)]
+            j = min(i + mb, n - 1)
+            for step in range(1, mb + 1):
+                j = i + step
+                if j >= n:
+                    j = n - 1; exit_p = c[j]; break
+                hit_sl = (d == 1 and l[j] <= sl) or (d == -1 and h[j] >= sl)
+                hit_tp = (d == 1 and h[j] >= tp) or (d == -1 and l[j] <= tp)
+                if hit_sl:
+                    exit_p = sl; break
+                if hit_tp:
+                    exit_p = tp; break
+                if step == mb:
+                    exit_p = c[j]
+
+            ret = (exit_p - entry) / entry * d * self.leverage - fee
+            capital = max(capital + capital * self.position_size * ret, 10.0)
+            if capital > peak:
+                peak = capital
+            max_dd = max(max_dd, (peak - capital) / peak)
+            total_bars += (j - i)
+            last_exit = j
+
+            year = ts.iloc[i].year
+            yr = yearly.setdefault(year, {"w": 0, "t": 0, "pnl": 0.0})
+            yr["t"] += 1
+            if ret > 0:
+                yr["w"] += 1
+            yr["pnl"] += ret * self.initial_capital
+
+        all_t = sum(v["t"] for v in yearly.values())
+        all_w = sum(v["w"] for v in yearly.values())
+        if all_t == 0:
+            return None
+
+        yearly_stats = [YearlyStats(y, v["t"], v["w"], v["pnl"]) for y, v in sorted(yearly.items())]
+        return BacktestResult(
+            strategy_name=self.name, asset=asset, timeframe=tf, params=params,
+            trades=all_t, wins=all_w, pnl=sum(v["pnl"] for v in yearly.values()),
+            capital=capital, initial_capital=self.initial_capital,
+            max_dd=max_dd * 100, time_in_market_pct=total_bars / n * 100,
+            avg_trade_duration_h=total_bars / all_t * TF_MINUTES.get(tf, 60) / 60,
+            years_active=len(yearly), yearly=yearly_stats,
+        )
+
+
+if __name__ == "__main__":
+    strat = BollingerFade()
+    print(f"{'=' * 110}")
+    print(f"  MR01 BOLLINGER FADE — netto fee {strat.fee_rt*100:.2f}% RT, leva {strat.leverage:.0f}x")
+    print(f"{'=' * 110}")
+    results = []
+    for asset in ["BTC", "ETH"]:
+        for k in [2.0, 2.5]:
+            r = strat.backtest(asset, "1h", bb_window=50, k=k, sl_atr=2.0, max_bars=24)
+            if r:
+                r.strategy_name = f"MR01 {asset} 1h n50 k{k}"
+                results.append(r)
+    for r in results:
+        r.print_summary()
+    if results:
+        results[0].print_yearly()

scripts/strategies/MR02_donchian_fade.py

@@ -0,0 +1,82 @@
+"""MR02 — Donchian Fade (mean-reversion sugli estremi del canale).
+
+L'opposto esatto del trend-following Donchian (che PERDE netto: vedi
+scripts/analysis/strategy_research.py). Coerente con la lezione squeeze:
+i breakout RIENTRANO, quindi si fada la rottura del canale verso il centro.
+
+Logica:
+    1. Canale Donchian: massimo/minimo delle ultime n barre (escludendo la corrente)
+    2. ENTRY: close rompe SOPRA il massimo del canale -> SHORT (fade);
+       close rompe SOTTO il minimo -> LONG. Ingresso a close[i] (eseguibile).
+    3. EXIT: take-profit al centro del canale (il rientro atteso),
+       stop-loss a sl_atr*ATR oltre l'estremo, time-limit max_bars.
+
+Validazione (netto, fee 0.10% RT reale Deribit, leva 3x, OOS = ultimo 30%):
+    BTC 1h n=20: +879% FULL / +171% OOS, DD 30%, 8/9 anni positivi
+    ETH 1h n=20: enorme FULL / +8452% OOS, DD 42%
+    Robusto su TUTTA la griglia n in {10,20,30,50} x sl_atr in {1.5,2.0,3.0}
+    (BTC+ETH 1h sempre positivo OOS) e su tutte le fee 0.00-0.20% RT.
+    Ricerca completa: scripts/analysis/strategy_research_v2.py.
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+
+from src.strategies.base import Signal
+from src.strategies.fade_base import FadeStrategy, atr, trend_distance
+
+
+class DonchianFade(FadeStrategy):
+    name = "MR02_donchian_fade"
+    description = "Mean-reversion: fada la rottura del canale Donchian, TP al centro"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["1h"]
+
+    def generate_signals(self, df: pd.DataFrame, ts: pd.DatetimeIndex,
+                         **params) -> list[Signal]:
+        n = params.get("n", 20)
+        sl_atr = params.get("sl_atr", 2.0)
+        max_bars = params.get("max_bars", 24)
+        trend_max = params.get("trend_max")          # None = filtro disattivo
+        ema_long = params.get("ema_long", 200)
+
+        h, l, c = df["high"].values, df["low"].values, df["close"].values
+        hh = pd.Series(h).rolling(n).max().shift(1).values
+        ll = pd.Series(l).rolling(n).min().shift(1).values
+        a = atr(df, 14)
+        td = trend_distance(df, ema_long) if trend_max is not None else None
+
+        signals: list[Signal] = []
+        for i in range(n + 14, len(c)):
+            if np.isnan(hh[i]) or np.isnan(a[i]):
+                continue
+            if td is not None and (np.isnan(td[i]) or td[i] > trend_max):
+                continue
+            mid = (hh[i] + ll[i]) / 2.0
+            if c[i] > hh[i] and c[i - 1] <= hh[i - 1]:          # rottura rialzista -> fade short
+                d, sl = -1, c[i] + sl_atr * a[i]
+            elif c[i] < ll[i] and c[i - 1] >= ll[i - 1]:        # rottura ribassista -> fade long
+                d, sl = 1, c[i] - sl_atr * a[i]
+            else:
+                continue
+            signals.append(Signal(
+                idx=i, direction=d, entry_price=c[i],
+                metadata={"tp": float(mid), "sl": float(sl), "max_bars": max_bars},
+            ))
+        return signals
+
+
+if __name__ == "__main__":
+    strat = DonchianFade()
+    print("=" * 110)
+    print(f"  MR02 DONCHIAN FADE — netto fee {strat.fee_rt*100:.2f}% RT, leva {strat.leverage:.0f}x")
+    print("=" * 110)
+    for asset in ["BTC", "ETH"]:
+        r = strat.backtest(asset, "1h", n=20, sl_atr=2.0, max_bars=24)
+        if r:
+            r.strategy_name = f"MR02 {asset} 1h n20"
+            r.print_summary()
+            r.print_yearly()

scripts/strategies/MR07_return_reversal.py

@@ -0,0 +1,88 @@
+"""MR07 — Return Reversal (fade del movimento di barra estremo).
+
+Meccanismo distinto da MR01/MR02/MR03: non guarda i LIVELLI di prezzo (bande,
+canali) ma la VOLATILITA' dei rendimenti. Quando una singola barra si muove di
+piu' di k deviazioni standard rolling dei rendimenti, e' un'over-reaction che
+tende a rientrare: si fada nella direzione opposta. Coerente con la lezione
+mean-reversion.
+
+Logica:
+    1. ret[i] = rendimento dell'ultima barra; sigma = std rolling(n) dei rendimenti
+    2. z = ret[i]/sigma. Se z <= -k (crollo) -> LONG; se z >= +k (spike) -> SHORT.
+       Ingresso a close[i] (eseguibile dal vivo, nessun look-ahead).
+    3. EXIT: take-profit a tp_atr*ATR a favore, stop-loss a sl_atr*ATR contro,
+       time-limit max_bars.
+
+Validazione (netto, fee 0.10% RT reale Deribit, leva 3x, OOS = ultimo 30%):
+    config robusta k=3.5 tp=2ATR sl=1.5ATR n=50:
+    BTC 1h: +447% FULL / +105% OOS, DD 25%
+    ETH 1h: +335% FULL / +195% OOS, DD 46%
+    L'intero blocco tp_atr=2.0 (k in {2.5,3.0,3.5} x sl in {1.5,2.0,2.5}) e'
+    positivo full+OOS su entrambi gli asset 1h.
+    Ricerca completa: scripts/analysis/strategy_research_v2.py.
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+
+from src.strategies.base import Signal
+from src.strategies.fade_base import FadeStrategy, atr, trend_distance
+
+
+class ReturnReversal(FadeStrategy):
+    name = "MR07_return_reversal"
+    description = "Mean-reversion: fada il movimento di barra estremo (z dei rendimenti)"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["1h"]
+
+    def generate_signals(self, df: pd.DataFrame, ts: pd.DatetimeIndex,
+                         **params) -> list[Signal]:
+        n = params.get("n", 50)
+        k = params.get("k", 3.5)
+        tp_atr = params.get("tp_atr", 2.0)
+        sl_atr = params.get("sl_atr", 1.5)
+        max_bars = params.get("max_bars", 24)
+        trend_max = params.get("trend_max")          # None = filtro disattivo
+        ema_long = params.get("ema_long", 200)
+
+        c = df["close"].values
+        ret = np.zeros_like(c)
+        ret[1:] = np.diff(c) / c[:-1]
+        sig = pd.Series(ret).rolling(n).std().values
+        a = atr(df, 14)
+        td = trend_distance(df, ema_long) if trend_max is not None else None
+
+        signals: list[Signal] = []
+        for i in range(n + 14, len(c)):
+            if np.isnan(sig[i]) or sig[i] == 0 or np.isnan(a[i]):
+                continue
+            if td is not None and (np.isnan(td[i]) or td[i] > trend_max):
+                continue
+            z = ret[i] / sig[i]
+            if z <= -k:                                # crollo di barra -> fade long
+                d, tp, sl = 1, c[i] + tp_atr * a[i], c[i] - sl_atr * a[i]
+            elif z >= k:                               # spike di barra -> fade short
+                d, tp, sl = -1, c[i] - tp_atr * a[i], c[i] + sl_atr * a[i]
+            else:
+                continue
+            signals.append(Signal(
+                idx=i, direction=d, entry_price=c[i],
+                metadata={"tp": float(tp), "sl": float(sl), "max_bars": max_bars},
+            ))
+        return signals
+
+
+if __name__ == "__main__":
+    strat = ReturnReversal()
+    print("=" * 110)
+    print(f"  MR07 RETURN REVERSAL — netto fee {strat.fee_rt*100:.2f}% RT, leva {strat.leverage:.0f}x")
+    print("=" * 110)
+    for asset in ["BTC", "ETH"]:
+        r = strat.backtest(asset, "1h", n=50, k=3.5, tp_atr=2.0, sl_atr=1.5, max_bars=24)
+        if r:
+            r.strategy_name = f"MR07 {asset} 1h k3.5"
+            r.print_summary()
+            r.print_yearly()

scripts/strategies/PORT01_portfolio.py

@@ -0,0 +1,65 @@
+"""PORT01 — Portafoglio combinato delle 3 strategie oneste (equal-weight, daily rebal).
+
+Sleeve (meccanismi anti-correlati):
+    DIP01     dip-buy reversion su BTC (1h)        regime: reversione
+    TR01      EMA 20/100 trend su paniere (4h)     regime: momentum singolo
+    ROT02     dual-momentum rotation (1d)          regime: forza relativa + risk-off
+
+La diversificazione e' il vero motore di risk-reduction: il DD del portafoglio
+scende SOTTO quello della sleeve meno rischiosa, mantenendo una CAGR alta e
+azzerando quasi gli anni negativi (il 2022 bear passa da -30% di ROT a -1%).
+
+Risultato (netto, 2021-2026, leva 3x pos 15% per sleeve; ROT02 ora top_k=3):
+    DIP01_BTC     +322%  DD 15%  CAGR 31%
+    TR01_basket   +591%  DD 27%  CAGR 43%
+    ROT02_dualmom +984%  DD 26%  CAGR 56%   (top_k=3: DD 40->26, PnL su)
+    PORTAFOGLIO   +676%  DD 13%  CAGR 46%   <-- DD piu' basso di ogni sleeve
+    Per-anno: 2021 +224 · 2022 +1 · 2023 +48 · 2024 +48 · 2025 +10 · 2026 -2
+Logica e ricostruzione: scripts/analysis/honest_improve2.py.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import pandas as pd
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.honest_improve import _dd  # noqa: E402
+from scripts.analysis.honest_improve2 import (  # noqa: E402
+    dip_market_gated, _daily_equity, _norm, _tr_basket_daily, _rot_daily_equity,
+)
+
+
+def run():
+    idx = pd.date_range("2021-01-01", "2026-05-26", freq="1D", tz="UTC")
+    d = dip_market_gated("BTC", market_n=0, return_equity=True)
+    members = {
+        "DIP01_BTC": _norm(_daily_equity(d["eq_ts"], d["eq_v"], idx)),
+        "TR01_basket": _norm(_tr_basket_daily(["BNB", "BTC", "DOGE", "SOL", "XRP"], idx)),
+        "ROT02_dualmom": _norm(_rot_daily_equity(idx)),
+    }
+    drets = pd.DataFrame({k: v.pct_change().fillna(0) for k, v in members.items()})
+    port_ret = drets.mean(axis=1)
+    combo = (1 + port_ret).cumprod()
+    yrs = (idx[-1] - idx[0]).days / 365.25
+
+    print("=" * 80)
+    print(f"  PORT01 — portafoglio equal-weight (daily rebal) | {idx[0].date()} -> {idx[-1].date()}")
+    print("=" * 80)
+    print(f"  {'sleeve':<16s}{'ret%':>9s}{'DD%':>7s}{'CAGR%':>8s}")
+    for name, s in members.items():
+        r = (s.iloc[-1] / s.iloc[0] - 1) * 100
+        cagr = ((s.iloc[-1] / s.iloc[0]) ** (1 / yrs) - 1) * 100
+        print(f"  {name:<16s}{r:>+9.0f}{_dd(s.values):>7.0f}{cagr:>8.0f}")
+    r = (combo.iloc[-1] / combo.iloc[0] - 1) * 100
+    cagr = ((combo.iloc[-1] / combo.iloc[0]) ** (1 / yrs) - 1) * 100
+    print(f"  {'PORTAFOGLIO':<16s}{r:>+9.0f}{_dd(combo.values):>7.0f}{cagr:>8.0f}")
+    pa = port_ret.groupby(port_ret.index.year).apply(lambda x: ((1 + x).prod() - 1) * 100)
+    print("  Per-anno: " + "  ".join(f"{y}:{v:+.0f}%" for y, v in pa.items()))
+
+
+if __name__ == "__main__":
+    run()

scripts/strategies/PORT02_fade_master.py

@@ -0,0 +1,52 @@
+"""PORT02 — Portafoglio FADE accorpato e migliorato (6 sleeve, equal-weight daily).
+
+Accorpa le 3 strategie fade mean-reversion su BTC e ETH, ciascuna MIGLIORATA con
+il filtro trend (salta i fade contro trend estremi: |close-EMA200|/ATR > 3.0):
+    MR01 Bollinger fade · MR02 Donchian fade · MR07 Return-reversal
+        x  {BTC, ETH}  ->  6 sleeve indipendenti, pos 0.15 ciascuna, leva 3x.
+    (MR03 Keltner spostata in waste: fade piu' debole e ridondante con MR01.)
+
+Le curve sono poco correlate fra loro (corr media intra-fade ~0.18): la
+diversificazione abbatte il DD aggregato ben sotto quello del singolo sleeve.
+
+Risultato (netto fee 0.10% RT, equal-weight ribilanciato ogni giorno, finestra
+comune 2021-2026, OOS = ultimo 30%):
+    Ret +666% / CAGR 46% / DD 8.2% FULL · OOS DD 5.9% / Sharpe 3.95 FULL / 4.09 OOS.
+Ricostruzione e confronto: scripts/analysis/combine_portfolio.py.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.combine_portfolio import (  # noqa: E402
+    build_all_sleeves, port_returns, metrics, yearly_returns, SPLIT, OOS_DATE, IDX,
+)
+
+
+def run():
+    sleeves = {k: v for k, v in build_all_sleeves().items() if k.startswith("MR")}
+    pr = port_returns(sleeves)
+    full, oos = metrics(pr), metrics(pr, lo=SPLIT)
+
+    print("=" * 84)
+    print(f"  PORT02 — FADE master (8 sleeve, equal-weight daily) | {IDX[0].date()} -> {IDX[-1].date()}")
+    print("=" * 84)
+    print(f"  {'sleeve':<14s}{'Ret%':>9s}{'DD%':>7s}{'Shrp':>7s}")
+    for name, s in sleeves.items():
+        m = metrics(s.pct_change().fillna(0.0))
+        print(f"  {name:<14s}{m['ret']:>+9.0f}{m['dd']:>7.1f}{m['sharpe']:>7.2f}")
+    print("  " + "-" * 80)
+    print(f"  {'PORTAFOGLIO':<14s}{full['ret']:>+9.0f}{full['dd']:>7.1f}{full['sharpe']:>7.2f}"
+          f"   CAGR {full['cagr']:.0f}%")
+    print(f"  {'  di cui OOS':<14s}{oos['ret']:>+9.0f}{oos['dd']:>7.1f}{oos['sharpe']:>7.2f}"
+          f"   (da {OOS_DATE})")
+    pa = yearly_returns(pr)
+    print("  Per-anno: " + "  ".join(f"{y}:{v:+.0f}%" for y, v in pa.items()))
+
+
+if __name__ == "__main__":
+    run()

scripts/strategies/PORT03_all_master.py

@@ -0,0 +1,78 @@
+"""PORT03 — Portafoglio MASTER: accorpa TUTTE le strategie (fade + honest).
+
+Unisce le due famiglie del progetto, quasi scorrelate (correlazione cross-famiglia
+~0.05), in un unico portafoglio diversificato:
+    FADE (reversione intraday 1h, long/short, BTC/ETH, col filtro trend):
+        MR01 Bollinger · MR02 Donchian · MR07 Return-reversal  -> 6 sleeve
+    HONEST (long-only multi-regime multi-crypto):
+        DIP01 dip-buy (1h) · TR01 EMA-trend (4h basket) · ROT02 dual-momentum (1d) -> 3 sleeve
+
+Combinare le due famiglie migliora il rischio/rendimento rispetto a ciascuna da
+sola: il DD scende e lo Sharpe sale (la honest, da sola piu' lumpy, viene
+stabilizzata dalle fade ad alta frequenza). Vedi scripts/analysis/combine_portfolio.py.
+
+Risultato (netto, equal-weight daily, finestra comune 2021-2026, OOS = ultimo 30%;
+ROT02 ora top_k=3 -> DD piu' basso):
+    FADE only (6)   DD  8.2%  Sharpe 3.95  (OOS DD 5.9 / Shrp 4.09)
+    HONEST only (3) DD 12.6%  Sharpe 2.20  (OOS DD 7.8 / Shrp 2.02)
+    MASTER eq (9)   DD  5.2%  Sharpe 4.23  (OOS DD 4.7 / Shrp 4.33)   <- miglior Sharpe
+    MASTER 50/50    DD  5.0%  Sharpe 3.69  (OOS DD 4.5)               <- miglior DD
+    CAGR ~47% mantenuta in entrambe le varianti combinate.
+
+Due varianti operative selezionabili:
+    weighting="equal" -> equal-weight sui 9 sleeve (massimo Sharpe)
+    weighting="5050"  -> 50% famiglia fade + 50% famiglia honest (minimo DD)
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.combine_portfolio import (  # noqa: E402
+    build_all_sleeves, port_returns, metrics, yearly_returns, SPLIT, OOS_DATE, IDX,
+)
+
+
+def master_returns(sleeves: dict, weighting: str = "equal"):
+    """Rendimenti giornalieri del portafoglio master.
+    equal = equal-weight su tutti gli 11 sleeve; 5050 = media fra le due famiglie."""
+    fade = {k: v for k, v in sleeves.items() if k.startswith("MR")}
+    honest = {k: v for k, v in sleeves.items() if not k.startswith("MR")}
+    if weighting == "5050":
+        return (port_returns(fade) + port_returns(honest)) / 2
+    return port_returns(sleeves)
+
+
+def run():
+    sleeves = build_all_sleeves()
+    fade = {k: v for k, v in sleeves.items() if k.startswith("MR")}
+    honest = {k: v for k, v in sleeves.items() if not k.startswith("MR")}
+
+    print("=" * 90)
+    print(f"  PORT03 — MASTER (fade + honest) | {IDX[0].date()} -> {IDX[-1].date()} | OOS da {OOS_DATE}")
+    print("=" * 90)
+    print(f"  {'portafoglio':<22s}{'Ret%':>9s}{'CAGR':>7s}{'DD%':>7s}{'Shrp':>7s}"
+          f"   | {'oRet%':>9s}{'oDD%':>7s}{'oShrp':>7s}")
+    print("  " + "-" * 86)
+
+    def line(label, pr):
+        f, o = metrics(pr), metrics(pr, lo=SPLIT)
+        print(f"  {label:<22s}{f['ret']:>+9.0f}{f['cagr']:>7.0f}{f['dd']:>7.1f}{f['sharpe']:>7.2f}"
+              f"   | {o['ret']:>+9.0f}{o['dd']:>7.1f}{o['sharpe']:>7.2f}")
+
+    line(f"FADE only ({len(fade)})", port_returns(fade))
+    line(f"HONEST only ({len(honest)})", port_returns(honest))
+    line(f"MASTER equal ({len(sleeves)})", master_returns(sleeves, "equal"))
+    line("MASTER 50/50 fam", master_returns(sleeves, "5050"))
+
+    print("  " + "-" * 86)
+    pa = yearly_returns(master_returns(sleeves, "equal"))
+    print("  MASTER equal per-anno: " + "  ".join(f"{y}:{v:+.0f}%" for y, v in pa.items()))
+    print("  -> combinare le due famiglie scorrelate (~0.05) abbassa il DD e alza lo Sharpe.")
+
+
+if __name__ == "__main__":
+    run()

scripts/strategies/PR01_pairs_reversion.py

@@ -0,0 +1,78 @@
+"""PR01 — Pairs / Spread Mean-Reversion fra cripto (market-neutral). FAMIGLIA NUOVA.
+
+Distinta da tutto l'esistente (single-asset direzionale): scommette sul RIENTRO del
+log-ratio di due cripto verso la sua media. Market-neutral (long A / short B) ->
+correlazione ~0.02 col mercato -> diversificatore prezioso.
+
+Logica (engine onesto verificato in scripts/analysis/pairs_research.py):
+    r[i] = log(closeA[i]/closeB[i]); z[i] = (r[i]-SMA_n(r)[i]) / STD_n(r)[i]   (causale)
+    z <= -z_in  -> LONG ratio (long A / short B)
+    z >= +z_in  -> SHORT ratio (short A / long B)
+    EXIT: |z| <= z_exit (rientro) o time-limit max_bars. Ingresso/uscita a close.
+    Fee su 2 GAMBE = 2*fee_rt*lev (0.20% RT/coppia). Filtro candele sporche (salto>8%).
+
+Validazione anti-overfit (netto, fee 0.20% RT/coppia a 2 gambe, leva 3x, OOS = ultimo
+30%, CONFIG UNIVERSALE n=50 z_in=2.0 z_exit=0.75 max_bars=72, 1h):
+    ETH/BTC : CAGR 158% / Sharpe 4.36 / 8/9 anni+   (regge fee 6x)
+    LTC/ETH : CAGR  92% / Sharpe 3.08 / 8/8 anni+
+    ADA/ETH : CAGR  91% / Sharpe 2.69 / 7/8 anni+
+    BTC/LTC : CAGR  60% / Sharpe 2.36 / 7/8 anni+   (robusta anche a 4h)
+    ETH/SOL : CAGR  74% / Sharpe 1.96 / 5/7 anni+   (la piu' debole, DD ~63%)
+  - No look-ahead verificato (z[i] invariato perturbando il futuro).
+  - PLATEAU non picco: heatmap n x z_in -> 20/20 celle Sharpe>1 (ETH/BTC, BTC/LTC).
+  - WALK-FORWARD (rolling train 2y / test 6m): ETH/BTC 11/12 finestre positive,
+    BTC/LTC 9/10 -> edge distribuito su tutta la storia, non un regime singolo.
+  - Stress costi: 5/6 reggono fee+slippage realistici; solo ETH/BTC regge fee 6x.
+  - Correlazione con BTC daily ~0.02-0.08 -> market-neutral.
+  - SCARTATA BNB/ETH: robusta solo coi suoi parametri (overfit), crolla con la universale.
+
+LIMITE OPERATIVO: e' una strategia a 2 gambe (long un perp + short l'altro), il worker
+attuale e' single-leg. Per tradarla serve: (a) eseguibilita' short del perp B su
+Deribit/Bybit, (b) gestione 2 ordini + fee doppie. Finche' il worker non supporta
+2 gambe, PR01 resta validata in backtest ma non wired nel paper trader.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.pairs_research import pairs_sim, OOS_FRAC  # noqa: E402
+
+# CONFIG UNIVERSALE (anti-overfit): un'unica terna per TUTTE le coppie, niente
+# cherry-picking per-coppia. Validata come ALTOPIANO (heatmap n x z_in: 20/20 celle
+# Sharpe>1 su ETH/BTC e BTC/LTC) e walk-forward (ETH/BTC 11/12 finestre+, BTC/LTC 9/10).
+UNIV = dict(n=50, z_in=2.0, z_exit=0.75, max_bars=72)
+
+# Coppie robuste con la config universale. Sempre alt-liquido vs major (mai alt/alt).
+# BNB/ETH SCARTATA: era robusta solo coi suoi parametri (n=30, z_exit=1.0) -> overfit;
+# con la config universale crolla (Sharpe 1.5, DD 71%) ed e' la prima a morire allo stress costi.
+PAIRS = [
+    ("ETH", "BTC", UNIV),    # la migliore (Sharpe 4.4 univ), regge fee 6x
+    ("LTC", "ETH", UNIV),
+    ("ADA", "ETH", UNIV),
+    ("BTC", "LTC", UNIV),    # robusta anche a 4h
+    ("ETH", "SOL", UNIV),    # piu' debole (DD ~63%, storia SOL corta) -> peso ridotto
+]
+
+
+def run():
+    print("=" * 92)
+    print("  PR01 — PAIRS spread reversion (market-neutral) | netto fee 0.20% RT/coppia, leva 3x")
+    print("=" * 92)
+    print(f"  {'coppia':<10s}{'trd':>5s}{'win%':>6s}{'CAGR%':>7s}{'OOS DD%':>8s}{'DD%':>6s}{'Shrp':>6s}{'anni+':>7s}")
+    for a, b, p in PAIRS:
+        f = pairs_sim(a, b, **p)
+        o = pairs_sim(a, b, **p, split_frac=1 - OOS_FRAC)
+        yrs = f["yearly"]; pos_y = sum(1 for v in yrs.values() if v > 0)
+        print(f"  {a+'/'+b:<10s}{f['trades']:>5d}{f['win']:>6.1f}{f['cagr']:>7.0f}{o['dd']:>8.0f}"
+              f"{f['dd']:>6.0f}{f['sharpe']:>6.2f}{f'{pos_y}/{len(yrs)}':>7s}")
+    print("\n  Market-neutral (corr ~0.02-0.08 col mercato) -> ottimo diversificatore di portafoglio.")
+    print("  Pattern: solo alt-liquido vs major (BTC/ETH); alt-vs-alt e' rumore.")
+    print("  NB: 2 gambe (long A / short B), fee doppie. Worker live da estendere prima del live.")
+
+
+if __name__ == "__main__":
+    run()

scripts/strategies/ROT02_dual_momentum.py

@@ -0,0 +1,46 @@
+"""ROT02 — Dual-Momentum Rotation (ROT01 + overlay di absolute momentum).
+
+Evoluzione di ROT01: alla rotazione cross-sectional (forza relativa) aggiunge un
+overlay di ABSOLUTE momentum sul mercato: se BTC e' sotto la sua media a `regime_n`
+giorni (mercato risk-off), va completamente in CASH. Cosi' si evitano i bear di
+sistema (2022, 2026 YTD) che erano gli unici anni rossi di ROT01.
+
+Risultato (netto, fee 0.10% RT, gross 0.45, OOS = ultimo 30%): MIGLIORA TUTTO
+rispetto a ROT01.
+    ROT01 base       : FULL  +679% / OOS +44% / DD 53%
+    ROT02 k2 SMA100  : FULL +1095% / OOS +98% / DD 40%   (versione iniziale)
+    ROT02 k3 SMA100  : FULL +1303% / OOS +68% / DD 26%   <-- DD quasi dimezzato, PnL su
+
+MIGLIORIA DD (top_k 2 -> 3): la versione iniziale concentrava il book su 2 soli
+asset; diversificare su 3 dimezza quasi il drawdown (40% -> 26%) e ALZA pure il
+ritorno full (+1095 -> +1303), ret/DD da 27 a 50. Il vol-target abbassa il DD ma
+sacrifica ritorno (solo de-leverage), quindi si tiene top_k=3 senza vol-target.
+Param-insensitive: top_k 3-4 e regime SMA100-150 danno risultati simili.
+Dettagli e sweep in scripts/analysis/honest_improve.py (rot_improved).
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.honest_improve import rot_improved  # noqa: E402
+
+LOOKBACK, TOP_K, REGIME_N = 60, 3, 100   # top_k=3 (era 2): dimezza il DD
+
+
+def run():
+    print("=" * 90)
+    print(f"  ROT02 DUAL-MOMENTUM | 1d lb={LOOKBACK} top{TOP_K} + cash se BTC<SMA{REGIME_N} | netto fee 0.10% RT")
+    print("=" * 90)
+    full = rot_improved(lookback=LOOKBACK, top_k=TOP_K, regime_n=REGIME_N)
+    oos = rot_improved(lookback=LOOKBACK, top_k=TOP_K, regime_n=REGIME_N, oos_frac=0.30)
+    print(f"  FULL: {full['ret']:+.0f}%   DD {full['dd']:.0f}%   ({full['pos_years']}/{full['n_years']} anni positivi)")
+    print(f"  OOS : {oos['ret']:+.0f}%   DD {oos['dd']:.0f}%")
+    print("  Per-anno: " + "  ".join(f"{y}:{v:+.0f}%" for y, v in sorted(full["yearly"].items())))
+
+
+if __name__ == "__main__":
+    run()

scripts/strategies/TR01_ema_trend.py

@@ -0,0 +1,50 @@
+"""TR01 — EMA Trend Following (long-only), timeframe 4h.
+
+Cavalca i trend rialzisti, si mette in cash nei downtrend. Niente short
+(shortare cripto perde OOS nel campione 2018-2026). Complementare a DIP01:
+DIP01 guadagna nei regimi di reversione, TR01 nei regimi di trend.
+
+Logica:
+    1. EMA fast (20) e EMA slow (100) sul close
+    2. LONG quando EMA_fast > EMA_slow (uptrend), altrimenti CASH
+    3. posizione continua, decisione a close[i] (no look-ahead);
+       fee solo sui cambi di stato (poche operazioni = fee non letali)
+
+Validazione (netto, fee 0.10% RT, leva 3x, pos 15%, OOS = ultimo 30%):
+    robusto FULL+OOS su 5/8 asset: BNB(+14), BTC(+27), DOGE(+53), SOL(+7), XRP(+29) OOS.
+    ETH ~flat, ADA/LTC negativi OOS -> preferire BNB/BTC/DOGE/SOL/XRP.
+Dettagli in scripts/analysis/honest_final.py / honest_trend.py.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.honest_trend import (  # noqa: E402
+    simulate_position, ema_dual_signal, oos as trend_oos,
+)
+from scripts.analysis.honest_lab import get_df
+
+ASSETS = ["BNB", "BTC", "DOGE", "SOL", "XRP"]
+FAST, SLOW, TF = 20, 100, "4h"
+
+
+def run():
+    print("=" * 90)
+    print(f"  TR01 EMA TREND {FAST}/{SLOW} long-only | {TF} | netto fee 0.10% RT leva 3x pos 15%")
+    print("=" * 90)
+    print(f"  {'Asset':<6s}{'Flip':>6s}{'FULL%':>9s}{'OOS%':>9s}{'DD%':>6s}{'Exp%':>6s}{'AnniPos':>9s}")
+    for a in ASSETS:
+        df = get_df(a, TF)
+        sig = ema_dual_signal(df, FAST, SLOW, long_only=True)
+        f = simulate_position(sig, df)
+        o = trend_oos(sig, df)
+        print(f"  {a:<6s}{f['flips']:>6d}{f['ret']:>+9.0f}{o['ret']:>+9.0f}"
+              f"{f['dd']:>6.0f}{f['exposure']:>6.0f}{str(f['pos_years'])+'/'+str(f['n_years']):>9s}")
+
+
+if __name__ == "__main__":
+    run()

scripts/waste/AD01_adaptive_squeeze.py

@@ -0,0 +1,205 @@
+"""AD01 — Adaptive Squeeze Threshold.
+
+Problema SQ02: sq_threshold fisso (0.8) non si adatta al regime di volatilità.
+Soluzione: threshold adattivo basato su volatilità recente.
+
+Logica:
+  - Calcola volatilità rolling (std dei rendimenti su finestra 100 barre)
+  - Confronta con percentile storico (rolling 500 barre)
+  - Alta vol (>70° percentile) → soglia BASSA (0.65) — squeeze più "lenti"
+  - Bassa vol (<30° percentile) → soglia ALTA (0.90) — squeeze "stretti"
+  - Vol media → soglia standard (0.80)
+
+Razionale: in mercati calmi, il BB si stringe molto → sq_threshold alto cattura
+segnali migliori. In mercati volatili, bastano squeeze minori per essere significativi.
+
+Anti-overfitting: solo 3 parametri (low_thr, mid_thr, high_thr), logica deterministica.
+Eredita antifakeout + volume da SQ02.
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+
+from src.strategies.base import Strategy, Signal, BacktestResult, YearlyStats, TF_MINUTES
+from src.strategies.indicators import keltner_ratio, ema
+from src.data.downloader import load_data
+
+
+def _adaptive_sq_threshold(close: np.ndarray,
+                            vol_window: int = 100,
+                            regime_window: int = 500,
+                            low_thr: float = 0.65,
+                            mid_thr: float = 0.80,
+                            high_thr: float = 0.90) -> np.ndarray:
+    """Calcola sq_threshold adattivo per ogni barra."""
+    n = len(close)
+    lr = np.diff(np.log(np.where(close <= 0, 1e-10, close)))
+    vol = np.full(n, np.nan)
+    for i in range(vol_window, n):
+        vol[i] = np.std(lr[i - vol_window:i])
+
+    # Percentile rolling della volatilità
+    thresh = np.full(n, mid_thr)
+    for i in range(regime_window, n):
+        if np.isnan(vol[i]):
+            continue
+        hist = vol[i - regime_window:i]
+        hist = hist[~np.isnan(hist)]
+        if len(hist) < 10:
+            continue
+        p30 = np.percentile(hist, 30)
+        p70 = np.percentile(hist, 70)
+        if vol[i] < p30:
+            thresh[i] = high_thr   # vol bassa → soglia alta
+        elif vol[i] > p70:
+            thresh[i] = low_thr    # vol alta → soglia bassa
+        else:
+            thresh[i] = mid_thr
+    return thresh
+
+
+def _detect_adaptive_squeezes(close, high, low, kcr, adaptive_thr,
+                               min_dur: int = 5) -> list[dict]:
+    """Squeeze con threshold adattivo per ogni barra."""
+    events = []
+    in_sq = False
+    sq_start = 0
+    for i in range(1, len(close)):
+        if np.isnan(kcr[i]) or np.isnan(adaptive_thr[i]):
+            continue
+        thr = adaptive_thr[i]
+        is_sq = kcr[i] < thr
+        if is_sq and not in_sq:
+            in_sq = True
+            sq_start = i
+        elif not is_sq and in_sq:
+            in_sq = False
+            dur = i - sq_start
+            if dur < min_dur:
+                continue
+            events.append({
+                "idx": i, "dur": dur, "sq_start": sq_start,
+                "kcr_at_release": kcr[i],
+                "thr_used": adaptive_thr[i],
+            })
+    return events
+
+
+class AdaptiveSqueeze(Strategy):
+    name = "AD01_adaptive_squeeze"
+    description = "Squeeze con threshold adattivo a regime volatilità"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m", "1h"]
+    fee_rt = 0.002
+    leverage = 3.0
+    position_size = 0.15
+    initial_capital = 1000.0
+
+    def generate_signals(self, df: pd.DataFrame, ts: pd.DatetimeIndex,
+                         **params) -> list[Signal]:
+        c = df["close"].values
+        h = df["high"].values
+        l = df["low"].values
+        v = df["volume"].values
+        n = len(c)
+
+        bb_w = params.get("bb_window", 14)
+        low_thr = params.get("low_thr", 0.65)
+        mid_thr = params.get("mid_thr", 0.80)
+        high_thr = params.get("high_thr", 0.90)
+        retrace_limit = params.get("retrace_limit", 0.6)
+        vol_mult = params.get("vol_multiplier", 1.3)
+        use_vol = params.get("use_vol", True)
+        vol_window = params.get("vol_window", 100)
+        regime_window = params.get("regime_window", 500)
+
+        kcr = keltner_ratio(c, h, l, bb_w)
+        adaptive_thr = _adaptive_sq_threshold(
+            c, vol_window, regime_window, low_thr, mid_thr, high_thr
+        )
+        events = _detect_adaptive_squeezes(c, h, l, kcr, adaptive_thr)
+
+        signals = []
+        for ev in events:
+            i = ev["idx"]
+            if i < 1 or i >= n:
+                continue
+
+            first_ret = (c[i] - c[i - 1]) / c[i - 1] if c[i - 1] > 0 else 0
+            if abs(first_ret) < 0.001:
+                continue
+            direction = 1 if first_ret > 0 else -1
+
+            # Anti-fakeout
+            br = h[i] - l[i]
+            if br > 0:
+                if direction == 1 and (h[i] - c[i]) / br > retrace_limit:
+                    continue
+                elif direction == -1 and (c[i] - l[i]) / br > retrace_limit:
+                    continue
+
+            # Volume confirm
+            if use_vol:
+                sq_start = ev["sq_start"]
+                avg_sq_v = np.mean(v[sq_start:i])
+                if avg_sq_v > 0 and v[i] <= avg_sq_v * vol_mult:
+                    continue
+
+            signals.append(Signal(
+                idx=i,
+                direction=direction,
+                entry_price=c[i - 1],
+                metadata={
+                    "dur": ev["dur"],
+                    "thr_used": ev.get("thr_used", mid_thr),
+                },
+            ))
+        return signals
+
+
+if __name__ == "__main__":
+    strategy = AdaptiveSqueeze()
+
+    configs = [
+        # low_thr, mid_thr, high_thr, use_vol
+        {"low_thr": 0.65, "mid_thr": 0.80, "high_thr": 0.90, "use_vol": True},
+        {"low_thr": 0.65, "mid_thr": 0.80, "high_thr": 0.90, "use_vol": False},
+        {"low_thr": 0.60, "mid_thr": 0.78, "high_thr": 0.92, "use_vol": True},
+        {"low_thr": 0.70, "mid_thr": 0.82, "high_thr": 0.90, "use_vol": True},
+        {"low_thr": 0.65, "mid_thr": 0.80, "high_thr": 0.95, "use_vol": True},
+        {"low_thr": 0.65, "mid_thr": 0.80, "high_thr": 0.90,
+         "use_vol": True, "vol_multiplier": 1.2},
+    ]
+
+    all_results = []
+    for cfg in configs:
+        for asset in ["BTC", "ETH"]:
+            for tf in ["15m", "1h"]:
+                for hold in [3, 6]:
+                    r = strategy.backtest(asset, tf, hold=hold, **cfg)
+                    if r and r.trades >= 20:
+                        lbl = (f"AD01 lt={cfg['low_thr']} ht={cfg['high_thr']} "
+                               f"v={cfg['use_vol']} h={hold}")
+                        r.strategy_name = lbl
+                        all_results.append(r)
+
+    all_results.sort(key=lambda r: r.accuracy, reverse=True)
+
+    print(f"\n{'=' * 130}")
+    print("  AD01 ADAPTIVE SQUEEZE THRESHOLD — TOP 20")
+    print(f"{'=' * 130}")
+    print(f"  {'Nome':<50s} {'A/T':>7s} {'Trades':>6s} {'Acc':>6s} "
+          f"{'PnL€':>10s} {'DD%':>6s} {'€/day':>7s} "
+          f"{'Mkt%':>5s} {'Dur':>5s} {'Anni':>4s}")
+    print(f"  {'─' * 120}")
+    for r in all_results[:20]:
+        r.print_summary()
+
+    if all_results:
+        all_results[0].print_yearly()
+
+    print(f"\n  BENCHMARK SQ02: 79.7% acc, 1250t, DD 6.5%, €5.23/day, 9 anni")
+    print(f"  BENCHMARK MT01: 82.7% acc, 503t, DD 5.9%")

scripts/waste/CM01_cross_market_momentum.py

@@ -0,0 +1,183 @@
+"""CM01 — Cross-Market Momentum Filter.
+
+Squeeze su asset primario, entra SOLO se l'altro asset (BTC↔ETH)
+mostra momentum short-term nella STESSA direzione.
+
+Differenza da MT01: MT01 usa EMA slope su 1h (trend lento).
+CM01 usa rendimento grezzo degli ultimi 3-6 bar sull'asset cross
+(momentum veloce, stesso timeframe).
+
+Razionale: BTC e ETH sono altamente correlati ma non perfettamente.
+Se BTC fa squeeze breakout UP e anche ETH sta salendo (momentum 3-6 bar),
+la probabilità di continuazione è maggiore perché c'è consenso di mercato.
+
+Anti-overfitting: 1 parametro chiave (cross_bars 3-6), logica deterministica.
+Eredita antifakeout + volume da SQ02.
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+
+from src.strategies.base import Strategy, Signal, BacktestResult, YearlyStats
+from src.strategies.indicators import keltner_ratio, detect_squeezes
+from src.data.downloader import load_data
+
+
+class CrossMarketMomentum(Strategy):
+    name = "CM01_cross_momentum"
+    description = "Squeeze + cross-asset short-term momentum filter"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m", "1h"]
+    fee_rt = 0.002
+    leverage = 3.0
+    position_size = 0.15
+    initial_capital = 1000.0
+
+    # Map asset → cross asset
+    _CROSS = {"BTC": "ETH", "ETH": "BTC"}
+
+    def generate_signals(self, df: pd.DataFrame, ts: pd.DatetimeIndex,
+                         **params) -> list[Signal]:
+        """Genera segnali con cross-market momentum."""
+        c = df["close"].values
+        h = df["high"].values
+        l = df["low"].values
+        v = df["volume"].values
+        n = len(c)
+        ts_ms = df["timestamp"].values
+
+        asset = params.get("asset", "BTC")
+        tf = params.get("tf", "15m")
+        bb_w = params.get("bb_window", 14)
+        sq_thr = params.get("sq_threshold", 0.8)
+        retrace_limit = params.get("retrace_limit", 0.6)
+        vol_mult = params.get("vol_multiplier", 1.3)
+        use_vol = params.get("use_vol", True)
+        cross_bars = params.get("cross_bars", 4)  # barre momentum cross
+        mom_min = params.get("mom_min", 0.0)      # momentum minimo (0 = solo direzione)
+
+        # Carica cross asset
+        cross_asset = self._CROSS.get(asset)
+        if cross_asset is None:
+            return []
+
+        try:
+            df_cross = load_data(cross_asset, tf)
+        except Exception:
+            return []
+
+        c_cross = df_cross["close"].values
+        ts_cross_ms = df_cross["timestamp"].values
+        n_cross = len(c_cross)
+
+        # Momentum cross: rendimento log su cross_bars barre
+        cross_mom = np.full(n_cross, np.nan)
+        for i in range(cross_bars, n_cross):
+            if c_cross[i - cross_bars] > 0:
+                cross_mom[i] = np.log(c_cross[i] / c_cross[i - cross_bars])
+
+        kcr = keltner_ratio(c, h, l, bb_w)
+        events = detect_squeezes(c, h, l, kcr, sq_thr)
+
+        signals = []
+        for ev in events:
+            i = ev["idx"]
+            if i < 1 or i >= n:
+                continue
+
+            first_ret = (c[i] - c[i - 1]) / c[i - 1] if c[i - 1] > 0 else 0
+            if abs(first_ret) < 0.001:
+                continue
+            direction = 1 if first_ret > 0 else -1
+
+            # Anti-fakeout
+            br = h[i] - l[i]
+            if br > 0:
+                if direction == 1 and (h[i] - c[i]) / br > retrace_limit:
+                    continue
+                elif direction == -1 and (c[i] - l[i]) / br > retrace_limit:
+                    continue
+
+            # Volume confirm
+            if use_vol:
+                sq_start = ev["sq_start"]
+                avg_sq_v = np.mean(v[sq_start:i])
+                if avg_sq_v > 0 and v[i] <= avg_sq_v * vol_mult:
+                    continue
+
+            # Cross-market momentum: trova indice cross corrispondente
+            i_cross = np.searchsorted(ts_cross_ms, ts_ms[i]) - 1
+            if i_cross < cross_bars or i_cross >= n_cross:
+                continue
+            mom = cross_mom[i_cross]
+            if np.isnan(mom):
+                continue
+
+            # Filtra per direzione concordante
+            if direction == 1 and mom <= mom_min:
+                continue
+            if direction == -1 and mom >= -mom_min:
+                continue
+
+            signals.append(Signal(
+                idx=i,
+                direction=direction,
+                entry_price=c[i - 1],
+                metadata={
+                    "dur": ev["dur"],
+                    "cross_mom": float(mom),
+                },
+            ))
+        return signals
+
+
+if __name__ == "__main__":
+    strategy = CrossMarketMomentum()
+
+    configs = [
+        # cross_bars, mom_min, use_vol
+        {"cross_bars": 3, "mom_min": 0.0, "use_vol": True},
+        {"cross_bars": 4, "mom_min": 0.0, "use_vol": True},
+        {"cross_bars": 6, "mom_min": 0.0, "use_vol": True},
+        {"cross_bars": 4, "mom_min": 0.001, "use_vol": True},
+        {"cross_bars": 4, "mom_min": 0.002, "use_vol": True},
+        {"cross_bars": 4, "mom_min": 0.0, "use_vol": False},
+        {"cross_bars": 3, "mom_min": 0.001, "use_vol": False},
+        {"cross_bars": 6, "mom_min": 0.001, "use_vol": True},
+    ]
+
+    all_results = []
+    for cfg in configs:
+        for asset in ["BTC", "ETH"]:
+            for tf in ["15m", "1h"]:
+                for hold in [3, 6]:
+                    r = strategy.backtest(asset, tf, hold=hold,
+                                          cross_bars=cfg["cross_bars"],
+                                          mom_min=cfg["mom_min"],
+                                          use_vol=cfg["use_vol"])
+                    if r and r.trades >= 20:
+                        lbl = (f"CM01 cb={cfg['cross_bars']} "
+                               f"mm={cfg['mom_min']} v={cfg['use_vol']} h={hold}")
+                        r.strategy_name = lbl
+                        all_results.append(r)
+
+    all_results.sort(key=lambda r: r.accuracy, reverse=True)
+
+    print(f"\n{'=' * 130}")
+    print("  CM01 CROSS-MARKET MOMENTUM — TOP 20")
+    print(f"{'=' * 130}")
+    print(f"  {'Nome':<50s} {'A/T':>7s} {'Trades':>6s} {'Acc':>6s} "
+          f"{'PnL€':>10s} {'DD%':>6s} {'€/day':>7s} "
+          f"{'Mkt%':>5s} {'Dur':>5s} {'Anni':>4s}")
+    print(f"  {'─' * 120}")
+    for r in all_results[:20]:
+        r.print_summary()
+
+    if all_results:
+        all_results[0].print_yearly()
+
+    print(f"\n  BENCHMARK SQ02: 79.7% acc, 1250t, DD 6.5%, €5.23/day, 9 anni")
+    print(f"  BENCHMARK MT01: 82.7% acc, 503t, DD 5.9%")

scripts/waste/ML01_squeeze_gbm.py

@@ -0,0 +1,266 @@
+"""ML01 — Squeeze + GBM (Gradient Boosting Machine) Walk-Forward.
+
+Strategia ibrida: squeeze breakout come pre-filtro (QUANDO tradare),
+GradientBoosting su features strutturali come conferma (QUALE direzione).
+
+Pipeline:
+    1. Rileva squeeze release (Bollinger esce da Keltner)
+    2. Estrai 44 features dalla finestra (structural multi-window + squeeze
+       metadata + price position + ATR + momentum breakout)
+    3. GBM walk-forward: train su 50% rolling, step 10%, predice direzione
+    4. Trade solo se ML ha confidenza ≥ ml_threshold
+
+IN:
+    - OHLCV DataFrame
+    - Parametri: bb_window (14), sq_threshold (0.8), brk_bars (3),
+                 ml_threshold (0.70), leverage (3), position_pct (0.15)
+
+OUT:
+    - BacktestResult con metriche walk-forward (no data leakage)
+    - Solo periodo di test (seconda metà dati)
+
+Risultati tipici:
+    ETH 15m bb14 ml=0.70: 76.9% acc, 1213 trades, DD 4.2%, €13.78/day
+    BTC 15m bb14 ml=0.70: 78.8% acc, 1964 trades, DD 7.0%, €5.51/day
+    BTC 1h  bb14 ml=0.70: 77.3% acc,  617 trades, DD 6.7%, €3.85/day
+
+Note:
+    - GBM = GradientBoostingClassifier di scikit-learn
+    - Walk-forward: nessun look-ahead, train sempre prima di test
+    - Il baseline squeeze puro ha accuracy più alta (~79.5%) ma DD peggiore
+    - Il valore del ML è filtrare breakout deboli → DD ridotto
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+from sklearn.ensemble import GradientBoostingClassifier
+from sklearn.preprocessing import StandardScaler
+
+from src.strategies.base import Strategy, Signal, BacktestResult, YearlyStats, TF_MINUTES
+from src.strategies.indicators import keltner_ratio, detect_squeezes
+from src.data.downloader import load_data
+
+
+def _build_features(df: pd.DataFrame, i: int, squeeze_info: dict) -> np.ndarray | None:
+    """44 features per il punto di squeeze release."""
+    if i < 100:
+        return None
+    o, h, l, c, v = (df["open"].values, df["high"].values, df["low"].values,
+                      df["close"].values, df["volume"].values)
+    feats = []
+    for w in [12, 24, 48]:
+        wc, wo = c[i-w:i], o[i-w:i]
+        wh, wl, wv = h[i-w:i], l[i-w:i], v[i-w:i]
+        mn, mx = wl.min(), max(wh.max(), wc.max())
+        rng = mx - mn if mx - mn > 0 else 1e-10
+        total = np.where(wh - wl == 0, 1e-10, wh - wl)
+        body = np.abs(wc - wo) / total
+        direction = np.sign(wc - wo)
+        log_c = np.log(np.where(wc == 0, 1e-10, wc))
+        rets = np.diff(log_c)
+        v_mean = np.mean(wv)
+        feats.extend([
+            np.mean(rets) if len(rets) > 0 else 0,
+            np.std(rets) if len(rets) > 0 else 0,
+            np.sum(rets) if len(rets) > 0 else 0,
+            float(pd.Series(rets).skew()) if len(rets) > 2 else 0,
+            float(pd.Series(rets).kurtosis()) if len(rets) > 3 else 0,
+            np.mean(body), np.std(body),
+            np.mean(direction), np.mean(direction[-min(3, w):]),
+            (wc[-1] - mn) / rng,
+            wv[-1] / v_mean if v_mean > 0 else 1,
+            np.corrcoef(rets[:-1], rets[1:])[0, 1] if len(rets) > 1 and np.std(rets) > 0 else 0,
+        ])
+    sq = squeeze_info
+    feats.extend([
+        sq["dur"], sq["dur"] / 24, sq["kcr_at_release"],
+        v[i-1] / sq.get("avg_vol", 1) if sq.get("avg_vol", 0) > 0 else 1,
+        np.mean(v[i:min(i+3, len(v))]) / sq.get("avg_vol", 1) if sq.get("avg_vol", 0) > 0 else 1,
+    ])
+    h48, l48 = np.max(h[max(0, i-48):i]), np.min(l[max(0, i-48):i])
+    r48 = h48 - l48
+    feats.append((c[i-1] - l48) / r48 if r48 > 0 else 0.5)
+    tr = np.maximum(h[i-14:i] - l[i-14:i],
+                    np.maximum(np.abs(h[i-14:i] - np.roll(c[i-14:i], 1)),
+                               np.abs(l[i-14:i] - np.roll(c[i-14:i], 1))))
+    feats.append(np.mean(tr[1:]) / c[i-1] if c[i-1] > 0 else 0)
+    feats.append((c[i] - c[i-1]) / c[i-1] if c[i-1] > 0 else 0)
+    return np.nan_to_num(np.array(feats), nan=0, posinf=1e6, neginf=-1e6)
+
+
+class SqueezeGBM(Strategy):
+    name = "ML01_squeeze_gbm"
+    description = "Squeeze + GBM walk-forward — ML filtra breakout deboli"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m", "1h"]
+    fee_ml = 0.001
+
+    def generate_signals(self, df, ts, **params):
+        raise NotImplementedError("ML01 usa backtest custom con walk-forward")
+
+    def backtest(self, asset: str, tf: str, hold: int = 3, **params) -> BacktestResult | None:
+        bb_w = params.get("bb_window", 14)
+        sq_thr = params.get("sq_threshold", 0.8 if tf == "1h" else 0.9)
+        brk = params.get("brk_bars", hold)
+        ml_thr = params.get("ml_threshold", 0.70)
+        lev = params.get("leverage", self.leverage)
+        pos = params.get("position_pct", self.position_size)
+
+        df = load_data(asset, tf)
+        close = df["close"].values
+        high = df["high"].values
+        low = df["low"].values
+        volume = df["volume"].values
+        n = len(df)
+        ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+
+        kcr = keltner_ratio(close, high, low, bb_w)
+        raw_events = detect_squeezes(close, high, low, kcr, sq_thr)
+
+        # Aggiungi avg_vol a ogni evento
+        events = []
+        for ev in raw_events:
+            ev["avg_vol"] = float(np.mean(volume[ev["sq_start"]:ev["idx"]]))
+            events.append(ev)
+
+        X_all, y_all, ev_all = [], [], []
+        for ev in events:
+            i = ev["idx"]
+            if i + brk >= n or i < 100:
+                continue
+            feats = _build_features(df, i, ev)
+            if feats is None:
+                continue
+            actual_ret = (close[i + brk - 1] - close[i - 1]) / close[i - 1]
+            X_all.append(feats)
+            y_all.append(1 if actual_ret > 0 else 0)
+            ev_all.append(ev)
+
+        if len(X_all) < 50:
+            return None
+
+        X, y = np.array(X_all), np.array(y_all)
+        TRAIN_SIZE = max(int(len(X) * 0.5), 50)
+        STEP_SIZE = max(int(len(X) * 0.1), 10)
+
+        yearly: dict[int, dict] = {}
+        capital = float(self.initial_capital)
+        peak = capital
+        max_dd = 0.0
+        total_bars = 0
+        all_t = all_w = 0
+
+        start = 0
+        while start + TRAIN_SIZE + STEP_SIZE <= len(X):
+            train_end = start + TRAIN_SIZE
+            test_end = min(train_end + STEP_SIZE, len(X))
+            X_tr, y_tr = X[start:train_end], y[start:train_end]
+            X_te = X[train_end:test_end]
+
+            if len(np.unique(y_tr)) < 2:
+                start += STEP_SIZE
+                continue
+
+            scaler = StandardScaler()
+            X_tr_s = scaler.fit_transform(X_tr)
+            X_te_s = scaler.transform(X_te)
+
+            model = GradientBoostingClassifier(
+                n_estimators=150, max_depth=4, min_samples_leaf=10,
+                learning_rate=0.05, subsample=0.8, random_state=42,
+            )
+            model.fit(X_tr_s, y_tr)
+            up_idx = list(model.classes_).index(1) if 1 in model.classes_ else -1
+            if up_idx < 0:
+                start += STEP_SIZE
+                continue
+
+            for j in range(len(X_te)):
+                proba = model.predict_proba(X_te_s[j:j+1])[0]
+                p_up = proba[up_idx]
+                ev = ev_all[train_end + j]
+                i = ev["idx"]
+                actual_ret = (close[i + brk - 1] - close[i - 1]) / close[i - 1]
+
+                if p_up >= ml_thr:
+                    direction = 1
+                elif p_up <= (1 - ml_thr):
+                    direction = -1
+                else:
+                    continue
+
+                is_correct = (direction == 1 and actual_ret > 0) or (direction == -1 and actual_ret < 0)
+                trade_ret = actual_ret * direction
+                net = trade_ret * lev - self.fee_ml * 2 * lev
+                capital += capital * pos * net
+                capital = max(capital, 10)
+                if capital > peak:
+                    peak = capital
+                dd = (peak - capital) / peak
+                max_dd = max(max_dd, dd)
+                total_bars += brk
+
+                all_t += 1
+                if is_correct:
+                    all_w += 1
+
+                year = ts.iloc[i].year
+                if year not in yearly:
+                    yearly[year] = {"w": 0, "t": 0, "pnl": 0.0}
+                yearly[year]["t"] += 1
+                if is_correct:
+                    yearly[year]["w"] += 1
+                yearly[year]["pnl"] += net * self.initial_capital
+
+            start += STEP_SIZE
+
+        if all_t == 0:
+            return None
+
+        yearly_stats = [
+            YearlyStats(year=y, trades=d["t"], wins=d["w"], pnl=d["pnl"])
+            for y, d in sorted(yearly.items())
+        ]
+
+        return BacktestResult(
+            strategy_name=self.name,
+            asset=asset,
+            timeframe=tf,
+            params={"bb_w": bb_w, "sq_thr": sq_thr, "ml_thr": ml_thr,
+                    "brk": brk, "lev": lev, "pos": pos},
+            trades=all_t,
+            wins=all_w,
+            pnl=sum(d["pnl"] for d in yearly.values()),
+            capital=capital,
+            initial_capital=self.initial_capital,
+            max_dd=max_dd * 100,
+            time_in_market_pct=total_bars / n * 100,
+            avg_trade_duration_h=brk * TF_MINUTES.get(tf, 60) / 60,
+            years_active=len(yearly),
+            yearly=yearly_stats,
+        )
+
+
+if __name__ == "__main__":
+    strategy = SqueezeGBM()
+    print("Training ML models...\n")
+    results = []
+    for asset in ["ETH", "BTC"]:
+        for tf in ["15m", "1h"]:
+            for ml_thr in [0.65, 0.70]:
+                r = strategy.backtest(asset, tf, ml_threshold=ml_thr)
+                if r and r.trades >= 20:
+                    r.strategy_name = f"ML01 {asset} {tf} ml={ml_thr}"
+                    results.append(r)
+    results.sort(key=lambda r: r.accuracy, reverse=True)
+
+    print(f"{'=' * 120}")
+    print(f"  ML01 SQUEEZE+GBM — RISULTATI")
+    print(f"{'=' * 120}")
+    for r in results:
+        r.print_summary()
+    if results:
+        results[0].print_yearly()

scripts/waste/MR03_keltner_fade.py

@@ -0,0 +1,82 @@
+"""MR03 — Keltner Fade (mean-reversion sul canale ATR).
+
+Stessa tesi di MR01 (i breakout rientrano) ma con banda costruita su ATR
+attorno a una EMA, invece che su deviazione standard attorno a una SMA.
+Reagisce diversamente a gap e code: edge indipendente, non ridondante con MR01.
+
+Logica:
+    1. Canale di Keltner: EMA(n) +/- k*ATR(n)
+    2. ENTRY: close esce sotto la banda inferiore -> LONG (o sopra la superiore -> SHORT)
+       Ingresso a close[i] (eseguibile dal vivo, nessun look-ahead).
+    3. EXIT: take-profit alla EMA centrale (il rientro atteso),
+       stop-loss a sl_atr*ATR oltre l'estremo, time-limit max_bars.
+
+Validazione (netto, fee 0.10% RT reale Deribit, leva 3x, OOS = ultimo 30%):
+    BTC 1h n=30 k=2.0: +112% OOS, DD 20%
+    ETH 1h n=50 k=1.5: +1426% OOS, DD 20%
+    Robusto su TUTTA la griglia n in {14,20,30,50} x k in {1.5,2.0,2.5}
+    (BTC+ETH 1h sempre positivo OOS).
+    Ricerca completa: scripts/analysis/strategy_research_v2.py.
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+
+from src.strategies.base import Signal
+from src.strategies.fade_base import FadeStrategy, atr, trend_distance
+
+
+class KeltnerFade(FadeStrategy):
+    name = "MR03_keltner_fade"
+    description = "Mean-reversion: fada il canale di Keltner (ATR), TP alla EMA"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["1h"]
+
+    def generate_signals(self, df: pd.DataFrame, ts: pd.DatetimeIndex,
+                         **params) -> list[Signal]:
+        n = params.get("n", 30)
+        k = params.get("k", 2.0)
+        sl_atr = params.get("sl_atr", 2.0)
+        max_bars = params.get("max_bars", 24)
+        trend_max = params.get("trend_max")          # None = filtro disattivo
+        ema_long = params.get("ema_long", 200)
+
+        c = df["close"].values
+        e = pd.Series(c).ewm(span=n, adjust=False).mean().values
+        a = atr(df, n)
+        up, lo = e + k * a, e - k * a
+        td = trend_distance(df, ema_long) if trend_max is not None else None
+
+        signals: list[Signal] = []
+        for i in range(n + 1, len(c)):
+            if np.isnan(up[i]) or np.isnan(a[i]):
+                continue
+            if td is not None and (np.isnan(td[i]) or td[i] > trend_max):
+                continue
+            if c[i] < lo[i] and c[i - 1] >= lo[i - 1]:
+                d, sl = 1, c[i] - sl_atr * a[i]
+            elif c[i] > up[i] and c[i - 1] <= up[i - 1]:
+                d, sl = -1, c[i] + sl_atr * a[i]
+            else:
+                continue
+            signals.append(Signal(
+                idx=i, direction=d, entry_price=c[i],
+                metadata={"tp": float(e[i]), "sl": float(sl), "max_bars": max_bars},
+            ))
+        return signals
+
+
+if __name__ == "__main__":
+    strat = KeltnerFade()
+    print("=" * 110)
+    print(f"  MR03 KELTNER FADE — netto fee {strat.fee_rt*100:.2f}% RT, leva {strat.leverage:.0f}x")
+    print("=" * 110)
+    for asset, n, k in [("BTC", 30, 2.0), ("ETH", 50, 1.5)]:
+        r = strat.backtest(asset, "1h", n=n, k=k, sl_atr=2.0, max_bars=24)
+        if r:
+            r.strategy_name = f"MR03 {asset} 1h n{n} k{k}"
+            r.print_summary()
+            r.print_yearly()

scripts/waste/MT01_squeeze_mtf_momentum.py

@@ -0,0 +1,261 @@
+"""MT01 — Squeeze + Multi-Timeframe Momentum.
+
+Problema SQ02: entra al breakout 15m ma non sa se il trend 1h è allineato.
+Soluzione: squeeze su 15m + conferma momentum su 1h.
+
+Anti-overfitting: usa solo 2 indicatori (squeeze + EMA slope),
+nessun parametro complesso.
+
+IN:
+    - OHLCV 15m + 1h per lo stesso asset
+    - Parametri: sq_threshold, ema_period_1h, min_slope
+
+OUT:
+    - Signal al breakout 15m confermato da trend 1h
+    - BacktestResult
+
+Logica:
+    1. Squeeze release su 15m (come SQ01)
+    2. Antifakeout filter (come SQ02)
+    3. Check 1h: EMA slope positiva per long, negativa per short
+    4. Check 1h: prezzo sopra/sotto EMA per conferma trend
+    5. Entra solo se 15m e 1h concordano
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+from src.strategies.base import Strategy, Signal, BacktestResult, YearlyStats, TF_MINUTES
+from src.strategies.indicators import keltner_ratio, detect_squeezes, ema
+from src.data.downloader import load_data
+
+
+class SqueezeMTFMomentum(Strategy):
+    name = "MT01_squeeze_mtf"
+    description = "Squeeze 15m + momentum trend 1h — multi-timeframe"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m"]
+    fee_rt = 0.002
+
+    def generate_signals(self, df, ts, **params):
+        """Genera segnali squeeze 15m confermati da trend 1h."""
+        c = df["close"].values
+        h = df["high"].values
+        l = df["low"].values
+        v = df["volume"].values
+        n = len(c)
+
+        asset = params.get("asset", "BTC")
+        sq_thr = params.get("sq_threshold", 0.8)
+        ema_period = params.get("ema_period", 50)
+        min_slope_val = params.get("min_slope", 0.001)
+        use_antifake = params.get("antifake", True)
+        use_vol = params.get("vol_filter", False)
+
+        kcr = keltner_ratio(c, h, l, 14)
+        events = detect_squeezes(c, h, l, kcr, sq_thr)
+
+        df_1h = params.get("df_1h")
+        if df_1h is None:
+            df_1h = load_data(asset, "1h")
+        c1h = df_1h["close"].values
+        ts1h_ms = df_1h["timestamp"].values
+        n1h = len(c1h)
+        ema_1h = ema(c1h, ema_period)
+        ema_slope_arr = np.full(n1h, np.nan)
+        for i in range(5, n1h):
+            if not np.isnan(ema_1h[i]) and not np.isnan(ema_1h[i-5]) and ema_1h[i-5] > 0:
+                ema_slope_arr[i] = (ema_1h[i] - ema_1h[i-5]) / ema_1h[i-5]
+
+        ts_ms = df["timestamp"].values
+        signals = []
+
+        for ev in events:
+            i = ev["idx"]
+            if i < 1 or i >= n:
+                continue
+            first_ret = (c[i] - c[i-1]) / c[i-1] if c[i-1] > 0 else 0
+            if abs(first_ret) < 0.001:
+                continue
+            if use_antifake:
+                br = h[i] - l[i]
+                if br > 0:
+                    if c[i] > c[i-1] and (h[i] - c[i]) / br > 0.6:
+                        continue
+                    elif c[i] <= c[i-1] and (c[i] - l[i]) / br > 0.6:
+                        continue
+            if use_vol:
+                avg_v = np.mean(v[ev["sq_start"]:i])
+                if avg_v > 0 and v[i] <= avg_v * 1.3:
+                    continue
+
+            direction = 1 if first_ret > 0 else -1
+            i1h = np.searchsorted(ts1h_ms, ts_ms[i]) - 1
+            if i1h < ema_period or i1h >= n1h:
+                continue
+            if np.isnan(ema_1h[i1h]) or np.isnan(ema_slope_arr[i1h]):
+                continue
+            if direction == 1:
+                if c1h[i1h] < ema_1h[i1h] or ema_slope_arr[i1h] < min_slope_val:
+                    continue
+            else:
+                if c1h[i1h] > ema_1h[i1h] or ema_slope_arr[i1h] > -min_slope_val:
+                    continue
+
+            signals.append(Signal(idx=i, direction=direction, entry_price=c[i-1]))
+
+        return signals
+
+    def backtest(self, asset, tf="15m", hold=3, **params):
+        sq_thr = params.get("sq_threshold", 0.8)
+        ema_period = params.get("ema_period", 50)
+        min_slope = params.get("min_slope", 0.001)
+        use_antifake = params.get("antifake", True)
+        use_vol = params.get("vol_filter", False)
+
+        # Carica 15m e 1h
+        df_15m = load_data(asset, "15m")
+        df_1h = load_data(asset, "1h")
+
+        c15 = df_15m["close"].values
+        h15 = df_15m["high"].values
+        l15 = df_15m["low"].values
+        v15 = df_15m["volume"].values
+        n15 = len(c15)
+        ts15 = pd.to_datetime(df_15m["timestamp"], unit="ms", utc=True)
+        ts15_ms = df_15m["timestamp"].values
+
+        c1h = df_1h["close"].values
+        ts1h_ms = df_1h["timestamp"].values
+        n1h = len(c1h)
+
+        kcr = keltner_ratio(c15, h15, l15, 14)
+        events = detect_squeezes(c15, h15, l15, kcr, sq_thr)
+
+        # EMA su 1h
+        ema_1h = ema(c1h, ema_period)
+
+        # EMA slope (variazione percentuale su 5 barre)
+        ema_slope = np.full(n1h, np.nan)
+        for i in range(5, n1h):
+            if not np.isnan(ema_1h[i]) and not np.isnan(ema_1h[i - 5]) and ema_1h[i - 5] > 0:
+                ema_slope[i] = (ema_1h[i] - ema_1h[i - 5]) / ema_1h[i - 5]
+
+        yearly = {}
+        capital = float(self.initial_capital)
+        peak = capital
+        max_dd = 0.0
+        total_bars = 0
+
+        for ev in events:
+            i = ev["idx"]
+            if i + hold + 1 >= n15 or i < 1:
+                continue
+
+            first_ret = (c15[i] - c15[i - 1]) / c15[i - 1] if c15[i - 1] > 0 else 0
+            if abs(first_ret) < 0.001:
+                continue
+
+            # Antifake
+            if use_antifake:
+                br = h15[i] - l15[i]
+                if br > 0:
+                    if c15[i] > c15[i - 1] and (h15[i] - c15[i]) / br > 0.6:
+                        continue
+                    elif c15[i] <= c15[i - 1] and (c15[i] - l15[i]) / br > 0.6:
+                        continue
+
+            # Volume filter
+            if use_vol:
+                avg_v = np.mean(v15[ev["sq_start"]:i])
+                if avg_v > 0 and v15[i] <= avg_v * 1.3:
+                    continue
+
+            direction = 1 if first_ret > 0 else -1
+
+            # Trova indice 1h corrispondente
+            i1h = np.searchsorted(ts1h_ms, ts15_ms[i]) - 1
+            if i1h < ema_period or i1h >= n1h or np.isnan(ema_1h[i1h]) or np.isnan(ema_slope[i1h]):
+                continue
+
+            # Conferma trend 1h
+            if direction == 1:
+                if c1h[i1h] < ema_1h[i1h]:
+                    continue
+                if ema_slope[i1h] < min_slope:
+                    continue
+            else:
+                if c1h[i1h] > ema_1h[i1h]:
+                    continue
+                if ema_slope[i1h] > -min_slope:
+                    continue
+
+            entry = c15[i - 1]
+            exit_price = c15[min(i + hold - 1, n15 - 1)]
+            actual = (exit_price - entry) / entry * direction
+            net = actual * self.leverage - self.fee_rt * self.leverage
+
+            capital += capital * self.position_size * net
+            capital = max(capital, 10)
+            if capital > peak: peak = capital
+            dd = (peak - capital) / peak
+            max_dd = max(max_dd, dd)
+            total_bars += hold
+
+            year = ts15.iloc[i].year
+            if year not in yearly:
+                yearly[year] = {"w": 0, "t": 0, "pnl": 0.0}
+            yearly[year]["t"] += 1
+            if actual > 0: yearly[year]["w"] += 1
+            yearly[year]["pnl"] += net * self.initial_capital
+
+        all_t = sum(d["t"] for d in yearly.values())
+        all_w = sum(d["w"] for d in yearly.values())
+        if all_t == 0:
+            return None
+
+        yearly_stats = [YearlyStats(y, d["t"], d["w"], d["pnl"]) for y, d in sorted(yearly.items())]
+        return BacktestResult(
+            strategy_name=self.name, asset=asset, timeframe="15m", params=params,
+            trades=all_t, wins=all_w, pnl=sum(d["pnl"] for d in yearly.values()),
+            capital=capital, initial_capital=self.initial_capital,
+            max_dd=max_dd * 100, time_in_market_pct=total_bars / n15 * 100,
+            avg_trade_duration_h=hold * 15 / 60, years_active=len(yearly), yearly=yearly_stats,
+        )
+
+
+if __name__ == "__main__":
+    strategy = SqueezeMTFMomentum()
+
+    configs = [
+        ("ema50 sl0.1%", {"ema_period": 50, "min_slope": 0.001}),
+        ("ema50 sl0.05%", {"ema_period": 50, "min_slope": 0.0005}),
+        ("ema50 sl0.2%", {"ema_period": 50, "min_slope": 0.002}),
+        ("ema20 sl0.1%", {"ema_period": 20, "min_slope": 0.001}),
+        ("ema50 sl0.1%+vol", {"ema_period": 50, "min_slope": 0.001, "vol_filter": True}),
+        ("ema20 sl0.1%+vol", {"ema_period": 20, "min_slope": 0.001, "vol_filter": True}),
+        ("ema50 noAF", {"ema_period": 50, "min_slope": 0.001, "antifake": False}),
+        ("ema100 sl0.05%", {"ema_period": 100, "min_slope": 0.0005}),
+    ]
+
+    all_results = []
+    for label, params in configs:
+        for asset in ["BTC", "ETH"]:
+            for hold in [3, 6]:
+                r = strategy.backtest(asset, "15m", hold=hold, **params)
+                if r and r.trades >= 30:
+                    r.strategy_name = f"MT01 {label} h={hold}"
+                    all_results.append(r)
+
+    all_results.sort(key=lambda r: r.accuracy, reverse=True)
+    print(f"\n{'=' * 130}")
+    print(f"  MT01 SQUEEZE + MTF MOMENTUM — TOP 20")
+    print(f"{'=' * 130}")
+    for r in all_results[:20]:
+        r.print_summary()
+    if all_results:
+        all_results[0].print_yearly()
+
+    print(f"\n  BENCHMARK SQ02: 79.7% acc, 1250t, DD 6.5%, 9 anni, €5.23/day")

scripts/waste/PD01_price_volume_divergence.py

@@ -0,0 +1,158 @@
+"""PD01 — Price-Volume Divergence Squeeze.
+
+Estende SQ02 con volume TREND come filtro:
+  - Breakout UP con volume CRESCENTE (ultimi 3 bar vs media squeeze) → ENTRA
+  - Breakout UP con volume CALANTE → SALTA (divergenza bearish)
+  - Viceversa per short
+
+Logica anti-fakeout:
+  1. Squeeze rilascio (come SQ02)
+  2. Anti-fakeout candela (come SQ02)
+  3. Volume al breakout > media squeeze (come SQ02)
+  4. NUOVO: volume trending UP nelle ultime 3 barre prima del breakout
+
+Parametri semplici, nessun overfitting.
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+
+from src.strategies.base import Strategy, Signal
+from src.strategies.indicators import keltner_ratio, detect_squeezes
+
+
+class PriceVolumeDivergence(Strategy):
+    name = "PD01_price_vol_div"
+    description = "Squeeze + antifakeout + volume trend confirmation"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m", "1h"]
+    fee_rt = 0.002
+    leverage = 3.0
+    position_size = 0.15
+    initial_capital = 1000.0
+
+    def generate_signals(self, df: pd.DataFrame, ts: pd.DatetimeIndex,
+                         **params) -> list[Signal]:
+        c = df["close"].values
+        h = df["high"].values
+        l = df["low"].values
+        v = df["volume"].values
+        n = len(c)
+
+        bb_w = params.get("bb_window", 14)
+        sq_thr = params.get("sq_threshold", 0.8)
+        retrace_limit = params.get("retrace_limit", 0.6)
+        vol_mult = params.get("vol_multiplier", 1.3)
+        vol_trend_bars = params.get("vol_trend_bars", 3)  # barre per trend volume
+
+        kcr = keltner_ratio(c, h, l, bb_w)
+        events = detect_squeezes(c, h, l, kcr, sq_thr)
+
+        signals = []
+        for ev in events:
+            i = ev["idx"]
+            if i < vol_trend_bars + 1 or i >= n:
+                continue
+
+            # Direzione breakout
+            first_ret = (c[i] - c[i - 1]) / c[i - 1] if c[i - 1] > 0 else 0
+            if abs(first_ret) < 0.001:
+                continue
+            direction = 1 if first_ret > 0 else -1
+
+            # Anti-fakeout
+            br = h[i] - l[i]
+            if br > 0:
+                if direction == 1 and (h[i] - c[i]) / br > retrace_limit:
+                    continue
+                elif direction == -1 and (c[i] - l[i]) / br > retrace_limit:
+                    continue
+
+            # Volume al breakout > media squeeze
+            sq_start = ev["sq_start"]
+            avg_sq_v = np.mean(v[sq_start:i])
+            if avg_sq_v <= 0 or v[i] <= avg_sq_v * vol_mult:
+                continue
+
+            # Volume TREND: slope delle ultime vol_trend_bars barre
+            # Usa regressione lineare semplice (rank correlation del volume)
+            recent_v = v[i - vol_trend_bars:i + 1]  # include breakout bar
+            if len(recent_v) < vol_trend_bars:
+                continue
+            # slope: media seconda metà vs prima metà
+            mid = len(recent_v) // 2
+            v_early = np.mean(recent_v[:mid])
+            v_late = np.mean(recent_v[mid:])
+            vol_trending_up = v_late > v_early
+            vol_trending_down = v_early > v_late
+
+            # Concordanza: long richiede volume trending up, short trending down
+            if direction == 1 and not vol_trending_up:
+                continue
+            if direction == -1 and not vol_trending_down:
+                continue
+
+            signals.append(Signal(
+                idx=i,
+                direction=direction,
+                entry_price=c[i - 1],
+                metadata={
+                    "dur": ev["dur"],
+                    "vol_ratio": v[i] / avg_sq_v if avg_sq_v > 0 else 0,
+                    "vol_trend": v_late / v_early if v_early > 0 else 1,
+                },
+            ))
+        return signals
+
+
+if __name__ == "__main__":
+    strategy = PriceVolumeDivergence()
+
+    configs = [
+        {"bb_window": 14, "sq_threshold": 0.8, "retrace_limit": 0.6,
+         "vol_multiplier": 1.3, "vol_trend_bars": 3},
+        {"bb_window": 14, "sq_threshold": 0.8, "retrace_limit": 0.6,
+         "vol_multiplier": 1.2, "vol_trend_bars": 3},
+        {"bb_window": 14, "sq_threshold": 0.8, "retrace_limit": 0.6,
+         "vol_multiplier": 1.3, "vol_trend_bars": 5},
+        {"bb_window": 14, "sq_threshold": 0.8, "retrace_limit": 0.5,
+         "vol_multiplier": 1.3, "vol_trend_bars": 3},
+        {"bb_window": 14, "sq_threshold": 0.75, "retrace_limit": 0.6,
+         "vol_multiplier": 1.3, "vol_trend_bars": 3},
+        {"bb_window": 20, "sq_threshold": 0.8, "retrace_limit": 0.6,
+         "vol_multiplier": 1.3, "vol_trend_bars": 3},
+    ]
+
+    all_results = []
+    for cfg in configs:
+        for asset in ["BTC", "ETH"]:
+            for tf in ["15m", "1h"]:
+                for hold in [3, 6]:
+                    r = strategy.backtest(asset, tf, hold=hold, **cfg)
+                    if r and r.trades >= 20:
+                        lbl = (f"PD01 vtb={cfg['vol_trend_bars']} "
+                               f"vm={cfg['vol_multiplier']} "
+                               f"sq={cfg['sq_threshold']} h={hold}")
+                        r.strategy_name = lbl
+                        all_results.append(r)
+
+    all_results.sort(key=lambda r: r.accuracy, reverse=True)
+
+    print(f"\n{'=' * 130}")
+    print("  PD01 PRICE-VOLUME DIVERGENCE — TOP 20")
+    print(f"{'=' * 130}")
+    print(f"  {'Nome':<50s} {'A/T':>7s} {'Trades':>6s} {'Acc':>6s} "
+          f"{'PnL€':>10s} {'DD%':>6s} {'€/day':>7s} "
+          f"{'Mkt%':>5s} {'Dur':>5s} {'Anni':>4s}")
+    print(f"  {'─' * 120}")
+    for r in all_results[:20]:
+        r.print_summary()
+
+    if all_results:
+        all_results[0].print_yearly()
+
+    print(f"\n  BENCHMARK SQ02: 79.7% acc, 1250t, DD 6.5%, €5.23/day, 9 anni")
+    print(f"  BENCHMARK MT01: 82.7% acc, 503t, DD 5.9%")

scripts/waste/REF_s3_01_squeeze_improved.py

@@ -0,0 +1,317 @@
+"""S3-01: Squeeze Migliorato — test per-anno, dati reali.
+Miglioramenti rispetto al squeeze base:
+1. Cross-asset: squeeze su BTC + ETH contemporaneo = segnale più forte
+2. Timing orario: accuracy per fascia oraria
+3. Squeeze duration weighted: squeeze lunghi → breakout più forti
+4. Dual-timeframe: squeeze su 1h confermato da 15m
+5. Anti-fakeout: skip se candela post-breakout ritraccia >50%
+6. Dynamic exit: trailing stop basato su ATR
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+from src.data.downloader import load_data
+
+FEE_RT = 0.002
+INITIAL = 1000
+LEVERAGE = 3
+
+
+def keltner_ratio(close, high, low, window=14):
+    n = len(close)
+    r = np.full(n, np.nan)
+    for i in range(window, n):
+        wc, wh, wl = close[i-window:i], high[i-window:i], low[i-window:i]
+        ma = np.mean(wc)
+        bb_std = np.std(wc)
+        tr = np.maximum(wh-wl, np.maximum(np.abs(wh-np.roll(wc,1)), np.abs(wl-np.roll(wc,1))))
+        atr = np.mean(tr[1:])
+        kc = (ma+1.5*atr)-(ma-1.5*atr)
+        bb = (ma+2*bb_std)-(ma-2*bb_std)
+        if kc > 0:
+            r[i] = bb/kc
+    return r
+
+
+def atr_calc(high, low, close, period=14):
+    tr = np.maximum(high-low, np.maximum(np.abs(high-np.roll(close,1)), np.abs(low-np.roll(close,1))))
+    tr[0] = high[0]-low[0]
+    r = np.full(len(close), np.nan)
+    r[period-1] = np.mean(tr[:period])
+    k = 2/(period+1)
+    for i in range(period, len(close)):
+        r[i] = tr[i]*k + r[i-1]*(1-k)
+    return r
+
+
+def detect_squeezes(close, high, low, volume, kcr, sq_thr=0.8, min_dur=5):
+    """Ritorna lista di squeeze events con metadata."""
+    events = []
+    in_sq = False
+    sq_start = 0
+    n = len(close)
+
+    for i in range(1, n):
+        if np.isnan(kcr[i]):
+            continue
+        is_sq = kcr[i] < sq_thr
+        if is_sq and not in_sq:
+            in_sq = True
+            sq_start = i
+        elif not is_sq and in_sq:
+            in_sq = False
+            dur = i - sq_start
+            if dur < min_dur:
+                continue
+            avg_vol = np.mean(volume[sq_start:i])
+            # Range durante squeeze
+            sq_range = (np.max(high[sq_start:i]) - np.min(low[sq_start:i])) / close[sq_start] if close[sq_start] > 0 else 0
+            events.append({
+                "release_idx": i,
+                "duration": dur,
+                "avg_vol": avg_vol,
+                "squeeze_range": sq_range,
+                "kcr_at_release": kcr[i],
+            })
+    return events
+
+
+def run_improved_squeeze(primary_asset, tf="1h"):
+    # Carica asset primario
+    df = load_data(primary_asset, tf)
+    c, h, l, v = df["close"].values, df["high"].values, df["low"].values, df["volume"].values
+    n = len(df)
+    ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+    ts_ms = df["timestamp"].values
+
+    kcr = keltner_ratio(c, h, l, 14)
+    atr_14 = atr_calc(h, l, c, 14)
+    events = detect_squeezes(c, h, l, v, kcr)
+
+    # Carica asset secondario per cross-check
+    secondary = "BTC" if primary_asset == "ETH" else "ETH"
+    df2 = load_data(secondary, tf)
+    c2, h2, l2 = df2["close"].values, df2["high"].values, df2["low"].values
+    ts2_ms = df2["timestamp"].values
+    kcr2 = keltner_ratio(c2, h2, l2, 14)
+
+    # Mappa ts2 → indici per allineare
+    def find_idx2(ts_val):
+        idx = np.searchsorted(ts2_ms, ts_val)
+        return min(idx, len(c2)-1)
+
+    # Carica 15m per dual-TF
+    if tf == "1h":
+        df_15m = load_data(primary_asset, "15m")
+        c15 = df_15m["close"].values
+        h15 = df_15m["high"].values
+        l15 = df_15m["low"].values
+        ts15 = df_15m["timestamp"].values
+        kcr_15m = keltner_ratio(c15, h15, l15, 14)
+    else:
+        kcr_15m = None
+        ts15 = None
+
+    # ================================================================
+    #  CONFIGURAZIONI
+    # ================================================================
+    configs = [
+        # (name, use_cross, use_timing, use_duration, use_dual_tf, use_antifake, use_trailing, hold, stop_atr)
+        ("BASE",                False, False, False, False, False, False, 3, 0),
+        ("cross_asset",         True,  False, False, False, False, False, 3, 0),
+        ("timing_filter",       False, True,  False, False, False, False, 3, 0),
+        ("long_squeeze",        False, False, True,  False, False, False, 3, 0),
+        ("dual_tf",             False, False, False, True,  False, False, 3, 0),
+        ("anti_fakeout",        False, False, False, False, True,  False, 3, 0),
+        ("trailing_stop",       False, False, False, False, False, True,  6, 1.5),
+        ("cross+timing",        True,  True,  False, False, False, False, 3, 0),
+        ("cross+long+timing",   True,  True,  True,  False, False, False, 3, 0),
+        ("cross+dual_tf",       True,  False, False, True,  False, False, 3, 0),
+        ("ALL_FILTERS",         True,  True,  True,  True,  True,  False, 3, 0),
+        ("ALL+trailing",        True,  True,  True,  True,  True,  True,  6, 1.5),
+        ("cross+antifake",      True,  False, False, False, True,  False, 3, 0),
+        ("timing+antifake",     False, True,  False, False, True,  False, 3, 0),
+        ("cross+timing+antifk", True,  True,  False, False, True,  False, 3, 0),
+        ("cross+timing+trail",  True,  True,  False, False, False, True,  6, 1.5),
+    ]
+
+    print(f"\n{'#'*75}")
+    print(f"  {primary_asset} {tf} — SQUEEZE MIGLIORATO")
+    print(f"{'#'*75}")
+
+    results = []
+
+    for name, f_cross, f_timing, f_dur, f_dual, f_antifake, f_trail, hold, stop_atr_m in configs:
+        yearly = {}
+        capital = float(INITIAL)
+        peak = capital
+        max_dd = 0
+
+        for ev in events:
+            i = ev["release_idx"]
+            if i + hold + 2 >= n:
+                continue
+
+            # --- FILTRI ---
+            skip = False
+
+            # Cross-asset: secondary deve anche essere in squeeze recente o breakout
+            if f_cross:
+                i2 = find_idx2(ts_ms[i])
+                if i2 >= 5:
+                    sec_in_squeeze = any(not np.isnan(kcr2[j]) and kcr2[j] < 0.85 for j in range(max(0,i2-10), i2+1))
+                    if not sec_in_squeeze:
+                        skip = True
+
+            # Timing: solo certe ore (testato: 6-14 UTC migliori)
+            if f_timing:
+                hour = ts.iloc[i].hour
+                if hour < 4 or hour > 16:
+                    skip = True
+
+            # Duration: solo squeeze > 10 barre
+            if f_dur:
+                if ev["duration"] < 10:
+                    skip = True
+
+            # Dual-TF: squeeze anche su 15m
+            if f_dual and kcr_15m is not None and ts15 is not None:
+                i15 = np.searchsorted(ts15, ts_ms[i])
+                if i15 >= 5:
+                    sq_15m = any(not np.isnan(kcr_15m[j]) and kcr_15m[j] < 0.85 for j in range(max(0,i15-20), i15+1))
+                    if not sq_15m:
+                        skip = True
+
+            # Anti-fakeout: prima candela post-breakout non deve ritracciare >50%
+            if f_antifake and i + 1 < n:
+                breakout_bar_range = h[i] - l[i]
+                if breakout_bar_range > 0:
+                    if c[i] > c[i-1]:  # breakout up
+                        retrace = (h[i] - c[i]) / breakout_bar_range
+                    else:  # breakout down
+                        retrace = (c[i] - l[i]) / breakout_bar_range
+                    if retrace > 0.6:
+                        skip = True
+
+            if skip:
+                continue
+
+            # --- DIREZIONE ---
+            first_ret = (c[i] - c[i-1]) / c[i-1]
+            if abs(first_ret) < 0.001:
+                continue
+            direction = 1 if first_ret > 0 else -1
+
+            # --- EXIT ---
+            entry = c[i-1]
+            if f_trail and not np.isnan(atr_14[i]):
+                # Trailing stop
+                trail_dist = atr_14[i] * stop_atr_m
+                best_price = entry
+                exit_price = c[min(i+hold, n-1)]
+                for j in range(i, min(i+hold+1, n)):
+                    if direction == 1:
+                        best_price = max(best_price, h[j])
+                        if l[j] <= best_price - trail_dist:
+                            exit_price = best_price - trail_dist
+                            break
+                    else:
+                        best_price = min(best_price, l[j])
+                        if h[j] >= best_price + trail_dist:
+                            exit_price = best_price + trail_dist
+                            break
+                    exit_price = c[j]
+            else:
+                exit_price = c[min(i+hold-1, n-1)]
+
+            actual = (exit_price - entry) / entry * direction
+            net = actual * LEVERAGE - FEE_RT * LEVERAGE
+
+            capital += capital * 0.15 * net
+            capital = max(capital, 10)
+            if capital > peak: peak = capital
+            dd = (peak - capital) / peak
+            max_dd = max(max_dd, dd)
+
+            year = ts.iloc[i].year
+            if year not in yearly:
+                yearly[year] = {"wins": 0, "total": 0, "pnls": []}
+            yearly[year]["total"] += 1
+            if actual > 0:
+                yearly[year]["wins"] += 1
+            yearly[year]["pnls"].append(net * INITIAL)
+
+        all_t = sum(d["total"] for d in yearly.values())
+        all_w = sum(d["wins"] for d in yearly.values())
+        if all_t < 30:
+            continue
+
+        acc = all_w / all_t * 100
+        all_pnls = [p for d in yearly.values() for p in d["pnls"]]
+        tot_pnl = sum(all_pnls)
+
+        # Worst year
+        worst_y_acc = 100
+        worst_y = ""
+        for y, d in yearly.items():
+            ya = d["wins"]/d["total"]*100 if d["total"] > 0 else 0
+            if ya < worst_y_acc:
+                worst_y_acc = ya
+                worst_y = str(y)
+
+        results.append({
+            "name": name, "trades": all_t, "acc": acc, "pnl": tot_pnl,
+            "max_dd": max_dd*100, "capital": capital,
+            "worst": f"{worst_y}({worst_y_acc:.0f}%)",
+            "yearly": yearly,
+        })
+
+    # Sort by accuracy
+    results.sort(key=lambda x: x["acc"], reverse=True)
+
+    print(f"\n  {'Name':.<26s} {'Trades':>7s} {'Acc':>6s} {'PnL€':>9s} {'DD%':>6s} {'Capital':>10s} {'Worst':>12s}")
+    print(f"  {'-'*80}")
+    for r in results:
+        tag = "✅✅" if r["acc"] >= 80 else "✅" if r["acc"] >= 76 else ""
+        print(f"  {r['name']:.<26s} {r['trades']:>7d} {r['acc']:>5.1f}% €{r['pnl']:>+8.0f} {r['max_dd']:>5.1f}% €{r['capital']:>9,.0f} {r['worst']:>12s} {tag}")
+
+    # Dettaglio per anno del migliore
+    if results:
+        best = results[0]
+        print(f"\n  MIGLIORE: {best['name']} → {best['acc']:.1f}% acc")
+        print(f"  {'Anno':>6s} {'Trades':>7s} {'Acc':>6s} {'PnL€':>9s}")
+        for y in sorted(best["yearly"]):
+            d = best["yearly"][y]
+            ya = d["wins"]/d["total"]*100 if d["total"] > 0 else 0
+            yp = sum(d["pnls"])
+            tag = " ← CRASH" if y in [2020,2021,2022] else ""
+            print(f"  {y:>6d} {d['total']:>7d} {ya:>5.1f}% €{yp:>+8.0f}{tag}")
+
+    return results
+
+
+# Run su entrambi gli asset e timeframe
+all_results = {}
+for asset in ["ETH", "BTC"]:
+    for tf in ["1h", "15m"]:
+        key = f"{asset}_{tf}"
+        all_results[key] = run_improved_squeeze(asset, tf)
+
+# Classifica globale
+print(f"\n\n{'='*75}")
+print(f"  CLASSIFICA GLOBALE — TOP 15")
+print(f"{'='*75}")
+
+global_list = []
+for key, results in all_results.items():
+    for r in results:
+        global_list.append({**r, "asset_tf": key})
+
+global_list.sort(key=lambda x: x["acc"], reverse=True)
+print(f"\n  {'Asset_TF':.<12s} {'Name':.<26s} {'Trades':>6s} {'Acc':>6s} {'PnL€':>9s} {'DD%':>5s} {'Worst':>12s}")
+for r in global_list[:15]:
+    tag = "✅✅" if r["acc"] >= 80 else "✅" if r["acc"] >= 76 else ""
+    print(f"  {r['asset_tf']:.<12s} {r['name']:.<26s} {r['trades']:>6d} {r['acc']:>5.1f}% €{r['pnl']:>+8.0f} {r['max_dd']:>4.1f}% {r['worst']:>12s} {tag}")

scripts/waste/REF_s3_02_lead_lag.py

@@ -0,0 +1,290 @@
+"""S3-02: Lead-lag multi-asset squeeze.
+Quando BTC fa squeeze breakout, ETH/SOL spesso seguono.
+Usa il breakout di BTC per anticipare entrata su ETH (e viceversa).
+Testa anche correlazione inter-asset per conferma segnale.
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+from src.data.downloader import load_data
+
+FEE_RT = 0.002
+INITIAL = 1000
+LEVERAGE = 3
+
+
+def keltner_ratio(close, high, low, window=14):
+    n = len(close)
+    r = np.full(n, np.nan)
+    for i in range(window, n):
+        wc, wh, wl = close[i-window:i], high[i-window:i], low[i-window:i]
+        ma = np.mean(wc)
+        bb_std = np.std(wc)
+        tr = np.maximum(wh-wl, np.maximum(np.abs(wh-np.roll(wc,1)), np.abs(wl-np.roll(wc,1))))
+        atr = np.mean(tr[1:])
+        kc = (ma+1.5*atr)-(ma-1.5*atr)
+        bb = (ma+2*bb_std)-(ma-2*bb_std)
+        if kc > 0: r[i] = bb/kc
+    return r
+
+
+def load_aligned(assets, tf):
+    """Carica e allinea dati multi-asset per timestamp."""
+    dfs = {}
+    for asset in assets:
+        try:
+            if asset == "SOL":
+                df = pd.read_parquet(f"data/raw/sol_{tf}.parquet")
+                df["datetime"] = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+            else:
+                df = load_data(asset, tf)
+            dfs[asset] = df
+        except Exception:
+            pass
+
+    if len(dfs) < 2:
+        return None
+
+    # Allinea per timestamp
+    common_ts = set(dfs[list(dfs.keys())[0]]["timestamp"].values)
+    for df in dfs.values():
+        common_ts &= set(df["timestamp"].values)
+    common_ts = sorted(common_ts)
+
+    aligned = {}
+    for asset, df in dfs.items():
+        mask = df["timestamp"].isin(common_ts)
+        aligned[asset] = df[mask].sort_values("timestamp").reset_index(drop=True)
+
+    return aligned
+
+
+def detect_breakouts(close, high, low, volume, kcr, sq_thr=0.8, min_dur=5):
+    """Detect squeeze breakout events."""
+    events = []
+    in_sq = False
+    sq_start = 0
+    for i in range(1, len(close)):
+        if np.isnan(kcr[i]):
+            continue
+        is_sq = kcr[i] < sq_thr
+        if is_sq and not in_sq:
+            in_sq = True
+            sq_start = i
+        elif not is_sq and in_sq:
+            in_sq = False
+            if i - sq_start < min_dur:
+                continue
+            first_ret = (close[i] - close[i-1]) / close[i-1] if close[i-1] > 0 else 0
+            if abs(first_ret) < 0.001:
+                continue
+            events.append({
+                "idx": i,
+                "duration": i - sq_start,
+                "direction": 1 if first_ret > 0 else -1,
+                "first_ret": first_ret,
+            })
+    return events
+
+
+print("=" * 75)
+print("  S3-02: LEAD-LAG MULTI-ASSET SQUEEZE")
+print("=" * 75)
+
+for tf in ["1h", "15m"]:
+    aligned = load_aligned(["BTC", "ETH", "SOL"], tf)
+    if aligned is None:
+        continue
+
+    n = len(aligned["BTC"])
+    ts = pd.to_datetime(aligned["BTC"]["timestamp"], unit="ms", utc=True)
+
+    print(f"\n  Timeframe: {tf}, Candles allineate: {n}")
+
+    # Calcola squeeze per ogni asset
+    asset_data = {}
+    for asset in aligned:
+        df = aligned[asset]
+        c, h, l, v = df["close"].values, df["high"].values, df["low"].values, df["volume"].values
+        kcr = keltner_ratio(c, h, l, 14)
+        events = detect_breakouts(c, h, l, v, kcr)
+        asset_data[asset] = {"close": c, "high": h, "low": l, "vol": v, "kcr": kcr, "events": events}
+        print(f"    {asset}: {len(events)} squeeze breakouts")
+
+    # ================================================================
+    #  STRATEGIA A: Leader-follower
+    #  Quando BTC fa breakout, entra su ETH/SOL nella stessa direzione
+    # ================================================================
+    print(f"\n  --- LEADER-FOLLOWER ({tf}) ---")
+
+    for leader, follower in [("BTC", "ETH"), ("BTC", "SOL"), ("ETH", "BTC"), ("ETH", "SOL")]:
+        if leader not in asset_data or follower not in asset_data:
+            continue
+
+        leader_events = asset_data[leader]["events"]
+        fc = asset_data[follower]["close"]
+
+        for hold in [3, 6]:
+            for delay in [0, 1, 2]:
+                yearly = {}
+
+                for ev in leader_events:
+                    i = ev["idx"] + delay
+                    if i + hold >= n:
+                        continue
+
+                    # Anti-fakeout su follower
+                    entry = fc[i]
+                    exit_price = fc[min(i + hold, n - 1)]
+                    direction = ev["direction"]
+                    actual = (exit_price - entry) / entry * direction
+                    net = actual * LEVERAGE - FEE_RT * LEVERAGE
+
+                    year = ts.iloc[min(i, n-1)].year
+                    if year not in yearly:
+                        yearly[year] = {"w": 0, "t": 0, "pnls": []}
+                    yearly[year]["t"] += 1
+                    if actual > 0:
+                        yearly[year]["w"] += 1
+                    yearly[year]["pnls"].append(net * INITIAL)
+
+                all_t = sum(d["t"] for d in yearly.values())
+                all_w = sum(d["w"] for d in yearly.values())
+                if all_t < 30:
+                    continue
+                acc = all_w / all_t * 100
+                pnl = sum(p for d in yearly.values() for p in d["pnls"])
+                worst_y = min(yearly.items(), key=lambda x: x[1]["w"]/x[1]["t"] if x[1]["t"]>0 else 0)
+                worst_acc = worst_y[1]["w"]/worst_y[1]["t"]*100 if worst_y[1]["t"]>0 else 0
+                tag = "✅" if acc >= 76 else ""
+                print(f"    {leader}→{follower} d={delay} h={hold}: trades={all_t:5d} acc={acc:.1f}% pnl=€{pnl:+.0f} worst={worst_y[0]}({worst_acc:.0f}%) {tag}")
+
+    # ================================================================
+    #  STRATEGIA B: Consensus multi-asset
+    #  Trade solo quando 2+ asset hanno squeeze breakout nello stesso momento
+    # ================================================================
+    print(f"\n  --- CONSENSUS MULTI-ASSET ({tf}) ---")
+
+    # Build event map: timestamp → list of (asset, direction)
+    event_map = {}
+    for asset, data in asset_data.items():
+        for ev in data["events"]:
+            idx = ev["idx"]
+            if idx not in event_map:
+                event_map[idx] = []
+            event_map[idx].append((asset, ev["direction"]))
+
+    for target in ["BTC", "ETH", "SOL"]:
+        if target not in asset_data:
+            continue
+        tc = asset_data[target]["close"]
+
+        for min_consensus in [2, 3]:
+            for window_bars in [1, 3, 5]:
+                yearly = {}
+                daily_done = set()
+
+                for idx in sorted(event_map.keys()):
+                    if idx + 6 >= n:
+                        continue
+
+                    day = ts.iloc[idx].strftime("%Y-%m-%d")
+                    if day in daily_done:
+                        continue
+
+                    # Count consensus within window
+                    nearby_events = []
+                    for j in range(max(0, idx - window_bars), idx + window_bars + 1):
+                        if j in event_map:
+                            nearby_events.extend(event_map[j])
+
+                    # Unique assets
+                    unique_assets = set(a for a, d in nearby_events)
+                    if len(unique_assets) < min_consensus:
+                        continue
+
+                    # Majority direction
+                    dirs = [d for a, d in nearby_events]
+                    majority = 1 if sum(dirs) > 0 else -1
+
+                    entry = tc[idx]
+                    exit_price = tc[min(idx + 3, n - 1)]
+                    actual = (exit_price - entry) / entry * majority
+                    net = actual * LEVERAGE - FEE_RT * LEVERAGE
+
+                    year = ts.iloc[idx].year
+                    if year not in yearly:
+                        yearly[year] = {"w": 0, "t": 0, "pnls": []}
+                    yearly[year]["t"] += 1
+                    if actual > 0:
+                        yearly[year]["w"] += 1
+                    yearly[year]["pnls"].append(net * INITIAL)
+                    daily_done.add(day)
+
+                all_t = sum(d["t"] for d in yearly.values())
+                all_w = sum(d["w"] for d in yearly.values())
+                if all_t < 20:
+                    continue
+                acc = all_w / all_t * 100
+                pnl = sum(p for d in yearly.values() for p in d["pnls"])
+                tag = "✅" if acc >= 76 else ""
+                print(f"    {target} consensus>={min_consensus} w={window_bars}: trades={all_t:4d} acc={acc:.1f}% pnl=€{pnl:+.0f} {tag}")
+
+    # ================================================================
+    #  STRATEGIA C: Correlation-weighted squeeze
+    #  Peso il segnale squeeze in base alla correlazione rolling con BTC
+    # ================================================================
+    print(f"\n  --- CORRELATION-WEIGHTED ({tf}) ---")
+
+    for target in ["ETH", "SOL"]:
+        if target not in asset_data:
+            continue
+        tc = asset_data[target]["close"]
+        btc_c = asset_data["BTC"]["close"]
+
+        # Rolling correlation
+        corr_window = 48  # 48 bars
+        rolling_corr = np.full(n, np.nan)
+        ret_t = np.diff(np.log(np.where(tc == 0, 1e-10, tc)))
+        ret_b = np.diff(np.log(np.where(btc_c == 0, 1e-10, btc_c)))
+        for i in range(corr_window, len(ret_t)):
+            c_val = np.corrcoef(ret_t[i-corr_window:i], ret_b[i-corr_window:i])[0, 1]
+            rolling_corr[i + 1] = c_val if np.isfinite(c_val) else 0
+
+        events = asset_data[target]["events"]
+
+        for corr_thr in [0.5, 0.6, 0.7, 0.8]:
+            yearly = {}
+            for ev in events:
+                i = ev["idx"]
+                if i + 3 >= n or np.isnan(rolling_corr[i]):
+                    continue
+
+                # Solo quando correlazione con BTC è alta
+                if abs(rolling_corr[i]) < corr_thr:
+                    continue
+
+                entry = tc[i - 1]
+                exit_price = tc[min(i + 2, n - 1)]
+                actual = (exit_price - entry) / entry * ev["direction"]
+                net = actual * LEVERAGE - FEE_RT * LEVERAGE
+
+                year = ts.iloc[i].year
+                if year not in yearly:
+                    yearly[year] = {"w": 0, "t": 0, "pnls": []}
+                yearly[year]["t"] += 1
+                if actual > 0:
+                    yearly[year]["w"] += 1
+                yearly[year]["pnls"].append(net * INITIAL)
+
+            all_t = sum(d["t"] for d in yearly.values())
+            all_w = sum(d["w"] for d in yearly.values())
+            if all_t < 20:
+                continue
+            acc = all_w / all_t * 100
+            pnl = sum(p for d in yearly.values() for p in d["pnls"])
+            tag = "✅" if acc >= 76 else ""
+            print(f"    {target} corr>={corr_thr}: trades={all_t:4d} acc={acc:.1f}% pnl=€{pnl:+.0f} {tag}")

scripts/waste/REF_s3_03_ultimate.py

@@ -0,0 +1,256 @@
+"""S3-03: Ultimate Squeeze — combina TUTTI i filtri migliori.
+Filtri che funzionano (testati singolarmente):
+- Anti-fakeout (+1% acc)
+- Long squeeze duration (+1% acc)
+- Cross-asset squeeze simultaneo (+0.5%)
+- Timing 4-16 UTC (+0.5%)
+- Correlation ETH-BTC alta per ETH trades (+1%)
+- Volume confirmation al breakout
+
+Nuovi filtri da testare:
+- Volume delta: up_volume - down_volume al breakout
+- Momentum confirmation: breakout nella direzione del trend 1h
+- Volatility regime: skip in regime estremo (RV > 100%)
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+from src.data.downloader import load_data
+
+FEE_RT = 0.002
+INITIAL = 1000
+LEVERAGE = 3
+
+
+def keltner_ratio(close, high, low, window=14):
+    n = len(close)
+    r = np.full(n, np.nan)
+    for i in range(window, n):
+        wc, wh, wl = close[i-window:i], high[i-window:i], low[i-window:i]
+        ma = np.mean(wc)
+        bb_std = np.std(wc)
+        tr = np.maximum(wh-wl, np.maximum(np.abs(wh-np.roll(wc,1)), np.abs(wl-np.roll(wc,1))))
+        atr = np.mean(tr[1:])
+        kc = (ma+1.5*atr)-(ma-1.5*atr)
+        bb = (ma+2*bb_std)-(ma-2*bb_std)
+        if kc > 0: r[i] = bb/kc
+    return r
+
+
+def ema(arr, period):
+    r = np.full(len(arr), np.nan)
+    k = 2/(period+1)
+    r[period-1] = np.mean(arr[:period])
+    for i in range(period, len(arr)):
+        r[i] = arr[i]*k + r[i-1]*(1-k)
+    return r
+
+
+def rv_ann(close, window):
+    lr = np.diff(np.log(np.where(close == 0, 1e-10, close)))
+    r = np.full(len(close), np.nan)
+    for i in range(window, len(lr)):
+        r[i+1] = np.std(lr[i-window:i]) * np.sqrt(24*365)
+    return r
+
+
+def run_ultimate(primary, tf="15m"):
+    secondary = "ETH" if primary == "BTC" else "BTC"
+
+    df = load_data(primary, tf)
+    c, h, l, v = df["close"].values, df["high"].values, df["low"].values, df["volume"].values
+    n = len(df)
+    ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+
+    df2 = load_data(secondary, tf)
+    c2, ts2 = df2["close"].values, df2["timestamp"].values
+
+    kcr = keltner_ratio(c, h, l, 14)
+    kcr2 = keltner_ratio(c2, df2["high"].values, df2["low"].values, 14)
+
+    ema_50 = ema(c, 50)
+    rv_48 = rv_ann(c, 48)
+
+    # Rolling correlation
+    ret1 = np.diff(np.log(np.where(c == 0, 1e-10, c)))
+    ret2 = np.diff(np.log(np.where(c2[:len(c)] == 0, 1e-10, c2[:len(c)])))
+    min_len = min(len(ret1), len(ret2))
+    ret1 = ret1[:min_len]
+    ret2 = ret2[:min_len]
+    corr = np.full(n, np.nan)
+    for i in range(48, min_len):
+        cv = np.corrcoef(ret1[i-48:i], ret2[i-48:i])[0,1]
+        corr[i+1] = cv if np.isfinite(cv) else 0
+
+    # Detect squeezes
+    events = []
+    in_sq = False
+    sq_start = 0
+    for i in range(15, n):
+        if np.isnan(kcr[i]): continue
+        is_sq = kcr[i] < 0.8
+        if is_sq and not in_sq:
+            in_sq = True
+            sq_start = i
+        elif not is_sq and in_sq:
+            in_sq = False
+            dur = i - sq_start
+            if dur < 5 or i + 6 >= n:
+                continue
+            events.append({"idx": i, "dur": dur, "sq_start": sq_start})
+
+    print(f"\n{'#'*70}")
+    print(f"  {primary} {tf} — ULTIMATE SQUEEZE ({len(events)} squeeze events)")
+    print(f"{'#'*70}")
+
+    filters_map = {
+        "antifake": lambda ev, i: not _antifake(c, h, l, i),
+        "long_sq": lambda ev, i: ev["dur"] >= 10,
+        "timing": lambda ev, i: 4 <= ts.iloc[i].hour <= 16,
+        "cross": lambda ev, i: _cross_squeeze(kcr2, i, ts, ts2),
+        "corr_high": lambda ev, i: not np.isnan(corr[i]) and abs(corr[i]) >= 0.6,
+        "vol_confirm": lambda ev, i: _vol_confirm(v, i, ev["sq_start"]),
+        "trend_align": lambda ev, i: _trend_align(c, ema_50, i),
+        "low_rv": lambda ev, i: not np.isnan(rv_48[i]) and rv_48[i] < 1.5,
+    }
+
+    def _antifake(c, h, l, i):
+        if i + 1 >= len(c): return False
+        br = h[i] - l[i]
+        if br <= 0: return False
+        if c[i] > c[i-1]:
+            return (h[i] - c[i]) / br > 0.6
+        return (c[i] - l[i]) / br > 0.6
+
+    def _cross_squeeze(kcr2, i, ts1, ts2_arr):
+        i2 = np.searchsorted(ts2_arr, ts.values[i].astype("int64") // 10**6)
+        i2 = min(i2, len(kcr2)-1)
+        return any(not np.isnan(kcr2[j]) and kcr2[j] < 0.85 for j in range(max(0,i2-10), i2+1))
+
+    def _vol_confirm(v, i, sq_start):
+        avg = np.mean(v[sq_start:i])
+        return avg > 0 and v[i] > avg * 1.3
+
+    def _trend_align(c, ema_val, i):
+        if np.isnan(ema_val[i]): return True
+        first_ret = (c[i] - c[i-1]) / c[i-1] if c[i-1] > 0 else 0
+        if first_ret > 0:
+            return c[i] > ema_val[i]
+        return c[i] < ema_val[i]
+
+    # Test combinazioni incrementali
+    combos = [
+        ("BASE", []),
+        ("antifake", ["antifake"]),
+        ("long_sq", ["long_sq"]),
+        ("antifake+long", ["antifake", "long_sq"]),
+        ("antifake+timing", ["antifake", "timing"]),
+        ("antifake+cross", ["antifake", "cross"]),
+        ("antifake+corr", ["antifake", "corr_high"]),
+        ("antifake+vol", ["antifake", "vol_confirm"]),
+        ("antifake+trend", ["antifake", "trend_align"]),
+        ("af+long+timing", ["antifake", "long_sq", "timing"]),
+        ("af+long+cross", ["antifake", "long_sq", "cross"]),
+        ("af+long+corr", ["antifake", "long_sq", "corr_high"]),
+        ("af+long+trend", ["antifake", "long_sq", "trend_align"]),
+        ("af+long+cross+time", ["antifake", "long_sq", "cross", "timing"]),
+        ("af+long+corr+time", ["antifake", "long_sq", "corr_high", "timing"]),
+        ("af+long+corr+trend", ["antifake", "long_sq", "corr_high", "trend_align"]),
+        ("ALL_NO_VOL", ["antifake", "long_sq", "cross", "timing", "corr_high", "trend_align", "low_rv"]),
+        ("ALL", ["antifake", "long_sq", "cross", "timing", "corr_high", "vol_confirm", "trend_align", "low_rv"]),
+        ("BEST_5", ["antifake", "long_sq", "corr_high", "trend_align", "low_rv"]),
+    ]
+
+    results = []
+    for combo_name, filter_names in combos:
+        yearly = {}
+        capital = float(INITIAL)
+        peak = capital
+        max_dd = 0
+
+        for ev in events:
+            i = ev["idx"]
+            first_ret = (c[i] - c[i-1]) / c[i-1] if c[i-1] > 0 else 0
+            if abs(first_ret) < 0.001:
+                continue
+
+            skip = False
+            for fn in filter_names:
+                if fn in filters_map and not filters_map[fn](ev, i):
+                    skip = True
+                    break
+            if skip:
+                continue
+
+            direction = 1 if first_ret > 0 else -1
+            entry = c[i-1]
+            exit_price = c[min(i+2, n-1)]
+            actual = (exit_price - entry) / entry * direction
+            net = actual * LEVERAGE - FEE_RT * LEVERAGE
+
+            capital += capital * 0.15 * net
+            capital = max(capital, 10)
+            if capital > peak: peak = capital
+            dd = (peak - capital) / peak
+            max_dd = max(max_dd, dd)
+
+            year = ts.iloc[i].year
+            if year not in yearly:
+                yearly[year] = {"w": 0, "t": 0, "pnls": []}
+            yearly[year]["t"] += 1
+            if actual > 0: yearly[year]["w"] += 1
+            yearly[year]["pnls"].append(net * INITIAL)
+
+        all_t = sum(d["t"] for d in yearly.values())
+        all_w = sum(d["w"] for d in yearly.values())
+        if all_t < 20: continue
+
+        acc = all_w / all_t * 100
+        pnl = sum(p for d in yearly.values() for p in d["pnls"])
+        worst = min(yearly.items(), key=lambda x: x[1]["w"]/x[1]["t"] if x[1]["t"]>0 else 0)
+        wa = worst[1]["w"]/worst[1]["t"]*100 if worst[1]["t"]>0 else 0
+
+        results.append({
+            "name": combo_name, "trades": all_t, "acc": acc, "pnl": pnl,
+            "dd": max_dd*100, "capital": capital, "worst": f"{worst[0]}({wa:.0f}%)",
+            "yearly": yearly,
+        })
+
+    results.sort(key=lambda x: x["acc"], reverse=True)
+
+    print(f"\n  {'Name':.<28s} {'Trades':>6s} {'Acc':>6s} {'PnL€':>9s} {'DD%':>5s} {'Worst':>12s}")
+    print(f"  {'-'*70}")
+    for r in results[:20]:
+        tag = "✅✅" if r["acc"] >= 80 else "✅" if r["acc"] >= 78 else ""
+        print(f"  {r['name']:.<28s} {r['trades']:>6d} {r['acc']:>5.1f}% €{r['pnl']:>+8.0f} {r['dd']:>4.1f}% {r['worst']:>12s} {tag}")
+
+    # Dettaglio migliore
+    if results:
+        best = results[0]
+        print(f"\n  MIGLIORE: {best['name']} → {best['acc']:.1f}% acc, DD {best['dd']:.1f}%")
+        for y in sorted(best["yearly"]):
+            d = best["yearly"][y]
+            ya = d["w"]/d["t"]*100 if d["t"]>0 else 0
+            tag = " ← CRASH" if y in [2020,2021,2022] else ""
+            print(f"    {y}: {d['t']:4d}t {ya:5.1f}% €{sum(d['pnls']):+.0f}{tag}")
+
+    return results
+
+
+all_r = []
+for asset in ["BTC", "ETH"]:
+    for tf in ["15m", "1h"]:
+        r = run_ultimate(asset, tf)
+        for x in r:
+            all_r.append({**x, "key": f"{asset}_{tf}"})
+
+all_r.sort(key=lambda x: x["acc"], reverse=True)
+print(f"\n\n{'='*70}")
+print(f"  TOP 10 GLOBALE")
+print(f"{'='*70}")
+for r in all_r[:10]:
+    tag = "✅✅" if r["acc"] >= 80 else "✅" if r["acc"] >= 78 else ""
+    print(f"  {r['key']:.<10s} {r['name']:.<28s} {r['trades']:>5d} {r['acc']:>5.1f}% €{r['pnl']:>+8.0f} DD {r['dd']:.1f}% {r['worst']:>12s} {tag}")

scripts/waste/ROT01_xsect_rotation.py

@@ -0,0 +1,48 @@
+"""ROT01 — Cross-Sectional Momentum Rotation (multi-crypto, long-only), 1d.
+
+UNA strategia che usa l'INTERO paniere di crypto in un solo book: ogni giorno
+ordina gli asset per momentum (rendimento sugli ultimi `lookback` giorni) e alloca
+il capitale in parti uguali ai `top_k` con momentum positivo; il resto in cash.
+Cattura la dispersione tra crypto (gli alt forti corrono molto piu' di BTC nei bull)
+senza shortare nulla. Meccanismo distinto da DIP01/TR01 -> vera diversificazione.
+
+Onesto: i pesi a close[i] usano solo rendimenti passati; il rendimento del giorno
+i->i+1 e' realizzato con quei pesi. Fee sul turnover. Allineamento per timestamp.
+
+Validazione (netto, fee 0.10% RT, gross 0.45, OOS = ultimo 30%):
+    lb=60 top2 -> FULL +679% / OOS +44% / DD 53% / 5-7 anni positivi.
+    Param-insensitive (tutte le lb/k positive) e regge fee fino 0.20% RT (OOS +41%).
+    Per-anno: 2020+33 2021+181 2022-29 2023+43 2024+59 2025+6 2026-10 (i negativi = bear).
+Dettagli in scripts/analysis/honest_rotation.py / honest_final.py.
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from scripts.analysis.honest_rotation import build_panel, simulate_rotation  # noqa: E402
+from scripts.analysis.honest_lab import available_assets
+
+LOOKBACK, TOP_K, TF = 60, 2, "1d"
+
+
+def run():
+    assets = available_assets()
+    panel = build_panel(assets, TF)
+    print("=" * 90)
+    print(f"  ROT01 ROTAZIONE cross-sectional momentum | {TF} lb={LOOKBACK} top{TOP_K} | netto fee 0.10% RT")
+    print("=" * 90)
+    print(f"  Paniere: {list(panel.columns)}")
+    print(f"  Periodo: {panel.index[0].date()} -> {panel.index[-1].date()} ({panel.shape[0]} barre)")
+    full = simulate_rotation(panel, lookback=LOOKBACK, top_k=TOP_K, fee_rt=0.001)
+    oos = simulate_rotation(panel, lookback=LOOKBACK, top_k=TOP_K, fee_rt=0.001, oos_frac=0.30)
+    print(f"\n  FULL: {full['ret']:+.0f}%  DD {full['dd']:.0f}%  turnover {full['turnover']:.0f}")
+    print(f"  OOS : {oos['ret']:+.0f}%  DD {oos['dd']:.0f}%  ({full['pos_years']}/{full['n_years']} anni positivi)")
+    print("  Per-anno: " + "  ".join(f"{y}:{v:+.0f}%" for y, v in sorted(full["yearly"].items())))
+
+
+if __name__ == "__main__":
+    run()

scripts/waste/SQ01_squeeze_base.py

@@ -0,0 +1,68 @@
+"""SQ01 — Squeeze Breakout Base.
+
+Strategia strutturale: rileva compressione di volatilità (Bollinger dentro
+Keltner Channel) e segue la direzione del breakout al rilascio.
+
+IN:
+    - OHLCV DataFrame (da load_data)
+    - Parametri: bb_window (14), sq_threshold (0.8), min_squeeze_dur (5)
+
+OUT:
+    - Lista di Signal con direzione breakout (+1/-1)
+    - BacktestResult con equity, yearly breakdown, metriche
+
+Risultati tipici:
+    BTC 15m: 76.7% acc, 4062 trades, DD 6.7%, €9.32/day
+    ETH 15m: 76.4% acc, 2948 trades, DD 6.2%, €10.31/day
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+
+from src.strategies.base import Strategy, Signal
+from src.strategies.indicators import keltner_ratio, detect_squeezes
+
+
+class SqueezeBase(Strategy):
+    name = "SQ01_squeeze_base"
+    description = "Squeeze breakout puro — segui direzione al rilascio"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m", "1h"]
+
+    def generate_signals(self, df: pd.DataFrame, ts: pd.DatetimeIndex,
+                         **params) -> list[Signal]:
+        c = df["close"].values
+        h = df["high"].values
+        l = df["low"].values
+        n = len(c)
+
+        bb_w = params.get("bb_window", 14)
+        sq_thr = params.get("sq_threshold", 0.8)
+        min_dur = params.get("min_dur", 5)
+
+        kcr = keltner_ratio(c, h, l, bb_w)
+        events = detect_squeezes(c, h, l, kcr, sq_thr, min_dur)
+
+        signals = []
+        for ev in events:
+            i = ev["idx"]
+            if i < 1 or i >= n:
+                continue
+            first_ret = (c[i] - c[i - 1]) / c[i - 1] if c[i - 1] > 0 else 0
+            if abs(first_ret) < 0.001:
+                continue
+            signals.append(Signal(
+                idx=i,
+                direction=1 if first_ret > 0 else -1,
+                entry_price=c[i - 1],
+                metadata={"dur": ev["dur"], "kcr": ev["kcr_at_release"]},
+            ))
+        return signals
+
+
+if __name__ == "__main__":
+    strategy = SqueezeBase()
+    strategy.report()

scripts/waste/SQ02_squeeze_antifake_vol.py

@@ -0,0 +1,87 @@
+"""SQ02 — Squeeze Breakout + Anti-Fakeout + Volume Confirmation.
+
+Migliora SQ01 con due filtri:
+  1. Anti-fakeout: scarta breakout dove la candela ritraccia >60% del range
+  2. Volume confirm: volume al breakout deve essere >1.3× la media durante squeeze
+
+IN:
+    - OHLCV DataFrame
+    - Parametri: bb_window (14), sq_threshold (0.8), retrace_limit (0.6),
+                 vol_multiplier (1.3)
+
+OUT:
+    - Lista di Signal filtrati
+    - BacktestResult
+
+Risultati tipici:
+    BTC 15m: 79.7% acc, 1250 trades, DD 6.5%, €5.23/day — SOLIDO 9/9 anni
+    ETH 15m: 78.6% acc, 942 trades, DD 3.4%, €4.33/day
+    BTC 1h:  78.0% acc, 473 trades, DD 3.5%, Sharpe 6.57
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+
+from src.strategies.base import Strategy, Signal
+from src.strategies.indicators import keltner_ratio, detect_squeezes
+
+
+class SqueezeAntifakeVol(Strategy):
+    name = "SQ02_antifake_vol"
+    description = "Squeeze + antifakeout + volume confirmation"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m", "1h"]
+
+    def generate_signals(self, df: pd.DataFrame, ts: pd.DatetimeIndex,
+                         **params) -> list[Signal]:
+        c = df["close"].values
+        h = df["high"].values
+        l = df["low"].values
+        v = df["volume"].values
+        n = len(c)
+
+        bb_w = params.get("bb_window", 14)
+        sq_thr = params.get("sq_threshold", 0.8)
+        retrace_limit = params.get("retrace_limit", 0.6)
+        vol_mult = params.get("vol_multiplier", 1.3)
+
+        kcr = keltner_ratio(c, h, l, bb_w)
+        events = detect_squeezes(c, h, l, kcr, sq_thr)
+
+        signals = []
+        for ev in events:
+            i = ev["idx"]
+            if i < 1 or i >= n:
+                continue
+            first_ret = (c[i] - c[i - 1]) / c[i - 1] if c[i - 1] > 0 else 0
+            if abs(first_ret) < 0.001:
+                continue
+
+            br = h[i] - l[i]
+            if br > 0:
+                if c[i] > c[i - 1]:
+                    if (h[i] - c[i]) / br > retrace_limit:
+                        continue
+                else:
+                    if (c[i] - l[i]) / br > retrace_limit:
+                        continue
+
+            avg_v = np.mean(v[ev["sq_start"]:i])
+            if avg_v > 0 and v[i] <= avg_v * vol_mult:
+                continue
+
+            signals.append(Signal(
+                idx=i,
+                direction=1 if first_ret > 0 else -1,
+                entry_price=c[i - 1],
+                metadata={"dur": ev["dur"], "vol_ratio": v[i] / avg_v if avg_v > 0 else 0},
+            ))
+        return signals
+
+
+if __name__ == "__main__":
+    strategy = SqueezeAntifakeVol()
+    strategy.report()

scripts/waste/SQ03_squeeze_all_filters.py

@@ -0,0 +1,175 @@
+"""SQ03 — Squeeze con filtri selezionabili.
+
+Ogni filtro è opzionale e attivabile via parametro. Di default attiva solo
+antifake + long_squeeze (i due filtri con miglior rapporto accuracy/trade).
+Esegue tutte le combinazioni utili e classifica.
+
+Filtri disponibili:
+    - antifake:  scarta breakout con retrace >60% (guadagna ~+1% acc)
+    - long_sq:   solo squeeze durata ≥10 barre (+1% acc, dimezza trade)
+    - timing:    solo ore 4-16 UTC (+0.5% acc)
+    - cross:     asset secondario in squeeze nelle ultime 10 barre (+0.5%)
+    - vol:       volume al breakout >1.3× media squeeze (+1% acc)
+
+IN:
+    - OHLCV DataFrame (primario + secondario per cross-check)
+    - Parametri: filters (lista), bb_window, sq_threshold
+
+OUT:
+    - BacktestResult per ogni preset di filtri
+
+Risultati tipici (BTC 15m):
+    antifake+long:    77.3% acc, 2179 trades
+    antifake+vol:     79.7% acc, 1250 trades — SOLIDO
+    ALL_FILTERS:      79.2% acc,  696 trades (restrittivo)
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+
+from src.strategies.base import Strategy, Signal, BacktestResult, YearlyStats, TF_MINUTES
+from src.strategies.indicators import keltner_ratio, detect_squeezes
+from src.data.downloader import load_data
+
+
+PRESETS = {
+    "antifake":           ["antifake"],
+    "long_sq":            ["long_sq"],
+    "antifake+long":      ["antifake", "long_sq"],
+    "antifake+vol":       ["antifake", "vol"],
+    "antifake+timing":    ["antifake", "timing"],
+    "long+timing":        ["long_sq", "timing"],
+    "antifake+long+time": ["antifake", "long_sq", "timing"],
+    "antifake+cross":     ["antifake", "cross"],
+    "ALL_FILTERS":        ["antifake", "long_sq", "timing", "cross"],
+}
+
+
+class SqueezeFiltered(Strategy):
+    name = "SQ03_filtered"
+    description = "Squeeze + filtri selezionabili (antifake, long, timing, cross, vol)"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m", "1h"]
+
+    def generate_signals(self, df: pd.DataFrame, ts: pd.DatetimeIndex,
+                         **params) -> list[Signal]:
+        c = df["close"].values
+        h = df["high"].values
+        l = df["low"].values
+        v = df["volume"].values
+        n = len(c)
+
+        bb_w = params.get("bb_window", 14)
+        sq_thr = params.get("sq_threshold", 0.8)
+        filters = params.get("filters", ["antifake", "long_sq"])
+        asset = params.get("asset", "BTC")
+        tf = params.get("tf", "15m")
+
+        kcr = keltner_ratio(c, h, l, bb_w)
+        events = detect_squeezes(c, h, l, kcr, sq_thr)
+
+        kcr2 = None
+        ts2 = None
+        if "cross" in filters:
+            secondary = "ETH" if asset == "BTC" else "BTC"
+            df2 = load_data(secondary, tf)
+            kcr2 = keltner_ratio(df2["close"].values, df2["high"].values,
+                                 df2["low"].values, bb_w)
+            ts2 = df2["timestamp"].values
+
+        signals = []
+        for ev in events:
+            i = ev["idx"]
+            if i < 1 or i >= n:
+                continue
+            first_ret = (c[i] - c[i - 1]) / c[i - 1] if c[i - 1] > 0 else 0
+            if abs(first_ret) < 0.001:
+                continue
+
+            skip = False
+
+            if "antifake" in filters:
+                br = h[i] - l[i]
+                if br > 0:
+                    if c[i] > c[i - 1] and (h[i] - c[i]) / br > 0.6:
+                        skip = True
+                    elif c[i] <= c[i - 1] and (c[i] - l[i]) / br > 0.6:
+                        skip = True
+
+            if not skip and "long_sq" in filters:
+                if ev["dur"] < 10:
+                    skip = True
+
+            if not skip and "timing" in filters:
+                hour = ts.iloc[i].hour
+                if hour < 4 or hour > 16:
+                    skip = True
+
+            if not skip and "vol" in filters:
+                avg_v = np.mean(v[ev["sq_start"]:i])
+                if avg_v > 0 and v[i] <= avg_v * 1.3:
+                    skip = True
+
+            if not skip and "cross" in filters and kcr2 is not None and ts2 is not None:
+                i2 = np.searchsorted(ts2, ts.values[i].astype("int64") // 10**6)
+                i2 = min(i2, len(kcr2) - 1)
+                cross_ok = any(
+                    not np.isnan(kcr2[j]) and kcr2[j] < 0.85
+                    for j in range(max(0, i2 - 10), i2 + 1)
+                )
+                if not cross_ok:
+                    skip = True
+
+            if skip:
+                continue
+
+            signals.append(Signal(
+                idx=i,
+                direction=1 if first_ret > 0 else -1,
+                entry_price=c[i - 1],
+                metadata={"dur": ev["dur"], "filters": filters},
+            ))
+        return signals
+
+    def report_all_presets(self, assets=None, timeframes=None, hold=3):
+        """Esegue tutti i preset di filtri × asset × tf."""
+        assets = assets or self.default_assets
+        timeframes = timeframes or self.default_timeframes
+        all_results = []
+
+        for preset_name, filter_list in PRESETS.items():
+            for asset in assets:
+                for tf in timeframes:
+                    r = self.backtest(asset, tf, hold, filters=filter_list)
+                    if r and r.trades >= 20:
+                        r.strategy_name = f"SQ03 {preset_name}"
+                        all_results.append(r)
+
+        all_results.sort(key=lambda r: r.accuracy, reverse=True)
+
+        print(f"\n{'=' * 120}")
+        print(f"  SQ03 SQUEEZE FILTRATO — TUTTI I PRESET ({len(all_results)} config)")
+        print(f"  Fee: {self.fee_rt*100:.1f}% RT | Leva: {self.leverage:.0f}x | Pos: {self.position_size*100:.0f}%")
+        print(f"{'=' * 120}")
+        print(f"  {'Nome':<30s} {'A/T':>7s} {'Trades':>6s} {'Acc':>6s} "
+              f"{'PnL€':>10s} {'DD%':>6s} {'€/day':>7s} "
+              f"{'Mkt%':>5s} {'Dur':>5s} {'Worst':>12s} {'Anni':>4s}")
+        print(f"  {'─' * 110}")
+
+        for r in all_results:
+            r.print_summary()
+
+        if all_results:
+            print(f"\n  MIGLIORE: ", end="")
+            best = all_results[0]
+            best.print_yearly()
+
+        return all_results
+
+
+if __name__ == "__main__":
+    strategy = SqueezeFiltered()
+    strategy.report_all_presets()

scripts/waste/SQ04_squeeze_ultimate.py

@@ -0,0 +1,204 @@
+"""SQ04 — Ultimate Squeeze — combinazione incrementale di tutti i filtri.
+
+Testa combinazioni di filtri (antifake, long_sq, timing, cross-asset,
+correlation, volume, trend alignment, volatility regime) e classifica
+per accuracy.
+
+IN:
+    - OHLCV DataFrame (primario + secondario)
+    - Parametri: bb_window, sq_threshold, lista filtri da attivare
+
+OUT:
+    - BacktestResult per ogni combinazione di filtri
+    - Classifica globale
+
+Risultati tipici:
+    BTC 15m antifake+corr: 81.6% acc (ma concentrato 2018)
+    BTC 15m antifake+vol: 79.7% acc, 1250 trades — robusto
+    ETH 1h antifake+corr: 80.7% acc (solo 2018)
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+
+from src.strategies.base import Strategy, Signal
+from src.strategies.indicators import (
+    keltner_ratio, detect_squeezes, ema, rv_annualized, rolling_correlation,
+)
+from src.data.downloader import load_data
+
+
+class SqueezeUltimate(Strategy):
+    name = "SQ04_ultimate"
+    description = "Ultimate squeeze — tutti i filtri combinabili"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m", "1h"]
+
+    FILTER_PRESETS = {
+        "antifake+vol": ["antifake", "vol_confirm"],
+        "antifake+corr": ["antifake", "corr_high"],
+        "af+long+corr+trend": ["antifake", "long_sq", "corr_high", "trend_align"],
+        "ALL": ["antifake", "long_sq", "cross", "timing", "corr_high",
+                "vol_confirm", "trend_align", "low_rv"],
+    }
+
+    def generate_signals(self, df: pd.DataFrame, ts: pd.DatetimeIndex,
+                         **params) -> list[Signal]:
+        c = df["close"].values
+        h = df["high"].values
+        l = df["low"].values
+        v = df["volume"].values
+        n = len(c)
+
+        asset = params.get("asset", "BTC")
+        tf = params.get("tf", "15m")
+        filters = params.get("filters", ["antifake", "vol_confirm"])
+
+        kcr = keltner_ratio(c, h, l, 14)
+        events = detect_squeezes(c, h, l, kcr)
+
+        secondary = "ETH" if asset == "BTC" else "BTC"
+        df2 = load_data(secondary, tf)
+        c2 = df2["close"].values
+        kcr2 = keltner_ratio(c2, df2["high"].values, df2["low"].values, 14)
+        ts2 = df2["timestamp"].values
+
+        ema_50 = ema(c, 50)
+        rv_48 = rv_annualized(c, 48)
+        corr = rolling_correlation(c, c2)
+
+        signals = []
+        for ev in events:
+            i = ev["idx"]
+            if i < 1 or i >= n:
+                continue
+            first_ret = (c[i] - c[i - 1]) / c[i - 1] if c[i - 1] > 0 else 0
+            if abs(first_ret) < 0.001:
+                continue
+
+            skip = False
+            for f in filters:
+                if f == "antifake":
+                    br = h[i] - l[i]
+                    if br > 0:
+                        if c[i] > c[i-1] and (h[i] - c[i]) / br > 0.6:
+                            skip = True
+                        elif c[i] <= c[i-1] and (c[i] - l[i]) / br > 0.6:
+                            skip = True
+                elif f == "long_sq":
+                    if ev["dur"] < 10:
+                        skip = True
+                elif f == "timing":
+                    if ts.iloc[i].hour < 4 or ts.iloc[i].hour > 16:
+                        skip = True
+                elif f == "cross":
+                    i2 = np.searchsorted(ts2, ts.values[i].astype("int64") // 10**6)
+                    i2 = min(i2, len(kcr2) - 1)
+                    if not any(not np.isnan(kcr2[j]) and kcr2[j] < 0.85
+                               for j in range(max(0, i2 - 10), i2 + 1)):
+                        skip = True
+                elif f == "corr_high":
+                    if np.isnan(corr[i]) or abs(corr[i]) < 0.6:
+                        skip = True
+                elif f == "vol_confirm":
+                    avg_v = np.mean(v[ev["sq_start"]:i])
+                    if avg_v > 0 and v[i] <= avg_v * 1.3:
+                        skip = True
+                elif f == "trend_align":
+                    if not np.isnan(ema_50[i]):
+                        if first_ret > 0 and c[i] < ema_50[i]:
+                            skip = True
+                        elif first_ret < 0 and c[i] > ema_50[i]:
+                            skip = True
+                elif f == "low_rv":
+                    if not np.isnan(rv_48[i]) and rv_48[i] >= 1.5:
+                        skip = True
+                if skip:
+                    break
+
+            if skip:
+                continue
+
+            signals.append(Signal(
+                idx=i,
+                direction=1 if first_ret > 0 else -1,
+                entry_price=c[i - 1],
+                metadata={"dur": ev["dur"], "filters": filters},
+            ))
+        return signals
+
+    def backtest(self, asset: str, tf: str, hold: int = 3, **params):
+        params.setdefault("asset", asset)
+        params.setdefault("tf", tf)
+        df = load_data(asset, tf)
+        ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+        signals = self.generate_signals(df, ts, **params)
+        # Usa il backtest della base ma passando i segnali già generati
+        from src.strategies.base import BacktestResult, YearlyStats, TF_MINUTES
+        c = df["close"].values
+        n = len(c)
+        yearly: dict[int, dict] = {}
+        capital = float(self.initial_capital)
+        peak = capital
+        max_dd = 0.0
+        total_bars = 0
+        for sig in signals:
+            i = sig.idx
+            if i + hold >= n or i < 1:
+                continue
+            entry = sig.entry_price
+            exit_price = c[min(i + hold - 1, n - 1)]
+            actual = (exit_price - entry) / entry * sig.direction
+            net = actual * self.leverage - self.fee_rt * self.leverage
+            capital += capital * self.position_size * net
+            capital = max(capital, 10)
+            if capital > peak: peak = capital
+            dd = (peak - capital) / peak
+            max_dd = max(max_dd, dd)
+            total_bars += hold
+            year = ts.iloc[i].year
+            if year not in yearly:
+                yearly[year] = {"w": 0, "t": 0, "pnl": 0.0}
+            yearly[year]["t"] += 1
+            if actual > 0: yearly[year]["w"] += 1
+            yearly[year]["pnl"] += net * self.initial_capital
+        all_t = sum(d["t"] for d in yearly.values())
+        all_w = sum(d["w"] for d in yearly.values())
+        if all_t == 0: return None
+        yearly_stats = [YearlyStats(y, d["t"], d["w"], d["pnl"]) for y, d in sorted(yearly.items())]
+        return BacktestResult(
+            strategy_name=self.name, asset=asset, timeframe=tf, params=params,
+            trades=all_t, wins=all_w, pnl=sum(d["pnl"] for d in yearly.values()),
+            capital=capital, initial_capital=self.initial_capital,
+            max_dd=max_dd * 100, time_in_market_pct=total_bars / n * 100,
+            avg_trade_duration_h=hold * TF_MINUTES.get(tf, 60) / 60,
+            years_active=len(yearly), yearly=yearly_stats,
+        )
+
+    def report_all_presets(self):
+        """Esegue tutte le combinazioni preset × asset × tf."""
+        all_results = []
+        for preset_name, filter_list in self.FILTER_PRESETS.items():
+            for asset in self.default_assets:
+                for tf in self.default_timeframes:
+                    r = self.backtest(asset, tf, filters=filter_list)
+                    if r and r.trades >= 20:
+                        r.strategy_name = f"SQ04 {preset_name}"
+                        all_results.append(r)
+
+        all_results.sort(key=lambda r: r.accuracy, reverse=True)
+
+        print(f"\n{'=' * 120}")
+        print(f"  SQ04 ULTIMATE — TUTTI I PRESET")
+        print(f"{'=' * 120}")
+        for r in all_results:
+            r.print_summary()
+        return all_results
+
+
+if __name__ == "__main__":
+    strategy = SqueezeUltimate()
+    strategy.report_all_presets()

scripts/waste/W23_inside_bar.py

@@ -0,0 +1,131 @@
+"""IB01 — Inside Bar Breakout.
+
+Pattern di compressione a singola candela: quando una barra ha high < prev high
+E low > prev low, il prezzo si sta comprimendo. Al breakout del range della
+inside bar, segui la direzione.
+
+17% delle candele 15m sono inside bars → frequenza altissima.
+
+IN:
+    - OHLCV DataFrame
+    - Parametri: min_consecutive (N inside bars consecutivi),
+                 volume_filter, breakout_confirm
+
+OUT:
+    - Signal al breakout del range dell'inside bar
+    - BacktestResult
+
+Logica:
+    1. Identifica N inside bars consecutivi (compressione)
+    2. Quando il prezzo rompe il range → entra nella direzione del breakout
+    3. Filtro: volume al breakout > media
+    4. Hold fisso
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+from src.strategies.base import Strategy, Signal
+
+
+class InsideBarBreakout(Strategy):
+    name = "IB01_inside_bar"
+    description = "Inside bar breakout — compressione a singola candela"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m", "1h"]
+    fee_rt = 0.002
+
+    def generate_signals(self, df, ts, **params):
+        c = df["close"].values
+        h = df["high"].values
+        l = df["low"].values
+        v = df["volume"].values
+        n = len(c)
+
+        min_consec = params.get("min_consecutive", 2)
+        use_vol = params.get("vol_filter", False)
+        min_range_pct = params.get("min_range_pct", 0.002)
+
+        # Volume media
+        vol_ma = np.full(n, np.nan)
+        for i in range(20, n):
+            vol_ma[i] = np.mean(v[i - 20:i])
+
+        signals = []
+        consec = 0
+        mother_high = 0.0
+        mother_low = 0.0
+
+        for i in range(1, n - 1):
+            is_inside = h[i] <= h[i - 1] and l[i] >= l[i - 1]
+
+            if is_inside:
+                if consec == 0:
+                    mother_high = h[i - 1]
+                    mother_low = l[i - 1]
+                consec += 1
+            else:
+                if consec >= min_consec:
+                    range_pct = (mother_high - mother_low) / mother_low if mother_low > 0 else 0
+                    if range_pct < min_range_pct:
+                        consec = 0
+                        continue
+
+                    # Breakout detection sulla barra corrente
+                    if c[i] > mother_high:
+                        direction = 1
+                    elif c[i] < mother_low:
+                        direction = -1
+                    else:
+                        consec = 0
+                        continue
+
+                    # Volume filter
+                    if use_vol and not np.isnan(vol_ma[i]):
+                        if v[i] < vol_ma[i] * 1.2:
+                            consec = 0
+                            continue
+
+                    signals.append(Signal(
+                        idx=i, direction=direction, entry_price=c[i],
+                        metadata={"consec": consec, "range_pct": round(range_pct * 100, 3)},
+                    ))
+
+                consec = 0
+
+        return signals
+
+
+if __name__ == "__main__":
+    strategy = InsideBarBreakout()
+
+    configs = [
+        ("2ib", {"min_consecutive": 2}),
+        ("3ib", {"min_consecutive": 3}),
+        ("4ib", {"min_consecutive": 4}),
+        ("2ib+vol", {"min_consecutive": 2, "vol_filter": True}),
+        ("3ib+vol", {"min_consecutive": 3, "vol_filter": True}),
+        ("2ib r>0.3%", {"min_consecutive": 2, "min_range_pct": 0.003}),
+        ("3ib r>0.3%", {"min_consecutive": 3, "min_range_pct": 0.003}),
+    ]
+
+    all_results = []
+    for label, params in configs:
+        for asset in ["BTC", "ETH"]:
+            for tf in ["15m", "1h"]:
+                for hold in [3, 6]:
+                    r = strategy.backtest(asset, tf, hold=hold, **params)
+                    if r and r.trades >= 30:
+                        r.strategy_name = f"IB01 {label} h={hold}"
+                        all_results.append(r)
+
+    all_results.sort(key=lambda r: r.accuracy, reverse=True)
+    print(f"\n{'=' * 120}")
+    print(f"  IB01 INSIDE BAR BREAKOUT — TOP 20")
+    print(f"{'=' * 120}")
+    for r in all_results[:20]:
+        r.print_summary()
+    if all_results:
+        all_results[0].print_yearly()

scripts/waste/W24_donchian.py

@@ -0,0 +1,133 @@
+"""DC01 — Donchian Channel Breakout con filtri.
+
+Trend-following classico: quando il prezzo rompe il massimo/minimo degli
+ultimi N periodi, entra nella direzione del breakout.
+
+Completamente diverso dallo squeeze (che usa Bollinger/Keltner).
+Donchian cattura breakout di RANGE, non di VOLATILITÀ.
+
+IN:
+    - OHLCV DataFrame
+    - Parametri: channel_period, volume_filter, atr_stop, trend_filter
+
+OUT:
+    - Signal al breakout del canale Donchian
+    - BacktestResult
+
+Logica:
+    1. Donchian upper = max(high, N periodi), lower = min(low, N periodi)
+    2. Close > upper → LONG (breakout rialzista)
+    3. Close < lower → SHORT (breakout ribassista)
+    4. Filtri: volume, trend EMA, ATR minimo
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+from src.strategies.base import Strategy, Signal
+
+
+class DonchianBreakout(Strategy):
+    name = "DC01_donchian"
+    description = "Donchian Channel breakout — trend-following su range"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m", "1h"]
+    fee_rt = 0.002
+
+    def generate_signals(self, df, ts, **params):
+        c = df["close"].values
+        h = df["high"].values
+        l = df["low"].values
+        v = df["volume"].values
+        n = len(c)
+
+        period = params.get("channel_period", 48)
+        use_vol = params.get("vol_filter", False)
+        use_trend = params.get("trend_filter", False)
+        cooldown = params.get("cooldown", 6)
+
+        # EMA per trend filter
+        ema_50 = np.full(n, np.nan)
+        k = 2 / 51
+        ema_50[49] = np.mean(c[:50])
+        for i in range(50, n):
+            ema_50[i] = c[i] * k + ema_50[i - 1] * (1 - k)
+
+        # Volume media
+        vol_ma = np.full(n, np.nan)
+        for i in range(20, n):
+            vol_ma[i] = np.mean(v[i - 20:i])
+
+        signals = []
+        last_signal_idx = -cooldown
+
+        for i in range(period + 1, n):
+            if i - last_signal_idx < cooldown:
+                continue
+
+            upper = np.max(h[i - period:i])
+            lower = np.min(l[i - period:i])
+
+            direction = 0
+            if c[i] > upper:
+                direction = 1
+            elif c[i] < lower:
+                direction = -1
+
+            if direction == 0:
+                continue
+
+            # Trend filter: breakout must align with EMA trend
+            if use_trend and not np.isnan(ema_50[i]):
+                if direction == 1 and c[i] < ema_50[i]:
+                    continue
+                if direction == -1 and c[i] > ema_50[i]:
+                    continue
+
+            # Volume filter
+            if use_vol and not np.isnan(vol_ma[i]):
+                if v[i] < vol_ma[i] * 1.3:
+                    continue
+
+            signals.append(Signal(
+                idx=i, direction=direction, entry_price=c[i],
+                metadata={"upper": float(upper), "lower": float(lower)},
+            ))
+            last_signal_idx = i
+
+        return signals
+
+
+if __name__ == "__main__":
+    strategy = DonchianBreakout()
+
+    configs = [
+        ("p=24", {"channel_period": 24}),
+        ("p=48", {"channel_period": 48}),
+        ("p=96", {"channel_period": 96}),
+        ("p=48+trend", {"channel_period": 48, "trend_filter": True}),
+        ("p=48+vol", {"channel_period": 48, "vol_filter": True}),
+        ("p=48+t+v", {"channel_period": 48, "trend_filter": True, "vol_filter": True}),
+        ("p=96+t+v", {"channel_period": 96, "trend_filter": True, "vol_filter": True}),
+    ]
+
+    all_results = []
+    for label, params in configs:
+        for asset in ["BTC", "ETH"]:
+            for tf in ["15m", "1h"]:
+                for hold in [3, 6, 12]:
+                    r = strategy.backtest(asset, tf, hold=hold, **params)
+                    if r and r.trades >= 30:
+                        r.strategy_name = f"DC01 {label} h={hold}"
+                        all_results.append(r)
+
+    all_results.sort(key=lambda r: r.accuracy, reverse=True)
+    print(f"\n{'=' * 120}")
+    print(f"  DC01 DONCHIAN BREAKOUT — TOP 20")
+    print(f"{'=' * 120}")
+    for r in all_results[:20]:
+        r.print_summary()
+    if all_results:
+        all_results[0].print_yearly()

scripts/waste/W25_squeeze_retest.py

@@ -0,0 +1,163 @@
+"""SB01 — Squeeze Breakout con Retest.
+
+Il problema di SQ01/SQ02: entri al breakout, ma molti breakout sono fakeout.
+Soluzione: aspetta il RETEST. Dopo il breakout, il prezzo spesso torna a
+testare il livello di breakout prima di continuare.
+
+Più selettivo di SQ02 → meno trade ma più accurati.
+Anti-overfitting: meccanismo strutturale (retest è fenomeno di mercato reale).
+
+IN:
+    - OHLCV DataFrame
+    - Parametri: bb_window, sq_threshold, retest_window (quante barre aspettare
+                 il retest), retest_tolerance (quanto può tornare indietro)
+
+OUT:
+    - Signal al retest confermato (non al breakout iniziale)
+    - BacktestResult
+
+Logica:
+    1. Rileva squeeze release (come SQ01)
+    2. NON entrare subito — segna direzione e livello di breakout
+    3. Nelle N barre successive, aspetta che il prezzo torni verso il livello
+    4. Se il prezzo torna nel range di tolleranza e poi rimbalza → ENTRA
+    5. Se il prezzo non torna → skip (momentum troppo forte, entry persa)
+    6. Se il prezzo sfonda il livello → fakeout confermato, skip
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+from src.strategies.base import Strategy, Signal
+from src.strategies.indicators import keltner_ratio, detect_squeezes
+
+
+class SqueezeBreakoutRetest(Strategy):
+    name = "SB01_squeeze_retest"
+    description = "Squeeze breakout con retest — entra solo dopo pullback confermato"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m", "1h"]
+    fee_rt = 0.002
+
+    def generate_signals(self, df, ts, **params):
+        c = df["close"].values
+        h = df["high"].values
+        l = df["low"].values
+        v = df["volume"].values
+        n = len(c)
+
+        bb_w = params.get("bb_window", 14)
+        sq_thr = params.get("sq_threshold", 0.8)
+        retest_window = params.get("retest_window", 8)
+        retest_tol = params.get("retest_tolerance", 0.5)
+        use_vol = params.get("vol_filter", False)
+
+        kcr = keltner_ratio(c, h, l, bb_w)
+        events = detect_squeezes(c, h, l, kcr, sq_thr)
+
+        vol_ma = np.full(n, np.nan)
+        for i in range(20, n):
+            vol_ma[i] = np.mean(v[i - 20:i])
+
+        signals = []
+
+        for ev in events:
+            brk_idx = ev["idx"]
+            if brk_idx + retest_window + 3 >= n or brk_idx < 1:
+                continue
+
+            # Direzione breakout
+            first_ret = (c[brk_idx] - c[brk_idx - 1]) / c[brk_idx - 1]
+            if abs(first_ret) < 0.001:
+                continue
+
+            direction = 1 if first_ret > 0 else -1
+            breakout_level = c[brk_idx - 1]
+            breakout_move = abs(first_ret)
+
+            # Aspetta retest nelle prossime N barre
+            retest_found = False
+            retest_idx = -1
+
+            for j in range(brk_idx + 1, min(brk_idx + retest_window + 1, n)):
+                if direction == 1:
+                    # Long: il prezzo deve tornare GIÙ verso breakout_level
+                    pullback = (h[brk_idx] - l[j]) / (h[brk_idx] - breakout_level) if h[brk_idx] > breakout_level else 0
+                    if pullback >= retest_tol:
+                        # Tornato abbastanza — ora deve rimbalzare
+                        if c[j] > breakout_level:
+                            retest_found = True
+                            retest_idx = j
+                            break
+                        elif c[j] < breakout_level * 0.998:
+                            # Sfondato sotto → fakeout
+                            break
+                else:
+                    # Short: il prezzo deve tornare SU verso breakout_level
+                    pullback = (h[j] - l[brk_idx]) / (breakout_level - l[brk_idx]) if breakout_level > l[brk_idx] else 0
+                    if pullback >= retest_tol:
+                        if c[j] < breakout_level:
+                            retest_found = True
+                            retest_idx = j
+                            break
+                        elif c[j] > breakout_level * 1.002:
+                            break
+
+            if not retest_found or retest_idx < 0:
+                continue
+
+            # Volume filter al retest
+            if use_vol and not np.isnan(vol_ma[retest_idx]):
+                if v[retest_idx] < vol_ma[retest_idx] * 0.8:
+                    continue
+
+            signals.append(Signal(
+                idx=retest_idx, direction=direction,
+                entry_price=c[retest_idx],
+                metadata={
+                    "breakout_idx": brk_idx,
+                    "retest_bars": retest_idx - brk_idx,
+                    "breakout_move": round(breakout_move * 100, 3),
+                },
+            ))
+
+        return signals
+
+
+if __name__ == "__main__":
+    strategy = SqueezeBreakoutRetest()
+
+    configs = [
+        ("rt8 tol50%", {"retest_window": 8, "retest_tolerance": 0.5}),
+        ("rt6 tol50%", {"retest_window": 6, "retest_tolerance": 0.5}),
+        ("rt10 tol50%", {"retest_window": 10, "retest_tolerance": 0.5}),
+        ("rt8 tol30%", {"retest_window": 8, "retest_tolerance": 0.3}),
+        ("rt8 tol70%", {"retest_window": 8, "retest_tolerance": 0.7}),
+        ("rt8 tol50%+vol", {"retest_window": 8, "retest_tolerance": 0.5, "vol_filter": True}),
+        ("rt6 tol30%", {"retest_window": 6, "retest_tolerance": 0.3}),
+        ("rt12 tol50%", {"retest_window": 12, "retest_tolerance": 0.5}),
+    ]
+
+    all_results = []
+    for label, params in configs:
+        for asset in ["BTC", "ETH"]:
+            for tf in ["15m", "1h"]:
+                for hold in [3, 6]:
+                    r = strategy.backtest(asset, tf, hold=hold, **params)
+                    if r and r.trades >= 30:
+                        r.strategy_name = f"SB01 {label} h={hold}"
+                        all_results.append(r)
+
+    all_results.sort(key=lambda r: r.accuracy, reverse=True)
+    print(f"\n{'=' * 130}")
+    print(f"  SB01 SQUEEZE BREAKOUT RETEST — TOP 25")
+    print(f"{'=' * 130}")
+    for r in all_results[:25]:
+        r.print_summary()
+    if all_results:
+        all_results[0].print_yearly()
+
+    # Confronto con benchmark
+    print(f"\n  BENCHMARK SQ02: 79.7% acc, 1250 trades, DD 6.5%, 9/9 anni")

scripts/waste/W26_mean_reversion_rsi.py

@@ -0,0 +1,148 @@
+"""MR01 — Mean Reversion da estremi RSI.
+
+Approccio opposto allo squeeze: quando il prezzo va troppo lontano troppo veloce,
+scommetti che torni indietro. Autocorrelazione lag-1 negativa (-0.21 BTC, -0.35 ETH)
+conferma che il mercato a 15m è mean-reverting.
+
+IN:
+    - OHLCV DataFrame
+    - Parametri: rsi_period, rsi_oversold, rsi_overbought, hold_bars,
+                 volume_filter (volume > N× media), atr_filter (move > N×ATR)
+
+OUT:
+    - Signal: long quando RSI < oversold, short quando RSI > overbought
+    - BacktestResult con metriche
+
+Logica:
+    1. RSI scende sotto soglia oversold → LONG (prezzo tornerà su)
+    2. RSI sale sopra soglia overbought → SHORT (prezzo tornerà giù)
+    3. Filtro opzionale: volume spike conferma l'eccesso
+    4. Filtro opzionale: move recente > N×ATR (eccesso di prezzo)
+    5. Hold fisso, poi chiudi
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+from src.strategies.base import Strategy, Signal
+
+
+def rsi(close, period=14):
+    delta = np.diff(close)
+    gain = np.where(delta > 0, delta, 0)
+    loss = np.where(delta < 0, -delta, 0)
+    result = np.full(len(close), 50.0)
+    if len(gain) < period:
+        return result
+    ag = np.mean(gain[:period])
+    al = np.mean(loss[:period])
+    for i in range(period, len(delta)):
+        ag = (ag * (period - 1) + gain[i]) / period
+        al = (al * (period - 1) + loss[i]) / period
+        result[i + 1] = 100 if al == 0 else 100 - 100 / (1 + ag / al)
+    return result
+
+
+class MeanReversionRSI(Strategy):
+    name = "MR01_mean_reversion_rsi"
+    description = "Mean reversion da estremi RSI — fade eccessi direzionali"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m", "1h"]
+    fee_rt = 0.002
+
+    def generate_signals(self, df, ts, **params):
+        c = df["close"].values
+        h = df["high"].values
+        l = df["low"].values
+        v = df["volume"].values
+        n = len(c)
+
+        rsi_period = params.get("rsi_period", 14)
+        oversold = params.get("rsi_oversold", 25)
+        overbought = params.get("rsi_overbought", 75)
+        use_vol_filter = params.get("vol_filter", False)
+        use_atr_filter = params.get("atr_filter", False)
+        cooldown = params.get("cooldown", 4)
+
+        rsi_vals = rsi(c, rsi_period)
+
+        # Volume media rolling
+        vol_ma = np.full(n, np.nan)
+        for i in range(20, n):
+            vol_ma[i] = np.mean(v[i - 20:i])
+
+        # ATR
+        tr = np.maximum(h[1:] - l[1:],
+                        np.maximum(np.abs(h[1:] - c[:-1]), np.abs(l[1:] - c[:-1])))
+        atr_vals = np.full(n, np.nan)
+        for i in range(15, len(tr)):
+            atr_vals[i + 1] = np.mean(tr[i - 14:i])
+
+        signals = []
+        last_signal_idx = -cooldown
+
+        for i in range(20, n):
+            if i - last_signal_idx < cooldown:
+                continue
+
+            direction = 0
+            if rsi_vals[i] < oversold:
+                direction = 1  # oversold → long
+            elif rsi_vals[i] > overbought:
+                direction = -1  # overbought → short
+
+            if direction == 0:
+                continue
+
+            # Volume filter
+            if use_vol_filter and not np.isnan(vol_ma[i]):
+                if v[i] < vol_ma[i] * 1.5:
+                    continue
+
+            # ATR filter: il move recente deve essere > 1.5× ATR
+            if use_atr_filter and not np.isnan(atr_vals[i]):
+                recent_move = abs(c[i] - c[max(0, i - 3)]) / c[max(0, i - 3)]
+                if recent_move < atr_vals[i] / c[i] * 1.5:
+                    continue
+
+            signals.append(Signal(
+                idx=i, direction=direction, entry_price=c[i],
+                metadata={"rsi": float(rsi_vals[i])},
+            ))
+            last_signal_idx = i
+
+        return signals
+
+
+if __name__ == "__main__":
+    strategy = MeanReversionRSI()
+
+    configs = [
+        ("RSI25/75", {}),
+        ("RSI20/80", {"rsi_oversold": 20, "rsi_overbought": 80}),
+        ("RSI25/75+vol", {"vol_filter": True}),
+        ("RSI20/80+vol", {"rsi_oversold": 20, "rsi_overbought": 80, "vol_filter": True}),
+        ("RSI25/75+atr", {"atr_filter": True}),
+        ("RSI20/80+vol+atr", {"rsi_oversold": 20, "rsi_overbought": 80, "vol_filter": True, "atr_filter": True}),
+    ]
+
+    all_results = []
+    for label, params in configs:
+        for asset in ["BTC", "ETH"]:
+            for tf in ["15m", "1h"]:
+                for hold in [3, 6]:
+                    r = strategy.backtest(asset, tf, hold=hold, **params)
+                    if r and r.trades >= 30:
+                        r.strategy_name = f"MR01 {label} h={hold}"
+                        all_results.append(r)
+
+    all_results.sort(key=lambda r: r.accuracy, reverse=True)
+    print(f"\n{'=' * 120}")
+    print(f"  MR01 MEAN REVERSION RSI — TOP 20")
+    print(f"{'=' * 120}")
+    for r in all_results[:20]:
+        r.print_summary()
+    if all_results:
+        all_results[0].print_yearly()

scripts/waste/W27_volume_spike.py

@@ -0,0 +1,133 @@
+"""VO01 — Volume Spike Reversal.
+
+Quando il volume esplode (>3× media) con un forte move direzionale,
+il mercato è in eccesso → fade il move (mean reversion).
+
+Diverso dallo squeeze: non cerca compressione, cerca ECCESSO.
+Il volume spike indica panico/euforia → reversal probabile.
+
+IN:
+    - OHLCV DataFrame
+    - Parametri: vol_mult (3), move_threshold (0.005), hold
+
+OUT:
+    - Signal: fade la direzione del volume spike
+    - BacktestResult
+
+Logica:
+    1. Volume > vol_mult × media 20 periodi
+    2. Move nella candela > move_threshold (0.5%)
+    3. Direzione: opposta al move (mean reversion)
+    4. Filtro: non entrare se già in trend forte (EMA slope)
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+from src.strategies.base import Strategy, Signal
+
+
+class VolumeSpikeReversal(Strategy):
+    name = "VO01_vol_spike_reversal"
+    description = "Volume spike reversal — fade eccessi di volume/prezzo"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m", "1h"]
+    fee_rt = 0.002
+
+    def generate_signals(self, df, ts, **params):
+        c = df["close"].values
+        o = df["open"].values
+        h = df["high"].values
+        l = df["low"].values
+        v = df["volume"].values
+        n = len(c)
+
+        vol_mult = params.get("vol_mult", 3.0)
+        move_thr = params.get("move_threshold", 0.005)
+        use_trend_filter = params.get("trend_filter", False)
+        cooldown = params.get("cooldown", 4)
+
+        # Volume media rolling
+        vol_ma = np.full(n, np.nan)
+        for i in range(20, n):
+            vol_ma[i] = np.mean(v[i - 20:i])
+
+        # EMA per trend filter
+        ema_20 = np.full(n, np.nan)
+        k = 2 / 21
+        ema_20[19] = np.mean(c[:20])
+        for i in range(20, n):
+            ema_20[i] = c[i] * k + ema_20[i - 1] * (1 - k)
+
+        signals = []
+        last_idx = -cooldown
+
+        for i in range(21, n):
+            if i - last_idx < cooldown:
+                continue
+            if np.isnan(vol_ma[i]):
+                continue
+
+            # Volume spike
+            if v[i] < vol_ma[i] * vol_mult:
+                continue
+
+            # Price move
+            move = (c[i] - o[i]) / o[i] if o[i] > 0 else 0
+            if abs(move) < move_thr:
+                continue
+
+            # Fade: opposto al move
+            direction = -1 if move > 0 else 1
+
+            # Trend filter: non fare mean reversion contro trend forte
+            if use_trend_filter and not np.isnan(ema_20[i]):
+                ema_slope = (ema_20[i] - ema_20[max(0, i - 5)]) / ema_20[max(0, i - 5)]
+                if direction == -1 and ema_slope > 0.005:
+                    continue
+                if direction == 1 and ema_slope < -0.005:
+                    continue
+
+            signals.append(Signal(
+                idx=i, direction=direction, entry_price=c[i],
+                metadata={"vol_ratio": float(v[i] / vol_ma[i]), "move_pct": round(move * 100, 3)},
+            ))
+            last_idx = i
+
+        return signals
+
+
+if __name__ == "__main__":
+    strategy = VolumeSpikeReversal()
+
+    configs = [
+        ("v3x m0.5%", {"vol_mult": 3.0, "move_threshold": 0.005}),
+        ("v3x m1%", {"vol_mult": 3.0, "move_threshold": 0.01}),
+        ("v4x m0.5%", {"vol_mult": 4.0, "move_threshold": 0.005}),
+        ("v4x m1%", {"vol_mult": 4.0, "move_threshold": 0.01}),
+        ("v3x m0.5%+tf", {"vol_mult": 3.0, "move_threshold": 0.005, "trend_filter": True}),
+        ("v3x m1%+tf", {"vol_mult": 3.0, "move_threshold": 0.01, "trend_filter": True}),
+        ("v5x m1%", {"vol_mult": 5.0, "move_threshold": 0.01}),
+        ("v5x m1%+tf", {"vol_mult": 5.0, "move_threshold": 0.01, "trend_filter": True}),
+    ]
+
+    all_results = []
+    for label, params in configs:
+        for asset in ["BTC", "ETH"]:
+            for tf in ["15m", "1h"]:
+                for hold in [3, 6]:
+                    r = strategy.backtest(asset, tf, hold=hold, **params)
+                    if r and r.trades >= 30:
+                        r.strategy_name = f"VO01 {label} h={hold}"
+                        all_results.append(r)
+
+    all_results.sort(key=lambda r: r.accuracy, reverse=True)
+    print(f"\n{'=' * 120}")
+    print(f"  VO01 VOLUME SPIKE REVERSAL — TOP 20")
+    print(f"{'=' * 120}")
+    for r in all_results[:20]:
+        r.print_summary()
+    if all_results:
+        all_results[0].print_yearly()

scripts/waste/W28_squeeze_mr.py

@@ -0,0 +1,169 @@
+"""HY01 — Squeeze + Mean Reversion Ibrida.
+
+Insight: durante lo squeeze (bassa volatilità), il prezzo mean-reverte
+DENTRO il range compresso. Autocorrelazione negativa a 15m conferma.
+Invece di aspettare il BREAKOUT, tradi la MEAN REVERSION dentro lo squeeze.
+
+Completamente diverso da SQ01-SQ04 che aspettano il RILASCIO.
+
+IN:
+    - OHLCV DataFrame
+    - Parametri: bb_window, sq_threshold, rsi_period, rsi_levels,
+                 vol_filter, bb_touch (prezzo tocca banda Bollinger)
+
+OUT:
+    - Signal: long quando RSI oversold DURANTE squeeze, short quando overbought
+    - BacktestResult
+
+Logica:
+    1. Verifica che siamo IN squeeze (BB dentro KC)
+    2. Prezzo tocca banda inferiore BB → LONG (tornerà alla media)
+    3. Prezzo tocca banda superiore BB → SHORT (tornerà alla media)
+    4. Conferma RSI: deve essere estremo nella direzione
+    5. Hold corto (2-3 barre) — target: ritorno alla media
+    6. Stop: se prezzo rompe lo squeeze → chiudi subito
+"""
+from __future__ import annotations
+import sys
+sys.path.insert(0, ".")
+
+import numpy as np
+import pandas as pd
+from src.strategies.base import Strategy, Signal
+from src.strategies.indicators import keltner_ratio
+
+
+def rsi(close, period=14):
+    delta = np.diff(close)
+    gain = np.where(delta > 0, delta, 0)
+    loss = np.where(delta < 0, -delta, 0)
+    result = np.full(len(close), 50.0)
+    if len(gain) < period:
+        return result
+    ag = np.mean(gain[:period])
+    al = np.mean(loss[:period])
+    for i in range(period, len(delta)):
+        ag = (ag * (period - 1) + gain[i]) / period
+        al = (al * (period - 1) + loss[i]) / period
+        result[i + 1] = 100 if al == 0 else 100 - 100 / (1 + ag / al)
+    return result
+
+
+def bollinger(close, window=14):
+    n = len(close)
+    upper = np.full(n, np.nan)
+    lower = np.full(n, np.nan)
+    mid = np.full(n, np.nan)
+    for i in range(window, n):
+        wc = close[i - window:i]
+        m = np.mean(wc)
+        s = np.std(wc)
+        mid[i] = m
+        upper[i] = m + 2 * s
+        lower[i] = m - 2 * s
+    return upper, mid, lower
+
+
+class SqueezeMeanReversion(Strategy):
+    name = "HY01_squeeze_mr"
+    description = "Mean reversion DENTRO lo squeeze — fade estremi in range compresso"
+    default_assets = ["BTC", "ETH"]
+    default_timeframes = ["15m", "1h"]
+    fee_rt = 0.002
+
+    def generate_signals(self, df, ts, **params):
+        c = df["close"].values
+        h = df["high"].values
+        l = df["low"].values
+        v = df["volume"].values
+        n = len(c)
+
+        bb_w = params.get("bb_window", 14)
+        sq_thr = params.get("sq_threshold", 0.8)
+        rsi_period = params.get("rsi_period", 14)
+        rsi_low = params.get("rsi_oversold", 30)
+        rsi_high = params.get("rsi_overbought", 70)
+        use_bb_touch = params.get("bb_touch", True)
+        cooldown = params.get("cooldown", 3)
+
+        kcr = keltner_ratio(c, h, l, bb_w)
+        rsi_vals = rsi(c, rsi_period)
+        bb_upper, bb_mid, bb_lower = bollinger(c, bb_w)
+
+        signals = []
+        last_idx = -cooldown
+
+        for i in range(bb_w + 1, n):
+            if i - last_idx < cooldown:
+                continue
+            if np.isnan(kcr[i]) or np.isnan(bb_lower[i]):
+                continue
+
+            # Must be IN squeeze
+            if kcr[i] >= sq_thr:
+                continue
+
+            direction = 0
+
+            if use_bb_touch:
+                # Prezzo tocca/rompe BB lower → long (mean reversion up)
+                if c[i] <= bb_lower[i] and rsi_vals[i] < rsi_low:
+                    direction = 1
+                # Prezzo tocca/rompe BB upper → short (mean reversion down)
+                elif c[i] >= bb_upper[i] and rsi_vals[i] > rsi_high:
+                    direction = -1
+            else:
+                # Solo RSI
+                if rsi_vals[i] < rsi_low:
+                    direction = 1
+                elif rsi_vals[i] > rsi_high:
+                    direction = -1
+
+            if direction == 0:
+                continue
+
+            signals.append(Signal(
+                idx=i, direction=direction, entry_price=c[i],
+                metadata={
+                    "rsi": float(rsi_vals[i]),
+                    "kcr": float(kcr[i]),
+                    "bb_pos": "lower" if direction == 1 else "upper",
+                },
+            ))
+            last_idx = i
+
+        return signals
+
+
+if __name__ == "__main__":
+    strategy = SqueezeMeanReversion()
+
+    configs = [
+        ("bb+rsi30/70", {"bb_touch": True, "rsi_oversold": 30, "rsi_overbought": 70}),
+        ("bb+rsi25/75", {"bb_touch": True, "rsi_oversold": 25, "rsi_overbought": 75}),
+        ("bb+rsi35/65", {"bb_touch": True, "rsi_oversold": 35, "rsi_overbought": 65}),
+        ("rsi30/70 only", {"bb_touch": False, "rsi_oversold": 30, "rsi_overbought": 70}),
+        ("rsi25/75 only", {"bb_touch": False, "rsi_oversold": 25, "rsi_overbought": 75}),
+        ("sq<0.7 bb+rsi30", {"bb_touch": True, "sq_threshold": 0.7, "rsi_oversold": 30, "rsi_overbought": 70}),
+        ("sq<0.9 bb+rsi30", {"bb_touch": True, "sq_threshold": 0.9, "rsi_oversold": 30, "rsi_overbought": 70}),
+        ("sq<0.9 rsi35/65", {"bb_touch": False, "sq_threshold": 0.9, "rsi_oversold": 35, "rsi_overbought": 65}),
+    ]
+
+    all_results = []
+    for label, params in configs:
+        for asset in ["BTC", "ETH"]:
+            for tf in ["15m", "1h"]:
+                for hold in [2, 3, 4]:
+                    r = strategy.backtest(asset, tf, hold=hold, **params)
+                    if r and r.trades >= 30:
+                        r.strategy_name = f"HY01 {label} h={hold}"
+                        all_results.append(r)
+
+    all_results.sort(key=lambda r: r.accuracy, reverse=True)
+    print(f"\n{'=' * 130}")
+    print(f"  HY01 SQUEEZE MEAN REVERSION — TOP 25")
+    print(f"{'=' * 130}")
+    for r in all_results[:25]:
+        r.print_summary()
+    if all_results:
+        all_results[0].print_yearly()

src/live/multi_runner.py

@@ -0,0 +1,291 @@
+"""Multi-Strategy Paper Trader — orchestratore per N strategie in parallelo."""
+from __future__ import annotations
+
+import time
+import yaml
+from datetime import datetime, timedelta, timezone
+from pathlib import Path
+
+import pandas as pd
+
+from src.live.cerbero_client import CerberoClient
+from src.live.strategy_loader import load_strategy
+from src.live.strategy_worker import StrategyWorker
+from src.live.signal_engine import SignalEngine
+from src.live.telegram_notifier import send_telegram
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+DATA_DIR = PROJECT_ROOT / "data" / "paper_trades"
+
+RESOLUTION_MAP = {"15m": "15", "1h": "60", "5m": "5"}
+INSTRUMENT_MAP = {
+    "BTC": "BTC-PERPETUAL",
+    "ETH": "ETH-PERPETUAL",
+}
+
+
+class MLWorkerWrapper:
+    """Wrapper speciale per ML01 che usa SignalEngine con training."""
+
+    def __init__(self, worker: StrategyWorker, config: dict):
+        self.worker = worker
+        self.engine = SignalEngine(
+            bb_w=config.get("params", {}).get("bb_window", 14),
+            sq_thr=config.get("params", {}).get("sq_threshold", 0.8),
+            ml_thr=config.get("params", {}).get("ml_threshold", 0.70),
+        )
+        self.trained = False
+        self.last_train: datetime | None = None
+        self.retrain_hours = config.get("retrain_hours", 24)
+
+    def needs_training(self) -> bool:
+        if not self.trained:
+            return True
+        if self.last_train is None:
+            return True
+        elapsed = (datetime.now(timezone.utc) - self.last_train).total_seconds()
+        return elapsed > self.retrain_hours * 3600
+
+    def train(self, df: pd.DataFrame, hold: int = 3):
+        result = self.engine.train(df, lookahead=hold)
+        if "error" not in result:
+            self.trained = True
+            self.last_train = datetime.now(timezone.utc)
+            print(f"  [{self.worker.worker_id}] TRAIN OK: {result}")
+        else:
+            print(f"  [{self.worker.worker_id}] TRAIN FAIL: {result}")
+
+    def tick(self, df: pd.DataFrame):
+        if not self.trained:
+            return
+
+        worker = self.worker
+        c = df["close"].values
+        current_price = float(c[-1])
+        current_ts = int(df["timestamp"].iloc[-1])
+
+        if worker.in_position:
+            if current_ts > worker.last_bar_ts:
+                worker.bars_held += 1
+                worker.last_bar_ts = current_ts
+            if worker.bars_held >= worker.hold_bars:
+                worker._close_position(current_price, "hold_limit")
+            else:
+                pnl_pct = (current_price - worker.entry_price) / worker.entry_price * worker.direction
+                if pnl_pct <= -0.02:
+                    worker._close_position(current_price, "stop_loss")
+            worker._save_state()
+            return
+
+        signal = self.engine.check_signal(df)
+        if signal:
+            from src.strategies.base import Signal
+            direction = 1 if signal["direction"] == "buy" else -1
+            sig = Signal(idx=len(df)-1, direction=direction, entry_price=current_price)
+            worker._open_position(sig, current_price)
+            worker.last_bar_ts = current_ts
+
+        worker._save_state()
+
+
+def load_config(path: Path) -> dict:
+    with open(path) as f:
+        return yaml.safe_load(f)
+
+
+def build_workers(config: dict) -> tuple[list[StrategyWorker], list[MLWorkerWrapper]]:
+    """Crea worker da config YAML."""
+    defaults = config.get("defaults", {})
+    regular_workers: list[StrategyWorker] = []
+    ml_workers: list[MLWorkerWrapper] = []
+
+    for entry in config.get("strategies", []):
+        if not entry.get("enabled", True):
+            continue
+
+        name = entry["name"]
+        asset = entry["asset"]
+        tf = entry["tf"]
+        capital = entry.get("capital", defaults.get("capital", 1000))
+        pos_size = entry.get("position_size", defaults.get("position_size", 0.15))
+        leverage = entry.get("leverage", defaults.get("leverage", 3))
+        hold = entry.get("hold_bars", defaults.get("hold_bars", 3))
+        params = entry.get("params", {})
+
+        strategy = load_strategy(name)
+
+        worker = StrategyWorker(
+            strategy=strategy, asset=asset, tf=tf,
+            capital=capital, position_size=pos_size,
+            leverage=leverage, hold_bars=hold,
+            params=params, data_dir=DATA_DIR,
+        )
+
+        if name == "ML01_squeeze_gbm":
+            ml_wrapper = MLWorkerWrapper(worker, {**defaults, **entry})
+            ml_workers.append(ml_wrapper)
+        else:
+            regular_workers.append(worker)
+
+    return regular_workers, ml_workers
+
+
+def run():
+    config_path = PROJECT_ROOT / "strategies.yml"
+    if not config_path.exists():
+        print(f"ERRORE: {config_path} non trovato")
+        return
+
+    config = load_config(config_path)
+    defaults = config.get("defaults", {})
+    poll_seconds = defaults.get("poll_seconds", 60)
+    lookback_days = 60
+    train_lookback_days = 365
+
+    regular_workers, ml_workers = build_workers(config)
+    all_worker_count = len(regular_workers) + len(ml_workers)
+
+    if all_worker_count == 0:
+        print("Nessuna strategia abilitata in strategies.yml")
+        return
+
+    client = CerberoClient()
+
+    print("=" * 70)
+    print(f"  MULTI-STRATEGY PAPER TRADER")
+    print(f"  Strategie attive: {all_worker_count}")
+    print(f"  Poll: ogni {poll_seconds}s")
+    print(f"  Data dir: {DATA_DIR}")
+    print("=" * 70)
+
+    for w in regular_workers:
+        print(f"  • {w.status_summary}")
+    for mw in ml_workers:
+        print(f"  • {mw.worker.status_summary} [ML]")
+
+    send_telegram(f"🚀 Multi-Strategy avviato: {all_worker_count} strategie")
+
+    # Raccogli asset/tf unici per fetch raggruppato
+    def _get_data_keys() -> set[tuple[str, str]]:
+        keys = set()
+        for w in regular_workers:
+            keys.add((w.asset, w.tf))
+        for mw in ml_workers:
+            keys.add((mw.worker.asset, mw.worker.tf))
+        return keys
+
+    # Training iniziale ML
+    for mw in ml_workers:
+        asset = mw.worker.asset
+        instrument = INSTRUMENT_MAP.get(asset, f"{asset}-PERPETUAL")
+        resolution = RESOLUTION_MAP.get(mw.worker.tf, "15")
+        end = datetime.now(timezone.utc)
+        start = end - timedelta(days=train_lookback_days)
+        candles = client.get_historical(instrument, start.strftime("%Y-%m-%d"),
+                                         end.strftime("%Y-%m-%d"), resolution)
+        if candles:
+            df_train = pd.DataFrame(candles)
+            df_train["timestamp"] = df_train["timestamp"].astype("int64")
+            df_train = df_train.sort_values("timestamp").reset_index(drop=True)
+            mw.train(df_train, hold=mw.worker.hold_bars)
+
+    while True:
+        try:
+            data_keys = _get_data_keys()
+            candle_cache: dict[tuple[str, str], pd.DataFrame] = {}
+
+            for asset, tf in data_keys:
+                instrument = INSTRUMENT_MAP.get(asset, f"{asset}-PERPETUAL")
+                resolution = RESOLUTION_MAP.get(tf, "15")
+                end = datetime.now(timezone.utc)
+                start = end - timedelta(days=lookback_days)
+
+                candles = client.get_historical(
+                    instrument, start.strftime("%Y-%m-%d"),
+                    end.strftime("%Y-%m-%d"), resolution,
+                )
+                if candles:
+                    df = pd.DataFrame(candles)
+                    df["timestamp"] = df["timestamp"].astype("int64")
+                    df = df.sort_values("timestamp").reset_index(drop=True)
+                    candle_cache[(asset, tf)] = df
+
+            # Fetch 1h live per strategie multi-timeframe (es. MT01):
+            # il trend va preso da Cerbero, non dal parquet statico (che resta indietro).
+            htf_cache: dict[str, pd.DataFrame] = {}
+            mtf_assets = {w.asset for w in regular_workers if w.strategy.name.startswith("MT01")}
+            for asset in mtf_assets:
+                instrument = INSTRUMENT_MAP.get(asset, f"{asset}-PERPETUAL")
+                end = datetime.now(timezone.utc)
+                start = end - timedelta(days=lookback_days)
+                try:
+                    candles_1h = client.get_historical(
+                        instrument, start.strftime("%Y-%m-%d"),
+                        end.strftime("%Y-%m-%d"), "60",
+                    )
+                    if candles_1h:
+                        df1h = pd.DataFrame(candles_1h)
+                        df1h["timestamp"] = df1h["timestamp"].astype("int64")
+                        htf_cache[asset] = df1h.sort_values("timestamp").reset_index(drop=True)
+                except Exception as e:
+                    print(f"  [1h fetch {asset}] ERRORE: {e}")
+
+            # Tick regular workers
+            for w in regular_workers:
+                key = (w.asset, w.tf)
+                if key in candle_cache:
+                    try:
+                        w.tick(candle_cache[key], df_1h=htf_cache.get(w.asset))
+                    except Exception as e:
+                        print(f"  [{w.worker_id}] ERRORE: {e}")
+
+            # Tick ML workers
+            for mw in ml_workers:
+                key = (mw.worker.asset, mw.worker.tf)
+                if key not in candle_cache:
+                    continue
+
+                if mw.needs_training():
+                    mw.train(candle_cache[key], hold=mw.worker.hold_bars)
+
+                try:
+                    mw.tick(candle_cache[key])
+                except Exception as e:
+                    print(f"  [{mw.worker.worker_id}] ERRORE: {e}")
+
+            # Status periodico
+            now = datetime.now(timezone.utc)
+            if now.minute == 0 and now.second < poll_seconds:
+                lines = [f"📊 Status {now.strftime('%H:%M')} UTC"]
+                for w in regular_workers:
+                    lines.append(f"  {w.status_summary}")
+                for mw in ml_workers:
+                    lines.append(f"  {mw.worker.status_summary} [ML]")
+                send_telegram("\n".join(lines))
+
+        except KeyboardInterrupt:
+            print("\nShutdown...")
+            for w in regular_workers:
+                if w.in_position:
+                    df = candle_cache.get((w.asset, w.tf))
+                    if df is not None and not df.empty:
+                        w._close_position(float(df["close"].iloc[-1]), "shutdown")
+                w._save_state()
+            for mw in ml_workers:
+                if mw.worker.in_position:
+                    df = candle_cache.get((mw.worker.asset, mw.worker.tf))
+                    if df is not None and not df.empty:
+                        mw.worker._close_position(float(df["close"].iloc[-1]), "shutdown")
+                mw.worker._save_state()
+            send_telegram("🛑 Multi-Strategy arrestato")
+            break
+        except Exception as e:
+            print(f"  ERRORE GLOBALE: {e}")
+            import traceback
+            traceback.print_exc()
+
+        time.sleep(poll_seconds)
+
+
+if __name__ == "__main__":
+    run()

src/live/paper_trader.py

@@ -10,6 +10,7 @@ import pandas as pd
 
 from src.live.cerbero_client import CerberoClient
 from src.live.signal_engine import SignalEngine
+from src.live.telegram_notifier import notify_event
 
 LOG_DIR = Path(__file__).resolve().parents[2] / "data" / "paper_trades"
 INSTRUMENT = "ETH_USDC-PERPETUAL"
@@ -52,6 +53,7 @@ class PaperTrader:
         with open(self.log_path, "a") as f:
             f.write(json.dumps(entry) + "\n")
         print(f"  [{entry['timestamp'][:19]}] {event}: {json.dumps(data or {})}")
+        notify_event(event, data)
 
     def save_status(self):
         status = {

src/live/signal_engine.py

@@ -112,6 +112,54 @@ class SignalEngine:
         self.squeeze_start_idx = 0
         self.trained = False
 
+    def _new_model(self) -> GradientBoostingClassifier:
+        return GradientBoostingClassifier(
+            n_estimators=150, max_depth=4, min_samples_leaf=10,
+            learning_rate=0.05, subsample=0.8, random_state=42,
+        )
+
+    def _validate_oos(self, X: np.ndarray, y: np.ndarray, test_frac: float = 0.2) -> dict:
+        """Split temporale (no shuffle) per stimare la performance out-of-sample.
+
+        Allena su training iniziale e valuta sull'ultimo `test_frac` dei campioni.
+        Oltre all'accuratezza OOS, riporta la precisione sui soli segnali con
+        confidenza >= ml_thr — cioè i trade che la strategia aprirebbe davvero.
+        """
+        n_test = int(len(X) * test_frac)
+        n_train = len(X) - n_test
+        if n_train < 30 or n_test < 5:
+            return {"oos_warning": "test set troppo piccolo", "oos_test_samples": n_test}
+
+        scaler = StandardScaler()
+        X_tr = scaler.fit_transform(X[:n_train])
+        X_te = scaler.transform(X[n_train:])
+        y_tr, y_te = y[:n_train], y[n_train:]
+
+        model = self._new_model()
+        model.fit(X_tr, y_tr)
+
+        up_idx = list(model.classes_).index(1)
+        p_up = model.predict_proba(X_te)[:, up_idx]
+        test_acc = float(np.mean((p_up >= 0.5).astype(int) == y_te) * 100)
+        oos_train_acc = float(np.mean(model.predict(X_tr) == y_tr) * 100)
+
+        long_sig = p_up >= self.ml_thr
+        short_sig = p_up <= (1 - self.ml_thr)
+        n_sig = int((long_sig | short_sig).sum())
+        if n_sig > 0:
+            correct = int(((long_sig & (y_te == 1)) | (short_sig & (y_te == 0))).sum())
+            sig_prec = round(correct / n_sig * 100, 1)
+        else:
+            sig_prec = None
+
+        return {
+            "oos_train_accuracy": round(oos_train_acc, 1),
+            "oos_test_accuracy": round(test_acc, 1),
+            "oos_test_samples": n_test,
+            "oos_signals": n_sig,
+            "oos_signal_precision": sig_prec,
+        }
+
     def train(self, df: pd.DataFrame, lookahead: int = 3) -> dict:
         """Addestra il modello su dati storici."""
         close = df["close"].values
@@ -154,20 +202,24 @@ class SignalEngine:
         X = np.array(X_all)
         y = np.array(y_all)
 
+        oos = self._validate_oos(X, y)
+
         self.scaler = StandardScaler()
         X_s = self.scaler.fit_transform(X)
 
-        self.model = GradientBoostingClassifier(
-            n_estimators=150, max_depth=4, min_samples_leaf=10,
-            learning_rate=0.05, subsample=0.8, random_state=42,
-        )
+        self.model = self._new_model()
         self.model.fit(X_s, y)
         self.trained = True
 
         preds = self.model.predict(X_s)
-        train_acc = np.mean(preds == y) * 100
+        train_acc = float(np.mean(preds == y) * 100)
 
-        return {"samples": len(X), "up_ratio": np.mean(y) * 100, "train_accuracy": train_acc}
+        return {
+            "samples": len(X),
+            "up_ratio": round(float(np.mean(y) * 100), 1),
+            "train_accuracy": round(train_acc, 1),
+            **oos,
+        }
 
     def check_signal(self, df: pd.DataFrame) -> dict | None:
         """Controlla se c'è un segnale sulle ultime candele.

src/live/strategy_loader.py

@@ -0,0 +1,53 @@
+"""Import dinamico delle classi Strategy da scripts/strategies/."""
+from __future__ import annotations
+
+import importlib
+import sys
+from pathlib import Path
+
+from src.strategies.base import Strategy
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+STRATEGIES_DIR = PROJECT_ROOT / "scripts" / "strategies"
+
+_REGISTRY: dict[str, type[Strategy]] = {}
+
+# Solo strategie con edge netto validato out-of-sample (fee-aware).
+# La famiglia squeeze-breakout (SQ/MT/ML/AD/CM/PD) e' stata spostata in
+# scripts/waste/: l'edge storico era un artefatto di look-ahead
+# (vedi scripts/analysis/oos_validation.py).
+MODULE_MAP = {
+    "MR01_bollinger_fade": ("MR01_bollinger_fade", "BollingerFade"),
+    "MR02_donchian_fade": ("MR02_donchian_fade", "DonchianFade"),
+    "MR07_return_reversal": ("MR07_return_reversal", "ReturnReversal"),
+}
+
+
+def load_strategy(name: str) -> Strategy:
+    """Carica e istanzia una Strategy per nome."""
+    if name in _REGISTRY:
+        return _REGISTRY[name]()
+
+    if name not in MODULE_MAP:
+        raise ValueError(f"Strategia sconosciuta: {name}. Disponibili: {list(MODULE_MAP)}")
+
+    module_file, class_name = MODULE_MAP[name]
+    module_path = STRATEGIES_DIR / f"{module_file}.py"
+
+    if not module_path.exists():
+        raise FileNotFoundError(f"File strategia non trovato: {module_path}")
+
+    if str(PROJECT_ROOT) not in sys.path:
+        sys.path.insert(0, str(PROJECT_ROOT))
+
+    spec = importlib.util.spec_from_file_location(f"strategies.{module_file}", module_path)
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+
+    cls = getattr(module, class_name)
+    _REGISTRY[name] = cls
+    return cls()
+
+
+def list_available() -> list[str]:
+    return list(MODULE_MAP.keys())

src/live/strategy_worker.py

@@ -0,0 +1,273 @@
+"""Worker per singola strategia — paper trading con stato persistente."""
+from __future__ import annotations
+
+import json
+from datetime import datetime, timezone
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+from src.strategies.base import Strategy, Signal
+from src.live.telegram_notifier import notify_event
+
+FEE_RT = 0.002
+
+
+class StrategyWorker:
+    """Gestisce paper trading per una singola strategia/asset/tf."""
+
+    def __init__(
+        self,
+        strategy: Strategy,
+        asset: str,
+        tf: str,
+        capital: float = 1000.0,
+        position_size: float = 0.15,
+        leverage: float = 3.0,
+        hold_bars: int = 3,
+        params: dict | None = None,
+        data_dir: Path = Path("data/paper_trades"),
+    ):
+        self.strategy = strategy
+        self.asset = asset
+        self.tf = tf
+        self.initial_capital = capital
+        self.position_size = position_size
+        self.leverage = leverage
+        self.hold_bars = hold_bars
+        self.params = params or {}
+
+        self.worker_id = f"{strategy.name}__{asset}__{tf}"
+        self.work_dir = data_dir / self.worker_id
+        self.work_dir.mkdir(parents=True, exist_ok=True)
+        self.trades_path = self.work_dir / "trades.jsonl"
+        self.status_path = self.work_dir / "status.json"
+
+        self.capital = capital
+        self.in_position = False
+        self.direction: int = 0
+        self.entry_price: float = 0
+        self.entry_time: str = ""
+        self.bars_held: int = 0
+        self.total_trades: int = 0
+        self.total_wins: int = 0
+        self.started_at = datetime.now(timezone.utc).isoformat()
+        self.last_bar_ts: int = 0
+
+        # Exit guidati dalla strategia via Signal.metadata (0 = usa hold_bars/stop legacy)
+        self.tp: float = 0.0
+        self.sl: float = 0.0
+        self.max_bars: int = 0
+        # Fee dalla strategia (MR01 = 0.001 realistico Deribit), fallback al default modulo
+        self.fee_rt: float = float(getattr(strategy, "fee_rt", FEE_RT))
+
+        self._load_state()
+        self._save_state()
+
+    def _load_state(self):
+        """Riprende stato da status.json se esiste."""
+        if not self.status_path.exists():
+            self._log("INIT", {"capital": self.capital, "strategy": self.strategy.name,
+                               "asset": self.asset, "tf": self.tf})
+            return
+
+        with open(self.status_path) as f:
+            state = json.load(f)
+
+        self.capital = state.get("capital", self.initial_capital)
+        self.in_position = state.get("in_position", False)
+        self.direction = state.get("direction", 0)
+        self.entry_price = state.get("entry_price", 0)
+        self.entry_time = state.get("entry_time", "")
+        self.bars_held = state.get("bars_held", 0)
+        self.total_trades = state.get("total_trades", 0)
+        self.total_wins = state.get("total_wins", 0)
+        self.started_at = state.get("started_at", self.started_at)
+        self.last_bar_ts = state.get("last_bar_ts", 0)
+        self.tp = state.get("tp", 0.0)
+        self.sl = state.get("sl", 0.0)
+        self.max_bars = state.get("max_bars", 0)
+
+        self._log("RESUME", {"capital": round(self.capital, 2),
+                              "total_trades": self.total_trades,
+                              "in_position": self.in_position})
+
+    def _save_state(self):
+        state = {
+            "capital": round(self.capital, 2),
+            "in_position": self.in_position,
+            "direction": self.direction,
+            "entry_price": self.entry_price,
+            "entry_time": self.entry_time,
+            "bars_held": self.bars_held,
+            "total_trades": self.total_trades,
+            "total_wins": self.total_wins,
+            "started_at": self.started_at,
+            "last_bar_ts": self.last_bar_ts,
+            "tp": self.tp,
+            "sl": self.sl,
+            "max_bars": self.max_bars,
+            "last_update": datetime.now(timezone.utc).isoformat(),
+        }
+        with open(self.status_path, "w") as f:
+            json.dump(state, f, indent=2)
+
+    def _log(self, event: str, data: dict | None = None):
+        entry = {
+            "ts": datetime.now(timezone.utc).isoformat(),
+            "worker": self.worker_id,
+            "event": event,
+            **(data or {}),
+        }
+        with open(self.trades_path, "a") as f:
+            f.write(json.dumps(entry) + "\n")
+        print(f"  [{self.worker_id}] {event}: {json.dumps(data or {}, default=str)}")
+
+    def _notify(self, event: str, data: dict | None = None):
+        enriched = {"worker": self.worker_id, **(data or {})}
+        notify_event(event, enriched)
+
+    def _open_position(self, signal: Signal, current_price: float):
+        notional = self.capital * self.position_size * self.leverage
+        size = notional / current_price if current_price > 0 else 0
+
+        self.in_position = True
+        self.direction = signal.direction
+        self.entry_price = current_price
+        self.entry_time = datetime.now(timezone.utc).isoformat()
+        self.bars_held = 0
+
+        meta = signal.metadata or {}
+        self.tp = float(meta.get("tp", 0.0) or 0.0)
+        self.sl = float(meta.get("sl", 0.0) or 0.0)
+        self.max_bars = int(meta.get("max_bars", 0) or 0)
+
+        trade_data = {
+            "direction": "long" if signal.direction == 1 else "short",
+            "price": round(current_price, 2),
+            "size": round(size, 6),
+            "notional": round(notional, 2),
+            "capital": round(self.capital, 2),
+            "tp": round(self.tp, 2) if self.tp else None,
+            "sl": round(self.sl, 2) if self.sl else None,
+        }
+        self._log("OPEN", trade_data)
+        self._notify("OPENED", trade_data)
+
+    def _close_position(self, current_price: float, reason: str):
+        if not self.in_position:
+            return
+
+        price_change = (current_price - self.entry_price) / self.entry_price
+        trade_return = price_change * self.direction
+        net = trade_return * self.leverage - self.fee_rt * self.leverage
+        pnl = self.capital * self.position_size * net
+
+        is_win = trade_return > 0
+        self.capital += pnl
+        self.capital = max(self.capital, 0)
+        self.total_trades += 1
+        if is_win:
+            self.total_wins += 1
+
+        accuracy = self.total_wins / self.total_trades * 100 if self.total_trades > 0 else 0
+
+        trade_data = {
+            "reason": reason,
+            "direction": "long" if self.direction == 1 else "short",
+            "entry": round(self.entry_price, 2),
+            "exit": round(current_price, 2),
+            "pnl": round(pnl, 2),
+            "net_return": round(net * 100, 3),
+            "capital": round(self.capital, 2),
+            "bars_held": self.bars_held,
+            "win": is_win,
+            "total_trades": self.total_trades,
+            "accuracy": round(accuracy, 1),
+        }
+        self._log("CLOSE", trade_data)
+        self._notify("CLOSED", trade_data)
+
+        self.in_position = False
+        self.direction = 0
+        self.entry_price = 0
+        self.entry_time = ""
+        self.bars_held = 0
+        self.tp = 0.0
+        self.sl = 0.0
+        self.max_bars = 0
+
+    def tick(self, df: pd.DataFrame, df_1h: pd.DataFrame | None = None):
+        """Chiamato ad ogni poll con DataFrame OHLCV aggiornato.
+
+        df_1h: serie 1h live opzionale per strategie multi-timeframe (es. MT01),
+        passata ai generate_signals via params. Se None la strategia ricade sul
+        parquet statico.
+        """
+        if df.empty or len(df) < 100:
+            return
+
+        c = df["close"].values
+        current_price = float(c[-1])
+        current_ts = int(df["timestamp"].iloc[-1])
+        ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+
+        if self.in_position:
+            if current_ts > self.last_bar_ts:
+                self.bars_held += 1
+                self.last_bar_ts = current_ts
+
+            if self.tp and self.sl:
+                # Exit guidati dalla strategia: SL (conservativo, prima), poi TP, poi time-limit
+                if self.direction == 1:
+                    if current_price <= self.sl:
+                        self._close_position(current_price, "stop_loss")
+                    elif current_price >= self.tp:
+                        self._close_position(current_price, "take_profit")
+                    elif self.max_bars and self.bars_held >= self.max_bars:
+                        self._close_position(current_price, "time_limit")
+                else:
+                    if current_price >= self.sl:
+                        self._close_position(current_price, "stop_loss")
+                    elif current_price <= self.tp:
+                        self._close_position(current_price, "take_profit")
+                    elif self.max_bars and self.bars_held >= self.max_bars:
+                        self._close_position(current_price, "time_limit")
+            elif self.bars_held >= self.hold_bars:
+                self._close_position(current_price, "hold_limit")
+            else:
+                pnl_pct = (current_price - self.entry_price) / self.entry_price * self.direction
+                if pnl_pct <= -0.02:
+                    self._close_position(current_price, "stop_loss")
+
+            self._save_state()
+            return
+
+        # Genera segnali
+        extra = dict(self.params)
+        if df_1h is not None:
+            extra["df_1h"] = df_1h
+        signals = self.strategy.generate_signals(
+            df, ts, asset=self.asset, tf=self.tf, **extra
+        )
+
+        if not signals:
+            self._save_state()
+            return
+
+        last_signal = signals[-1]
+        last_idx = len(df) - 1
+
+        if last_signal.idx >= last_idx - 1:
+            self._open_position(last_signal, current_price)
+            self.last_bar_ts = current_ts
+
+        self._save_state()
+
+    @property
+    def status_summary(self) -> str:
+        acc = self.total_wins / self.total_trades * 100 if self.total_trades > 0 else 0
+        pos = "LONG" if self.direction == 1 else "SHORT" if self.direction == -1 else "FLAT"
+        return (f"{self.worker_id}: €{self.capital:.0f} | {self.total_trades}t "
+                f"{acc:.0f}% | {pos}")

src/live/telegram_notifier.py

@@ -0,0 +1,39 @@
+"""Notifiche Telegram per il paper trader."""
+from __future__ import annotations
+
+import os
+import urllib.request
+import urllib.parse
+import json
+
+BOT_TOKEN = os.environ.get("TELEGRAM_BOT_TOKEN", "")
+CHAT_ID = os.environ.get("TELEGRAM_CHAT_ID", "")
+
+NOTIFY_EVENTS = {
+    "SIGNAL", "OPENED", "CLOSED", "OPEN_FAILED", "CLOSE_FAILED",
+    "ERROR", "STARTUP", "SHUTDOWN", "TRAINING_FAILED",
+}
+
+
+def send_telegram(text: str) -> bool:
+    if not BOT_TOKEN or not CHAT_ID:
+        return False
+    try:
+        url = f"https://api.telegram.org/bot{BOT_TOKEN}/sendMessage"
+        data = urllib.parse.urlencode({"chat_id": CHAT_ID, "text": text, "parse_mode": "HTML"}).encode()
+        urllib.request.urlopen(url, data, timeout=10)
+        return True
+    except Exception:
+        return False
+
+
+def notify_event(event: str, data: dict | None = None):
+    if event not in NOTIFY_EVENTS:
+        return
+    lines = [f"📊 <b>{event}</b>"]
+    if data:
+        for k, v in data.items():
+            if k in ("signal",):
+                continue
+            lines.append(f"  {k}: {v}")
+    send_telegram("\n".join(lines))

src/strategies/__init__.py

@@ -0,0 +1,11 @@
+"""Strategie di trading — classe base e indicatori condivisi."""
+from src.strategies.base import Strategy, Signal, BacktestResult, YearlyStats
+from src.strategies.indicators import (
+    keltner_ratio, detect_squeezes, ema, atr, rv_annualized, rolling_correlation,
+)
+
+__all__ = [
+    "Strategy", "Signal", "BacktestResult", "YearlyStats",
+    "keltner_ratio", "detect_squeezes", "ema", "atr",
+    "rv_annualized", "rolling_correlation",
+]

src/strategies/base.py

@@ -0,0 +1,243 @@
+"""Classe base astratta per tutte le strategie di trading."""
+from __future__ import annotations
+
+from abc import ABC, abstractmethod
+from dataclasses import dataclass, field
+
+import numpy as np
+import pandas as pd
+
+from src.data.downloader import load_data
+
+
+@dataclass
+class Signal:
+    """Segnale di trading generato da una strategia."""
+    idx: int
+    direction: int          # +1 long, -1 short
+    entry_price: float
+    metadata: dict = field(default_factory=dict)
+
+
+@dataclass
+class YearlyStats:
+    year: int
+    trades: int
+    wins: int
+    pnl: float
+
+    @property
+    def accuracy(self) -> float:
+        return self.wins / self.trades * 100 if self.trades > 0 else 0.0
+
+
+@dataclass
+class BacktestResult:
+    """Risultato completo di un backtest."""
+    strategy_name: str
+    asset: str
+    timeframe: str
+    params: dict
+
+    trades: int
+    wins: int
+    pnl: float
+    capital: float
+    initial_capital: float
+    max_dd: float
+    time_in_market_pct: float
+    avg_trade_duration_h: float
+    years_active: int
+    yearly: list[YearlyStats]
+
+    @property
+    def accuracy(self) -> float:
+        return self.wins / self.trades * 100 if self.trades > 0 else 0.0
+
+    @property
+    def sharpe(self) -> float:
+        pnls = []
+        for ys in self.yearly:
+            pnls.append(ys.pnl)
+        if len(pnls) < 2 or np.std(pnls) == 0:
+            return 0.0
+        return float(np.mean(pnls) / np.std(pnls) * np.sqrt(len(pnls)))
+
+    @property
+    def daily_pnl(self) -> float:
+        return self.pnl / (self.years_active * 365) if self.years_active > 0 else 0.0
+
+    @property
+    def worst_year(self) -> YearlyStats | None:
+        valid = [y for y in self.yearly if y.trades >= 10]
+        if not valid:
+            valid = self.yearly
+        return min(valid, key=lambda y: y.accuracy) if valid else None
+
+    def print_summary(self):
+        worst = self.worst_year
+        worst_str = f"{worst.year}({worst.accuracy:.0f}%)" if worst else "N/A"
+        dur = f"{self.avg_trade_duration_h:.0f}h" if self.avg_trade_duration_h >= 1 else f"{self.avg_trade_duration_h * 60:.0f}m"
+        print(f"  {self.strategy_name:<30s} {self.asset:>3s} {self.timeframe:>3s} "
+              f"{self.trades:>5d}t {self.accuracy:>5.1f}% "
+              f"€{self.pnl:>+9.0f} DD {self.max_dd:>4.1f}% "
+              f"€/d {self.daily_pnl:>+6.2f} "
+              f"Mkt {self.time_in_market_pct:>4.1f}% {dur:>5s} "
+              f"worst={worst_str} {self.years_active}y")
+
+    def print_yearly(self):
+        print(f"\n  {self.strategy_name} [{self.asset} {self.timeframe}] — per anno:")
+        print(f"  {'Anno':>6s} {'Trades':>7s} {'Acc':>6s} {'PnL€':>9s}")
+        for ys in sorted(self.yearly, key=lambda y: y.year):
+            print(f"  {ys.year:>6d} {ys.trades:>7d} {ys.accuracy:>5.1f}% €{ys.pnl:>+8.0f}")
+
+
+TF_MINUTES = {"1m": 1, "5m": 5, "15m": 15, "1h": 60, "4h": 240, "1d": 1440}
+
+
+class Strategy(ABC):
+    """Classe base per tutte le strategie.
+
+    Sottoclassi devono implementare:
+      - name, description, default_assets, default_timeframes
+      - generate_signals(df, timestamps, **params) -> list[Signal]
+    """
+
+    name: str = "unnamed"
+    description: str = ""
+    default_assets: list[str] = ["BTC", "ETH"]
+    default_timeframes: list[str] = ["15m", "1h"]
+
+    # Parametri di backtest
+    fee_rt: float = 0.002
+    leverage: float = 3.0
+    position_size: float = 0.15
+    initial_capital: float = 1000.0
+
+    @abstractmethod
+    def generate_signals(self, df: pd.DataFrame, ts: pd.DatetimeIndex,
+                         **params) -> list[Signal]:
+        """Genera segnali di trading dal dataframe OHLCV.
+
+        Args:
+            df: DataFrame con colonne open, high, low, close, volume, timestamp
+            ts: DatetimeIndex UTC dei timestamp
+            **params: parametri specifici della strategia
+
+        Returns:
+            Lista di Signal con idx, direction, entry_price
+        """
+        ...
+
+    def backtest(self, asset: str, tf: str, hold: int = 3,
+                 **params) -> BacktestResult | None:
+        """Esegue backtest su un asset/timeframe."""
+        df = load_data(asset, tf)
+        c = df["close"].values
+        n = len(c)
+        ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+
+        sig_params = {**params, "asset": asset, "tf": tf}
+        signals = self.generate_signals(df, ts, **sig_params)
+        if not signals:
+            return None
+
+        yearly: dict[int, dict] = {}
+        capital = float(self.initial_capital)
+        peak = capital
+        max_dd = 0.0
+        total_bars = 0
+
+        for sig in signals:
+            i = sig.idx
+            if i + hold >= n or i < 1:
+                continue
+
+            entry = sig.entry_price
+            exit_price = c[min(i + hold - 1, n - 1)]
+            actual = (exit_price - entry) / entry * sig.direction
+            net = actual * self.leverage - self.fee_rt * self.leverage
+
+            capital += capital * self.position_size * net
+            capital = max(capital, 10)
+            if capital > peak:
+                peak = capital
+            dd = (peak - capital) / peak
+            max_dd = max(max_dd, dd)
+            total_bars += hold
+
+            year = ts.iloc[i].year
+            if year not in yearly:
+                yearly[year] = {"w": 0, "t": 0, "pnl": 0.0}
+            yearly[year]["t"] += 1
+            if actual > 0:
+                yearly[year]["w"] += 1
+            yearly[year]["pnl"] += net * self.initial_capital
+
+        all_t = sum(d["t"] for d in yearly.values())
+        all_w = sum(d["w"] for d in yearly.values())
+        if all_t == 0:
+            return None
+
+        yearly_stats = [
+            YearlyStats(year=y, trades=d["t"], wins=d["w"], pnl=d["pnl"])
+            for y, d in sorted(yearly.items())
+        ]
+
+        return BacktestResult(
+            strategy_name=self.name,
+            asset=asset,
+            timeframe=tf,
+            params=params,
+            trades=all_t,
+            wins=all_w,
+            pnl=sum(d["pnl"] for d in yearly.values()),
+            capital=capital,
+            initial_capital=self.initial_capital,
+            max_dd=max_dd * 100,
+            time_in_market_pct=total_bars / n * 100,
+            avg_trade_duration_h=hold * TF_MINUTES.get(tf, 60) / 60,
+            years_active=len(yearly),
+            yearly=yearly_stats,
+        )
+
+    def run_all(self, assets: list[str] | None = None,
+                timeframes: list[str] | None = None,
+                hold: int = 3, **params) -> list[BacktestResult]:
+        """Esegue backtest su tutte le combinazioni asset/timeframe."""
+        assets = assets or self.default_assets
+        timeframes = timeframes or self.default_timeframes
+        results = []
+        for asset in assets:
+            for tf in timeframes:
+                r = self.backtest(asset, tf, hold=hold, **params)
+                if r and r.trades >= 20:
+                    results.append(r)
+        results.sort(key=lambda r: r.accuracy, reverse=True)
+        return results
+
+    def report(self, results: list[BacktestResult] | None = None,
+               assets: list[str] | None = None,
+               timeframes: list[str] | None = None,
+               hold: int = 3, **params):
+        """Esegue e stampa report completo."""
+        if results is None:
+            results = self.run_all(assets, timeframes, hold, **params)
+
+        print(f"\n{'=' * 120}")
+        print(f"  {self.name} — {self.description}")
+        print(f"  Fee: {self.fee_rt*100:.1f}% RT | Leva: {self.leverage:.0f}x | Pos: {self.position_size*100:.0f}%")
+        print(f"{'=' * 120}")
+        print(f"  {'Nome':<30s} {'A/T':>7s} {'Trades':>6s} {'Acc':>6s} "
+              f"{'PnL€':>10s} {'DD%':>6s} {'€/day':>7s} "
+              f"{'Mkt%':>5s} {'Dur':>5s} {'Worst':>12s} {'Anni':>4s}")
+        print(f"  {'─' * 110}")
+
+        for r in results:
+            r.print_summary()
+
+        if results:
+            best = results[0]
+            best.print_yearly()
+
+        return results

src/strategies/fade_base.py

@@ -0,0 +1,116 @@
+"""Base condivisa per strategie mean-reversion con exit TP/SL/max_bars.
+
+Tutte le strategie fade (MR02/MR03/MR07) generano Signal con metadata
+{tp, sl, max_bars} e usano lo stesso backtest fedele: ingresso a close[i]
+(eseguibile dal vivo), uscita su take-profit / stop-loss intrabar (high/low)
+o time-limit, una posizione per volta (non-overlap), capitale composto,
+fee+leva nette. Identico all'engine di scripts/analysis/strategy_research.py.
+
+Le sottoclassi implementano solo generate_signals().
+"""
+from __future__ import annotations
+
+import numpy as np
+import pandas as pd
+
+from src.strategies.base import Strategy, BacktestResult, YearlyStats, TF_MINUTES
+from src.data.downloader import load_data
+
+
+def atr(df: pd.DataFrame, n: int = 14) -> np.ndarray:
+    h, l, c = df["high"].values, df["low"].values, df["close"].values
+    pc = np.roll(c, 1); pc[0] = c[0]
+    tr = np.maximum(h - l, np.maximum(np.abs(h - pc), np.abs(l - pc)))
+    return pd.Series(tr).rolling(n).mean().values
+
+
+def trend_distance(df: pd.DataFrame, ema_long: int = 200) -> np.ndarray:
+    """Distanza del close dalla EMA lunga, in multipli di ATR(14).
+
+    Misura quanto il prezzo e' esteso rispetto al trend di fondo. Le fade
+    falliscono quando si oppongono a un trend estremo (crolli/parabolic): il
+    filtro `trend_max` salta i segnali con distanza > soglia. Riduce DD e alza
+    l'accuratezza (validato OOS: scripts/analysis/risk_portfolio.py).
+    """
+    c = df["close"].values
+    a = atr(df, 14)
+    el = pd.Series(c).ewm(span=ema_long, adjust=False).mean().values
+    with np.errstate(divide="ignore", invalid="ignore"):
+        return np.abs(c - el) / np.where(a == 0, np.nan, a)
+
+
+class FadeStrategy(Strategy):
+    """Strategy con backtest intrabar TP/SL/max_bars (exit guidati dai metadata)."""
+
+    fee_rt = 0.001          # Deribit perp realistico (taker 0.05%/lato)
+    leverage = 3.0
+    position_size = 0.15
+    initial_capital = 1000.0
+
+    def backtest(self, asset: str, tf: str = "1h", hold: int = 3,
+                 **params) -> BacktestResult | None:
+        df = load_data(asset, tf)
+        ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
+        signals = self.generate_signals(df, ts, **params)
+        if not signals:
+            return None
+
+        h, l, c = df["high"].values, df["low"].values, df["close"].values
+        n = len(c)
+        fee = self.fee_rt * self.leverage
+        capital = peak = float(self.initial_capital)
+        max_dd = 0.0
+        total_bars = 0
+        last_exit = -1
+        yearly: dict[int, dict] = {}
+
+        for sig in signals:
+            i, d = sig.idx, sig.direction
+            if i <= last_exit or i + 1 >= n:
+                continue
+            entry = c[i]
+            tp, sl, mb = sig.metadata["tp"], sig.metadata["sl"], sig.metadata["max_bars"]
+            exit_p = c[min(i + mb, n - 1)]
+            j = min(i + mb, n - 1)
+            for step in range(1, mb + 1):
+                j = i + step
+                if j >= n:
+                    j = n - 1; exit_p = c[j]; break
+                hit_sl = (d == 1 and l[j] <= sl) or (d == -1 and h[j] >= sl)
+                hit_tp = (d == 1 and h[j] >= tp) or (d == -1 and l[j] <= tp)
+                if hit_sl:                          # conservativo: SL prima del TP nello stesso bar
+                    exit_p = sl; break
+                if hit_tp:
+                    exit_p = tp; break
+                if step == mb:
+                    exit_p = c[j]
+
+            ret = (exit_p - entry) / entry * d * self.leverage - fee
+            capital = max(capital + capital * self.position_size * ret, 10.0)
+            if capital > peak:
+                peak = capital
+            max_dd = max(max_dd, (peak - capital) / peak)
+            total_bars += (j - i)
+            last_exit = j
+
+            year = ts.iloc[i].year
+            yr = yearly.setdefault(year, {"w": 0, "t": 0, "pnl": 0.0})
+            yr["t"] += 1
+            if ret > 0:
+                yr["w"] += 1
+            yr["pnl"] += ret * self.initial_capital
+
+        all_t = sum(v["t"] for v in yearly.values())
+        all_w = sum(v["w"] for v in yearly.values())
+        if all_t == 0:
+            return None
+
+        yearly_stats = [YearlyStats(y, v["t"], v["w"], v["pnl"]) for y, v in sorted(yearly.items())]
+        return BacktestResult(
+            strategy_name=self.name, asset=asset, timeframe=tf, params=params,
+            trades=all_t, wins=all_w, pnl=sum(v["pnl"] for v in yearly.values()),
+            capital=capital, initial_capital=self.initial_capital,
+            max_dd=max_dd * 100, time_in_market_pct=total_bars / n * 100,
+            avg_trade_duration_h=total_bars / all_t * TF_MINUTES.get(tf, 60) / 60,
+            years_active=len(yearly), yearly=yearly_stats,
+        )

src/strategies/indicators.py

@@ -0,0 +1,102 @@
+"""Indicatori tecnici condivisi tra tutte le strategie."""
+from __future__ import annotations
+
+import numpy as np
+
+
+def keltner_ratio(close: np.ndarray, high: np.ndarray, low: np.ndarray,
+                  window: int = 14) -> np.ndarray:
+    """Rapporto Bollinger / Keltner. Sotto 1 = squeeze (BB dentro KC)."""
+    n = len(close)
+    r = np.full(n, np.nan)
+    for i in range(window, n):
+        wc = close[i - window:i]
+        wh = high[i - window:i]
+        wl = low[i - window:i]
+        ma = np.mean(wc)
+        bb_std = np.std(wc)
+        tr = np.maximum(
+            wh - wl,
+            np.maximum(np.abs(wh - np.roll(wc, 1)), np.abs(wl - np.roll(wc, 1))),
+        )
+        atr = np.mean(tr[1:])
+        kc = (ma + 1.5 * atr) - (ma - 1.5 * atr)
+        bb = (ma + 2 * bb_std) - (ma - 2 * bb_std)
+        if kc > 0:
+            r[i] = bb / kc
+    return r
+
+
+def detect_squeezes(close: np.ndarray, high: np.ndarray, low: np.ndarray,
+                    kcr: np.ndarray, sq_thr: float = 0.8,
+                    min_dur: int = 5) -> list[dict]:
+    """Rileva squeeze events: periodi dove BB sta dentro KC."""
+    events: list[dict] = []
+    in_sq = False
+    sq_start = 0
+    for i in range(1, len(close)):
+        if np.isnan(kcr[i]):
+            continue
+        is_sq = kcr[i] < sq_thr
+        if is_sq and not in_sq:
+            in_sq = True
+            sq_start = i
+        elif not is_sq and in_sq:
+            in_sq = False
+            dur = i - sq_start
+            if dur < min_dur:
+                continue
+            events.append({
+                "idx": i, "dur": dur, "sq_start": sq_start,
+                "kcr_at_release": kcr[i],
+            })
+    return events
+
+
+def ema(arr: np.ndarray, period: int) -> np.ndarray:
+    """Exponential Moving Average."""
+    r = np.full(len(arr), np.nan)
+    k = 2 / (period + 1)
+    r[period - 1] = np.mean(arr[:period])
+    for i in range(period, len(arr)):
+        r[i] = arr[i] * k + r[i - 1] * (1 - k)
+    return r
+
+
+def atr(high: np.ndarray, low: np.ndarray, close: np.ndarray,
+        period: int = 14) -> np.ndarray:
+    """Average True Range (EMA-smoothed)."""
+    tr = np.maximum(
+        high - low,
+        np.maximum(np.abs(high - np.roll(close, 1)), np.abs(low - np.roll(close, 1))),
+    )
+    tr[0] = high[0] - low[0]
+    r = np.full(len(close), np.nan)
+    r[period - 1] = np.mean(tr[:period])
+    k = 2 / (period + 1)
+    for i in range(period, len(close)):
+        r[i] = tr[i] * k + r[i - 1] * (1 - k)
+    return r
+
+
+def rv_annualized(close: np.ndarray, window: int) -> np.ndarray:
+    """Realized volatility annualizzata (hourly data assumed)."""
+    lr = np.diff(np.log(np.where(close == 0, 1e-10, close)))
+    r = np.full(len(close), np.nan)
+    for i in range(window, len(lr)):
+        r[i + 1] = np.std(lr[i - window:i]) * np.sqrt(24 * 365)
+    return r
+
+
+def rolling_correlation(close_a: np.ndarray, close_b: np.ndarray,
+                        window: int = 48) -> np.ndarray:
+    """Correlazione rolling tra rendimenti logaritmici di due asset."""
+    n = max(len(close_a), len(close_b))
+    ret_a = np.diff(np.log(np.where(close_a == 0, 1e-10, close_a)))
+    ret_b = np.diff(np.log(np.where(close_b[:len(close_a)] == 0, 1e-10, close_b[:len(close_a)])))
+    min_len = min(len(ret_a), len(ret_b))
+    corr = np.full(n, np.nan)
+    for i in range(window, min_len):
+        cv = np.corrcoef(ret_a[i - window:i], ret_b[i - window:i])[0, 1]
+        corr[i + 1] = cv if np.isfinite(cv) else 0
+    return corr

strategies.yml

@@ -0,0 +1,92 @@
+defaults:
+  capital: 1000
+  position_size: 0.15
+  leverage: 3
+  hold_bars: 3
+  poll_seconds: 60
+  retrain_hours: 24
+
+# Solo MR01 Bollinger fade (mean-reversion): unica con edge netto validato
+# out-of-sample e fee-aware. La famiglia squeeze e' in scripts/waste/.
+# ATTENZIONE: MR01 esce su TP-alla-media / SL-ad-ATR / max_bars (vedi metadata
+# dei Signal). Lo StrategyWorker attuale esce solo a hold_bars/stop -2% fisso:
+# va aggiornato per usare gli exit in metadata PRIMA di tradare MR01 dal vivo.
+strategies:
+  - name: MR01_bollinger_fade
+    asset: BTC
+    tf: 1h
+    enabled: true
+    params:
+      bb_window: 50
+      k: 2.5
+      sl_atr: 2.0
+      max_bars: 24
+      trend_max: 3.0       # salta fade contro trend estremo (|close-EMA200|/ATR>3): Acc+ DD-
+      ema_long: 200
+
+  # ETH: edge positivo ma DD piu' alto (~70%); leva piu' bassa consigliata
+  - name: MR01_bollinger_fade
+    asset: ETH
+    tf: 1h
+    enabled: true
+    params:
+      bb_window: 50
+      k: 2.5
+      sl_atr: 2.0
+      max_bars: 24
+      trend_max: 3.0       # salta fade contro trend estremo (|close-EMA200|/ATR>3): Acc+ DD-
+      ema_long: 200
+
+  # MR02 Donchian fade: fade rottura canale (estremi H/L). Robusto su tutta la
+  # griglia n x sl_atr e tutte le fee. BTC +879%/+171% OOS (8/9 anni), ETH enorme.
+  - name: MR02_donchian_fade
+    asset: BTC
+    tf: 1h
+    enabled: true
+    params:
+      n: 20
+      sl_atr: 2.0
+      max_bars: 24
+      trend_max: 3.0       # salta fade contro trend estremo (|close-EMA200|/ATR>3): Acc+ DD-
+      ema_long: 200
+  - name: MR02_donchian_fade
+    asset: ETH
+    tf: 1h
+    enabled: true
+    params:
+      n: 20
+      sl_atr: 2.0
+      max_bars: 24
+      trend_max: 3.0       # salta fade contro trend estremo (|close-EMA200|/ATR>3): Acc+ DD-
+      ema_long: 200
+
+  # MR03 Keltner fade -> spostata in scripts/waste/ (fade piu' debole e ridondante
+  # con MR01; il filtro trend la peggiorava su BTC). Vedi CLAUDE.md.
+
+  # MR07 Return reversal: fade movimento di barra estremo (z dei rendimenti).
+  # Meccanismo distinto (volatilita' rendimenti, non livelli). Esposizione bassa
+  # (~8%). BTC +447%/+105% OOS DD25%, ETH +335%/+195% OOS DD46%.
+  - name: MR07_return_reversal
+    asset: BTC
+    tf: 1h
+    enabled: true
+    params:
+      n: 50
+      k: 3.5
+      tp_atr: 2.0
+      sl_atr: 1.5
+      max_bars: 24
+      trend_max: 3.0       # salta fade contro trend estremo (|close-EMA200|/ATR>3): Acc+ DD-
+      ema_long: 200
+  - name: MR07_return_reversal
+    asset: ETH
+    tf: 1h
+    enabled: true
+    params:
+      n: 50
+      k: 3.5
+      tp_atr: 2.0
+      sl_atr: 1.5
+      max_bars: 24
+      trend_max: 3.0       # salta fade contro trend estremo (|close-EMA200|/ATR>3): Acc+ DD-
+      ema_long: 200

uv.lock

@@ -542,30 +542,30 @@ wheels = [
 
 [[package]]
 name = "cuda-bindings"
-version = "13.2.0"
+version = "13.3.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "cuda-pathfinder", marker = "sys_platform != 'emscripten' and sys_platform != 'win32'" },
 ]
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/e0/a9/3a8241c6e19483ac1f1dcf5c10238205dcb8a6e9d0d4d4709240dff28ff4/cuda_bindings-13.2.0-cp311-cp311-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:721104c603f059780d287969be3d194a18d0cc3b713ed9049065a1107706759d", size = 5730273, upload-time = "2026-03-11T00:12:37.18Z" },
-    { url = "https://files.pythonhosted.org/packages/e9/94/2748597f47bb1600cd466b20cab4159f1530a3a33fe7f70fee199b3abb9e/cuda_bindings-13.2.0-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:1eba9504ac70667dd48313395fe05157518fd6371b532790e96fbb31bbb5a5e1", size = 6313924, upload-time = "2026-03-11T00:12:39.462Z" },
-    { url = "https://files.pythonhosted.org/packages/52/c8/b2589d68acf7e3d63e2be330b84bc25712e97ed799affbca7edd7eae25d6/cuda_bindings-13.2.0-cp312-cp312-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:e865447abfb83d6a98ad5130ed3c70b1fc295ae3eeee39fd07b4ddb0671b6788", size = 5722404, upload-time = "2026-03-11T00:12:44.041Z" },
-    { url = "https://files.pythonhosted.org/packages/1f/92/f899f7bbb5617bb65ec52a6eac1e9a1447a86b916c4194f8a5001b8cde0c/cuda_bindings-13.2.0-cp312-cp312-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:46d8776a55d6d5da9dd6e9858fba2efcda2abe6743871dee47dd06eb8cb6d955", size = 6320619, upload-time = "2026-03-11T00:12:45.939Z" },
-    { url = "https://files.pythonhosted.org/packages/df/93/eef988860a3ca985f82c4f3174fc0cdd94e07331ba9a92e8e064c260337f/cuda_bindings-13.2.0-cp313-cp313-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:6629ca2df6f795b784752409bcaedbd22a7a651b74b56a165ebc0c9dcbd504d0", size = 5614610, upload-time = "2026-03-11T00:12:50.337Z" },
-    { url = "https://files.pythonhosted.org/packages/18/23/6db3aba46864aee357ab2415135b3fe3da7e9f1fa0221fa2a86a5968099c/cuda_bindings-13.2.0-cp313-cp313-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:7dca0da053d3b4cc4869eff49c61c03f3c5dbaa0bcd712317a358d5b8f3f385d", size = 6149914, upload-time = "2026-03-11T00:12:52.374Z" },
-    { url = "https://files.pythonhosted.org/packages/c0/87/87a014f045b77c6de5c8527b0757fe644417b184e5367db977236a141602/cuda_bindings-13.2.0-cp314-cp314-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:a6464b30f46692d6c7f65d4a0e0450d81dd29de3afc1bb515653973d01c2cd6e", size = 5685673, upload-time = "2026-03-11T00:12:56.371Z" },
-    { url = "https://files.pythonhosted.org/packages/ee/5e/c0fe77a73aaefd3fff25ffaccaac69c5a63eafdf8b9a4c476626ef0ac703/cuda_bindings-13.2.0-cp314-cp314-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:f4af9f3e1be603fa12d5ad6cfca7844c9d230befa9792b5abdf7dd79979c3626", size = 6191386, upload-time = "2026-03-11T00:12:58.965Z" },
-    { url = "https://files.pythonhosted.org/packages/5f/58/ed2c3b39c8dd5f96aa7a4abef0d47a73932c7a988e30f5fa428f00ed0da1/cuda_bindings-13.2.0-cp314-cp314t-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:df850a1ff8ce1b3385257b08e47b70e959932f5f432d0a4e46a355962b4e4771", size = 5507469, upload-time = "2026-03-11T00:13:04.063Z" },
-    { url = "https://files.pythonhosted.org/packages/1f/01/0c941b112ceeb21439b05895eace78ca1aa2eaaf695c8521a068fd9b4c00/cuda_bindings-13.2.0-cp314-cp314t-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:e8a16384c6494e5485f39314b0b4afb04bee48d49edb16d5d8593fd35bbd231b", size = 6059693, upload-time = "2026-03-11T00:13:06.003Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/52/50673d25e46d199556f827514bf646a49471d50538c5e577201245b348a9/cuda_bindings-13.3.0-cp311-cp311-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:244a167d81a9d07d3209d02c8e02e848c2b7c38f4d01e8e4d1f9620b173ae006", size = 6051409, upload-time = "2026-05-27T03:59:01.648Z" },
+    { url = "https://files.pythonhosted.org/packages/88/ee/e8f4bdfb808c3689539b7c035d63b6dac9f236b2d6f807f18c7f5f3ef879/cuda_bindings-13.3.0-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:820ab45be3f39a088f39c4a04eb8b852d0b339bff8c518da5c258882b8a4e21b", size = 6671833, upload-time = "2026-05-27T03:59:03.761Z" },
+    { url = "https://files.pythonhosted.org/packages/1f/e0/4b3fdba08ff177e9451f376a4ba2df18d76f9158e6a16cdc062bd83db9fa/cuda_bindings-13.3.0-cp312-cp312-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:f72f67f5790e7fa51e3f893e007e49567573ccdf4ed1ca988fbbbd36cd77847c", size = 6020531, upload-time = "2026-05-27T03:59:07.942Z" },
+    { url = "https://files.pythonhosted.org/packages/04/40/a2ea4d8f032bfd6c220d50b6f92cd61f33d48f31959da39ed1b178cfee54/cuda_bindings-13.3.0-cp312-cp312-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:a99c3b8d584f266c616bd0f30c7cd83e33553e3ef2abad41ff5a74fbc033a69a", size = 6653764, upload-time = "2026-05-27T03:59:09.981Z" },
+    { url = "https://files.pythonhosted.org/packages/ae/a0/156efe7816699c2de1ea2395031db7d010b7af23c243563a3ee6f0ecc1de/cuda_bindings-13.3.0-cp313-cp313-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:a7698fcc4577aa96372866f4d0c9a6cf686cd5c90eab94581c29d37fe6600542", size = 5914803, upload-time = "2026-05-27T03:59:14.011Z" },
+    { url = "https://files.pythonhosted.org/packages/51/91/510aae64d53227b5b36db6bfaea41514b66d92cd65ddc43aa49566f18313/cuda_bindings-13.3.0-cp313-cp313-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:abd908f651160d12c45c5714a38ee102a1173a55433c0d1509ec0e8293beb4a6", size = 6472506, upload-time = "2026-05-27T03:59:16.551Z" },
+    { url = "https://files.pythonhosted.org/packages/01/53/2ef49e5b3734a5531b2ba5d726cba724d9cbb262404e586ed61070604826/cuda_bindings-13.3.0-cp314-cp314-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:4a801fa30e75d25b74252123aefc746b6c4275624d2b8640632dd1dfeeaa1f88", size = 6008814, upload-time = "2026-05-27T03:59:20.921Z" },
+    { url = "https://files.pythonhosted.org/packages/2f/cb/3a9fcf0651e0a49b4d0f1955837ce079245b27086c22fb2f253039bdf324/cuda_bindings-13.3.0-cp314-cp314-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:94d40ef7b4bdd9dce0244a1baa132e0e538f1eb2c0d162fb3648a15e48515365", size = 6531477, upload-time = "2026-05-27T03:59:23.391Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/0f/6987c5ee98f117317a85650ddc79480a3fa59a573ae1c923d0722b56ae71/cuda_bindings-13.3.0-cp314-cp314t-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:e5911bea15810b749a8077f8c45423ed785d51618b8e8664dea1fc8f5a2a76c8", size = 5807073, upload-time = "2026-05-27T03:59:28.218Z" },
+    { url = "https://files.pythonhosted.org/packages/f6/ab/46ceee07dc19f18a5d1c28d592750ed9dbdc803077eb083576a442c9938c/cuda_bindings-13.3.0-cp314-cp314t-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:e2870fed7707a37f8af0c02364b05f355ebe8921604e8c68eb56cf66867e0798", size = 6354325, upload-time = "2026-05-27T03:59:30.715Z" },
 ]
 
 [[package]]
 name = "cuda-pathfinder"
-version = "1.5.4"
+version = "1.5.5"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/11/d0/c177e29701cf1d3008d7d2b16b5fc626592ce13bd535f8795c5f57187e0e/cuda_pathfinder-1.5.4-py3-none-any.whl", hash = "sha256:9563d3175ce1828531acf4b94e1c1c7d67208c347ca002493e2654878b26f4b7", size = 51657, upload-time = "2026-04-27T22:42:07.712Z" },
+    { url = "https://files.pythonhosted.org/packages/11/c8/26f2e4aae92f11522a96043892ba39a90eac610d5242523aa863212bc1c7/cuda_pathfinder-1.5.5-py3-none-any.whl", hash = "sha256:0228c023f95d1480f143ef5c8922d27a2ab052087a942e81dc289c9eb8f91689", size = 51671, upload-time = "2026-05-27T01:21:25.413Z" },
 ]
 
 [[package]]
@@ -2057,6 +2057,7 @@ dependencies = [
     { name = "numpy" },
     { name = "pandas" },
     { name = "pyarrow" },
+    { name = "pyyaml" },
     { name = "requests" },
     { name = "scikit-learn" },
     { name = "scipy" },
@@ -2081,6 +2082,7 @@ requires-dist = [
     { name = "pyarrow", specifier = ">=15.0" },
     { name = "pytest", marker = "extra == 'dev'", specifier = ">=8.0" },
     { name = "pytest-asyncio", marker = "extra == 'dev'", specifier = ">=0.24" },
+    { name = "pyyaml", specifier = ">=6.0" },
     { name = "requests", specifier = ">=2.31" },
     { name = "scikit-learn", specifier = ">=1.3" },
     { name = "scipy", specifier = ">=1.11" },
@@ -2101,6 +2103,61 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/ec/57/56b9bcc3c9c6a792fcbaf139543cee77261f3651ca9da0c93f5c1221264b/python_dateutil-2.9.0.post0-py2.py3-none-any.whl", hash = "sha256:a8b2bc7bffae282281c8140a97d3aa9c14da0b136dfe83f850eea9a5f7470427", size = 229892, upload-time = "2024-03-01T18:36:18.57Z" },
 ]
 
+[[package]]
+name = "pyyaml"
+version = "6.0.3"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/05/8e/961c0007c59b8dd7729d542c61a4d537767a59645b82a0b521206e1e25c2/pyyaml-6.0.3.tar.gz", hash = "sha256:d76623373421df22fb4cf8817020cbb7ef15c725b9d5e45f17e189bfc384190f", size = 130960, upload-time = "2025-09-25T21:33:16.546Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/6d/16/a95b6757765b7b031c9374925bb718d55e0a9ba8a1b6a12d25962ea44347/pyyaml-6.0.3-cp311-cp311-macosx_10_13_x86_64.whl", hash = "sha256:44edc647873928551a01e7a563d7452ccdebee747728c1080d881d68af7b997e", size = 185826, upload-time = "2025-09-25T21:31:58.655Z" },
+    { url = "https://files.pythonhosted.org/packages/16/19/13de8e4377ed53079ee996e1ab0a9c33ec2faf808a4647b7b4c0d46dd239/pyyaml-6.0.3-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:652cb6edd41e718550aad172851962662ff2681490a8a711af6a4d288dd96824", size = 175577, upload-time = "2025-09-25T21:32:00.088Z" },
+    { url = "https://files.pythonhosted.org/packages/0c/62/d2eb46264d4b157dae1275b573017abec435397aa59cbcdab6fc978a8af4/pyyaml-6.0.3-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:10892704fc220243f5305762e276552a0395f7beb4dbf9b14ec8fd43b57f126c", size = 775556, upload-time = "2025-09-25T21:32:01.31Z" },
+    { url = "https://files.pythonhosted.org/packages/10/cb/16c3f2cf3266edd25aaa00d6c4350381c8b012ed6f5276675b9eba8d9ff4/pyyaml-6.0.3-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:850774a7879607d3a6f50d36d04f00ee69e7fc816450e5f7e58d7f17f1ae5c00", size = 882114, upload-time = "2025-09-25T21:32:03.376Z" },
+    { url = "https://files.pythonhosted.org/packages/71/60/917329f640924b18ff085ab889a11c763e0b573da888e8404ff486657602/pyyaml-6.0.3-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:b8bb0864c5a28024fac8a632c443c87c5aa6f215c0b126c449ae1a150412f31d", size = 806638, upload-time = "2025-09-25T21:32:04.553Z" },
+    { url = "https://files.pythonhosted.org/packages/dd/6f/529b0f316a9fd167281a6c3826b5583e6192dba792dd55e3203d3f8e655a/pyyaml-6.0.3-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:1d37d57ad971609cf3c53ba6a7e365e40660e3be0e5175fa9f2365a379d6095a", size = 767463, upload-time = "2025-09-25T21:32:06.152Z" },
+    { url = "https://files.pythonhosted.org/packages/f2/6a/b627b4e0c1dd03718543519ffb2f1deea4a1e6d42fbab8021936a4d22589/pyyaml-6.0.3-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:37503bfbfc9d2c40b344d06b2199cf0e96e97957ab1c1b546fd4f87e53e5d3e4", size = 794986, upload-time = "2025-09-25T21:32:07.367Z" },
+    { url = "https://files.pythonhosted.org/packages/45/91/47a6e1c42d9ee337c4839208f30d9f09caa9f720ec7582917b264defc875/pyyaml-6.0.3-cp311-cp311-win32.whl", hash = "sha256:8098f252adfa6c80ab48096053f512f2321f0b998f98150cea9bd23d83e1467b", size = 142543, upload-time = "2025-09-25T21:32:08.95Z" },
+    { url = "https://files.pythonhosted.org/packages/da/e3/ea007450a105ae919a72393cb06f122f288ef60bba2dc64b26e2646fa315/pyyaml-6.0.3-cp311-cp311-win_amd64.whl", hash = "sha256:9f3bfb4965eb874431221a3ff3fdcddc7e74e3b07799e0e84ca4a0f867d449bf", size = 158763, upload-time = "2025-09-25T21:32:09.96Z" },
+    { url = "https://files.pythonhosted.org/packages/d1/33/422b98d2195232ca1826284a76852ad5a86fe23e31b009c9886b2d0fb8b2/pyyaml-6.0.3-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:7f047e29dcae44602496db43be01ad42fc6f1cc0d8cd6c83d342306c32270196", size = 182063, upload-time = "2025-09-25T21:32:11.445Z" },
+    { url = "https://files.pythonhosted.org/packages/89/a0/6cf41a19a1f2f3feab0e9c0b74134aa2ce6849093d5517a0c550fe37a648/pyyaml-6.0.3-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:fc09d0aa354569bc501d4e787133afc08552722d3ab34836a80547331bb5d4a0", size = 173973, upload-time = "2025-09-25T21:32:12.492Z" },
+    { url = "https://files.pythonhosted.org/packages/ed/23/7a778b6bd0b9a8039df8b1b1d80e2e2ad78aa04171592c8a5c43a56a6af4/pyyaml-6.0.3-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:9149cad251584d5fb4981be1ecde53a1ca46c891a79788c0df828d2f166bda28", size = 775116, upload-time = "2025-09-25T21:32:13.652Z" },
+    { url = "https://files.pythonhosted.org/packages/65/30/d7353c338e12baef4ecc1b09e877c1970bd3382789c159b4f89d6a70dc09/pyyaml-6.0.3-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:5fdec68f91a0c6739b380c83b951e2c72ac0197ace422360e6d5a959d8d97b2c", size = 844011, upload-time = "2025-09-25T21:32:15.21Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/9d/b3589d3877982d4f2329302ef98a8026e7f4443c765c46cfecc8858c6b4b/pyyaml-6.0.3-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:ba1cc08a7ccde2d2ec775841541641e4548226580ab850948cbfda66a1befcdc", size = 807870, upload-time = "2025-09-25T21:32:16.431Z" },
+    { url = "https://files.pythonhosted.org/packages/05/c0/b3be26a015601b822b97d9149ff8cb5ead58c66f981e04fedf4e762f4bd4/pyyaml-6.0.3-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:8dc52c23056b9ddd46818a57b78404882310fb473d63f17b07d5c40421e47f8e", size = 761089, upload-time = "2025-09-25T21:32:17.56Z" },
+    { url = "https://files.pythonhosted.org/packages/be/8e/98435a21d1d4b46590d5459a22d88128103f8da4c2d4cb8f14f2a96504e1/pyyaml-6.0.3-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:41715c910c881bc081f1e8872880d3c650acf13dfa8214bad49ed4cede7c34ea", size = 790181, upload-time = "2025-09-25T21:32:18.834Z" },
+    { url = "https://files.pythonhosted.org/packages/74/93/7baea19427dcfbe1e5a372d81473250b379f04b1bd3c4c5ff825e2327202/pyyaml-6.0.3-cp312-cp312-win32.whl", hash = "sha256:96b533f0e99f6579b3d4d4995707cf36df9100d67e0c8303a0c55b27b5f99bc5", size = 137658, upload-time = "2025-09-25T21:32:20.209Z" },
+    { url = "https://files.pythonhosted.org/packages/86/bf/899e81e4cce32febab4fb42bb97dcdf66bc135272882d1987881a4b519e9/pyyaml-6.0.3-cp312-cp312-win_amd64.whl", hash = "sha256:5fcd34e47f6e0b794d17de1b4ff496c00986e1c83f7ab2fb8fcfe9616ff7477b", size = 154003, upload-time = "2025-09-25T21:32:21.167Z" },
+    { url = "https://files.pythonhosted.org/packages/1a/08/67bd04656199bbb51dbed1439b7f27601dfb576fb864099c7ef0c3e55531/pyyaml-6.0.3-cp312-cp312-win_arm64.whl", hash = "sha256:64386e5e707d03a7e172c0701abfb7e10f0fb753ee1d773128192742712a98fd", size = 140344, upload-time = "2025-09-25T21:32:22.617Z" },
+    { url = "https://files.pythonhosted.org/packages/d1/11/0fd08f8192109f7169db964b5707a2f1e8b745d4e239b784a5a1dd80d1db/pyyaml-6.0.3-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:8da9669d359f02c0b91ccc01cac4a67f16afec0dac22c2ad09f46bee0697eba8", size = 181669, upload-time = "2025-09-25T21:32:23.673Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/16/95309993f1d3748cd644e02e38b75d50cbc0d9561d21f390a76242ce073f/pyyaml-6.0.3-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:2283a07e2c21a2aa78d9c4442724ec1eb15f5e42a723b99cb3d822d48f5f7ad1", size = 173252, upload-time = "2025-09-25T21:32:25.149Z" },
+    { url = "https://files.pythonhosted.org/packages/50/31/b20f376d3f810b9b2371e72ef5adb33879b25edb7a6d072cb7ca0c486398/pyyaml-6.0.3-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:ee2922902c45ae8ccada2c5b501ab86c36525b883eff4255313a253a3160861c", size = 767081, upload-time = "2025-09-25T21:32:26.575Z" },
+    { url = "https://files.pythonhosted.org/packages/49/1e/a55ca81e949270d5d4432fbbd19dfea5321eda7c41a849d443dc92fd1ff7/pyyaml-6.0.3-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:a33284e20b78bd4a18c8c2282d549d10bc8408a2a7ff57653c0cf0b9be0afce5", size = 841159, upload-time = "2025-09-25T21:32:27.727Z" },
+    { url = "https://files.pythonhosted.org/packages/74/27/e5b8f34d02d9995b80abcef563ea1f8b56d20134d8f4e5e81733b1feceb2/pyyaml-6.0.3-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:0f29edc409a6392443abf94b9cf89ce99889a1dd5376d94316ae5145dfedd5d6", size = 801626, upload-time = "2025-09-25T21:32:28.878Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/11/ba845c23988798f40e52ba45f34849aa8a1f2d4af4b798588010792ebad6/pyyaml-6.0.3-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:f7057c9a337546edc7973c0d3ba84ddcdf0daa14533c2065749c9075001090e6", size = 753613, upload-time = "2025-09-25T21:32:30.178Z" },
+    { url = "https://files.pythonhosted.org/packages/3d/e0/7966e1a7bfc0a45bf0a7fb6b98ea03fc9b8d84fa7f2229e9659680b69ee3/pyyaml-6.0.3-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:eda16858a3cab07b80edaf74336ece1f986ba330fdb8ee0d6c0d68fe82bc96be", size = 794115, upload-time = "2025-09-25T21:32:31.353Z" },
+    { url = "https://files.pythonhosted.org/packages/de/94/980b50a6531b3019e45ddeada0626d45fa85cbe22300844a7983285bed3b/pyyaml-6.0.3-cp313-cp313-win32.whl", hash = "sha256:d0eae10f8159e8fdad514efdc92d74fd8d682c933a6dd088030f3834bc8e6b26", size = 137427, upload-time = "2025-09-25T21:32:32.58Z" },
+    { url = "https://files.pythonhosted.org/packages/97/c9/39d5b874e8b28845e4ec2202b5da735d0199dbe5b8fb85f91398814a9a46/pyyaml-6.0.3-cp313-cp313-win_amd64.whl", hash = "sha256:79005a0d97d5ddabfeeea4cf676af11e647e41d81c9a7722a193022accdb6b7c", size = 154090, upload-time = "2025-09-25T21:32:33.659Z" },
+    { url = "https://files.pythonhosted.org/packages/73/e8/2bdf3ca2090f68bb3d75b44da7bbc71843b19c9f2b9cb9b0f4ab7a5a4329/pyyaml-6.0.3-cp313-cp313-win_arm64.whl", hash = "sha256:5498cd1645aa724a7c71c8f378eb29ebe23da2fc0d7a08071d89469bf1d2defb", size = 140246, upload-time = "2025-09-25T21:32:34.663Z" },
+    { url = "https://files.pythonhosted.org/packages/9d/8c/f4bd7f6465179953d3ac9bc44ac1a8a3e6122cf8ada906b4f96c60172d43/pyyaml-6.0.3-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:8d1fab6bb153a416f9aeb4b8763bc0f22a5586065f86f7664fc23339fc1c1fac", size = 181814, upload-time = "2025-09-25T21:32:35.712Z" },
+    { url = "https://files.pythonhosted.org/packages/bd/9c/4d95bb87eb2063d20db7b60faa3840c1b18025517ae857371c4dd55a6b3a/pyyaml-6.0.3-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:34d5fcd24b8445fadc33f9cf348c1047101756fd760b4dacb5c3e99755703310", size = 173809, upload-time = "2025-09-25T21:32:36.789Z" },
+    { url = "https://files.pythonhosted.org/packages/92/b5/47e807c2623074914e29dabd16cbbdd4bf5e9b2db9f8090fa64411fc5382/pyyaml-6.0.3-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:501a031947e3a9025ed4405a168e6ef5ae3126c59f90ce0cd6f2bfc477be31b7", size = 766454, upload-time = "2025-09-25T21:32:37.966Z" },
+    { url = "https://files.pythonhosted.org/packages/02/9e/e5e9b168be58564121efb3de6859c452fccde0ab093d8438905899a3a483/pyyaml-6.0.3-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:b3bc83488de33889877a0f2543ade9f70c67d66d9ebb4ac959502e12de895788", size = 836355, upload-time = "2025-09-25T21:32:39.178Z" },
+    { url = "https://files.pythonhosted.org/packages/88/f9/16491d7ed2a919954993e48aa941b200f38040928474c9e85ea9e64222c3/pyyaml-6.0.3-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:c458b6d084f9b935061bc36216e8a69a7e293a2f1e68bf956dcd9e6cbcd143f5", size = 794175, upload-time = "2025-09-25T21:32:40.865Z" },
+    { url = "https://files.pythonhosted.org/packages/dd/3f/5989debef34dc6397317802b527dbbafb2b4760878a53d4166579111411e/pyyaml-6.0.3-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:7c6610def4f163542a622a73fb39f534f8c101d690126992300bf3207eab9764", size = 755228, upload-time = "2025-09-25T21:32:42.084Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/ce/af88a49043cd2e265be63d083fc75b27b6ed062f5f9fd6cdc223ad62f03e/pyyaml-6.0.3-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:5190d403f121660ce8d1d2c1bb2ef1bd05b5f68533fc5c2ea899bd15f4399b35", size = 789194, upload-time = "2025-09-25T21:32:43.362Z" },
+    { url = "https://files.pythonhosted.org/packages/23/20/bb6982b26a40bb43951265ba29d4c246ef0ff59c9fdcdf0ed04e0687de4d/pyyaml-6.0.3-cp314-cp314-win_amd64.whl", hash = "sha256:4a2e8cebe2ff6ab7d1050ecd59c25d4c8bd7e6f400f5f82b96557ac0abafd0ac", size = 156429, upload-time = "2025-09-25T21:32:57.844Z" },
+    { url = "https://files.pythonhosted.org/packages/f4/f4/a4541072bb9422c8a883ab55255f918fa378ecf083f5b85e87fc2b4eda1b/pyyaml-6.0.3-cp314-cp314-win_arm64.whl", hash = "sha256:93dda82c9c22deb0a405ea4dc5f2d0cda384168e466364dec6255b293923b2f3", size = 143912, upload-time = "2025-09-25T21:32:59.247Z" },
+    { url = "https://files.pythonhosted.org/packages/7c/f9/07dd09ae774e4616edf6cda684ee78f97777bdd15847253637a6f052a62f/pyyaml-6.0.3-cp314-cp314t-macosx_10_13_x86_64.whl", hash = "sha256:02893d100e99e03eda1c8fd5c441d8c60103fd175728e23e431db1b589cf5ab3", size = 189108, upload-time = "2025-09-25T21:32:44.377Z" },
+    { url = "https://files.pythonhosted.org/packages/4e/78/8d08c9fb7ce09ad8c38ad533c1191cf27f7ae1effe5bb9400a46d9437fcf/pyyaml-6.0.3-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:c1ff362665ae507275af2853520967820d9124984e0f7466736aea23d8611fba", size = 183641, upload-time = "2025-09-25T21:32:45.407Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/5b/3babb19104a46945cf816d047db2788bcaf8c94527a805610b0289a01c6b/pyyaml-6.0.3-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:6adc77889b628398debc7b65c073bcb99c4a0237b248cacaf3fe8a557563ef6c", size = 831901, upload-time = "2025-09-25T21:32:48.83Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/cc/dff0684d8dc44da4d22a13f35f073d558c268780ce3c6ba1b87055bb0b87/pyyaml-6.0.3-cp314-cp314t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:a80cb027f6b349846a3bf6d73b5e95e782175e52f22108cfa17876aaeff93702", size = 861132, upload-time = "2025-09-25T21:32:50.149Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/5e/f77dc6b9036943e285ba76b49e118d9ea929885becb0a29ba8a7c75e29fe/pyyaml-6.0.3-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:00c4bdeba853cc34e7dd471f16b4114f4162dc03e6b7afcc2128711f0eca823c", size = 839261, upload-time = "2025-09-25T21:32:51.808Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/88/a9db1376aa2a228197c58b37302f284b5617f56a5d959fd1763fb1675ce6/pyyaml-6.0.3-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:66e1674c3ef6f541c35191caae2d429b967b99e02040f5ba928632d9a7f0f065", size = 805272, upload-time = "2025-09-25T21:32:52.941Z" },
+    { url = "https://files.pythonhosted.org/packages/da/92/1446574745d74df0c92e6aa4a7b0b3130706a4142b2d1a5869f2eaa423c6/pyyaml-6.0.3-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:16249ee61e95f858e83976573de0f5b2893b3677ba71c9dd36b9cf8be9ac6d65", size = 829923, upload-time = "2025-09-25T21:32:54.537Z" },
+    { url = "https://files.pythonhosted.org/packages/f0/7a/1c7270340330e575b92f397352af856a8c06f230aa3e76f86b39d01b416a/pyyaml-6.0.3-cp314-cp314t-win_amd64.whl", hash = "sha256:4ad1906908f2f5ae4e5a8ddfce73c320c2a1429ec52eafd27138b7f1cbe341c9", size = 174062, upload-time = "2025-09-25T21:32:55.767Z" },
+    { url = "https://files.pythonhosted.org/packages/f1/12/de94a39c2ef588c7e6455cfbe7343d3b2dc9d6b6b2f40c4c6565744c873d/pyyaml-6.0.3-cp314-cp314t-win_arm64.whl", hash = "sha256:ebc55a14a21cb14062aa4162f906cd962b28e2e9ea38f9b4391244cd8de4ae0b", size = 149341, upload-time = "2025-09-25T21:32:56.828Z" },
+]
+
 [[package]]
 name = "requests"
 version = "2.34.2"