fix(llm): inject system prompt into user message for OpusAgent

OpusAgent topic system_prompt is used for summarization context but not enforced as a binding instruction to Claude. Prepend [SYSTEM] block to the first message of each session so Claude sees the full JSON schema and grammar specification. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
feat(llm): add session_id for retries and summarize for run tracking
2026-05-26 09:42:26 +00:00 · 2026-05-26 09:18:11 +00:00 · 2026-05-26 09:09:54 +00:00 · 2026-05-20 15:19:13 +00:00 · 2026-05-20 10:00:31 +00:00 · 2026-05-19 14:16:07 +00:00
73 changed files with 11451 additions and 508 deletions
@@ -47,3 +47,6 @@ logs/
 build/
 dist/
 *.egg
+
+# Pythagoras source PDFs (local only, not tracked)
+src/strategy_pythagoras/Pythagoras/*.pdf
@@ -12,23 +12,23 @@ git clone ssh://git@git.tielogic.xyz:222/Adriano/Multi_Swarm_Coevolutive.git

 ## Layout monorepo (uv workspace)

-Il repo è un **workspace uv** con due member packages indipendenti:
+Il repo è un **workspace uv** con due member packages indipendenti, principio "**core = framework, strategy = contenuto**":

 ```
 multi_swarm_coevolutive/             repo root (workspace coordinator)
 ├── pyproject.toml                    workspace + dev deps + ruff/mypy/pytest
-├── docker-compose.yml                strategy-crypto-paper + strategy-crypto-gui
+├── docker-compose.yml                3 servizi su immagine condivisa
 ├── Dockerfile                        immagine multi-swarm-coevolutive:dev
 ├── uv.lock                           lock unico del workspace
 ├── data/, series/, state/            cache OHLCV + DB (runtime, gitignored)
 ├── scripts/                          CLI entrypoints (run_phase1, run_paper_trading, ...)
 └── src/
    ├── multi_swarm_core/             WORKSPACE MEMBER (wheel: multi-swarm-core)
-    │   ├── pyproject.toml             deps: pandas, numpy, openai, pydantic, ...
+    │   ├── pyproject.toml             core deps (pandas, numpy, openai, pydantic, nicegui, ...)
    │   ├── multi_swarm_core/          GA + genome + protocol + backtest + cerbero +
    │   │                              data + llm + agents + ga + orchestrator +
-    │   │                              metrics + persistence + config
-    │   ├── tests/                     unit + integration (182 test)
+    │   │                              metrics + persistence + config + dashboard (GA-only)
+    │   ├── tests/                     unit + integration
    │   └── docs/                      design/ + decisions/ + reports/
    │
    └── strategy_crypto/              WORKSPACE MEMBER (wheel: strategy-crypto)
@@ -36,27 +36,73 @@ multi_swarm_coevolutive/             repo root (workspace coordinator)
        ├── README.md                  overview strategia + pattern per nuove strategie
        ├── strategy_crypto/
        │   ├── backend/               paper-trading (executor, portfolio, persistence, schema)
-        │   ├── frontend/               NiceGUI dashboard dual-DB
-        │   └── strategies/             JSON freezate (btc_*.json, eth_*.json)
-        └── tests/                      smoke regression (import + json + schema)
+        │   ├── frontend/              NiceGUI paper-only dashboard
+        │   ├── strategies/            JSON freezate (btc_*.json, eth_*.json)
+        │   └── prompts.json           v3.2: agent_role/pattern_guidance/instruction/
+        │                              domain_warnings/anti_patterns/output_priorities +
+        │                              7 stili cognitive (directive + focus_metrics)
+        └── tests/                     smoke regression
 ```

 **DB separati per dominio:** `state/runs.db` (GA core universale) + `state/strategy_crypto.db` (paper della strategia crypto). Pattern scala a N strategie senza naming collision.

-**Pattern N strategie future:** aggiungere `src/strategy_<asset>/` con lo stesso scheletro (`backend/`, `frontend/`, `strategies/`, `tests/`), DB dedicato `state/strategy_<asset>.db`, servizi Docker `strategy-<asset>-paper` + `strategy-<asset>-gui`, GUI su `/strategy_<asset>_gui`.
+**Pattern N strategie future:** aggiungere `src/strategy_<asset>/` con stesso scheletro (`backend/`, `frontend/`, `strategies/`, `tests/`, `prompts.json`), DB dedicato `state/strategy_<asset>.db`, servizi Docker `strategy-<asset>-paper` + `strategy-<asset>-gui`, GUI su `/strategy_<asset>_gui`. **Zero modifiche al core** richieste.
+
+## Architettura prompt (v3.2)
+
+**Compositor**: il SYSTEM prompt al LLM viene COMPOSTO at-runtime da scaffold core + contenuto strategy:
+
+```
+[1] agent_role           ← strategy (prompts.json — chi è l'agente)
+[2] cognitive_style + directive  ← genome (evoluti dal GA)
+[3] GRAMMAR_SPEC         ← core scaffold (operatori, indicatori, units rules)
+[4] pattern_guidance     ← strategy (forme di curva astratte, no indicatori prescritti)
+[5] domain_warnings      ← strategy (es. "crypto trada 24/7, NON inferire funding rate")
+[6] CONSTRAINTS          ← core scaffold (validator semantics)
+[7] anti_patterns        ← strategy (7 voci: no >4 AND, no chattering, isteresi, ecc.)
+[8] output_priorities    ← strategy (5 voci, #1 coerenza con lente cognitiva)
+[9] EXAMPLE              ← core scaffold
+```
+
+**Input USER (calcolato da `build_market_summary`):**
+- Base (5): mean, std, skew, kurt, vol_regime
+- Regime recente rolling 500 (6): autocorr_lag1 (recent + baseline), hurst, vol_percentile, seasonality (hour + dow)
+- Geometria & frattali (7): efficiency_ratio (Kaufman), tail_index (left + right Hill), structural_uptrend (HH/HL), compression, spectral_entropy, dominant_cycle (gated)
+- Feature accessibili dal genome + lookback_window
+- **Focus per la tua lente**: blocco style-aware (4 metriche prioritarie da `focus_metrics` di prompts.json)
+- Instruction finale (da strategy)
+
+**Grammar protocol JSON (8 indicatori):**
+| Indicatore | Output | Range |
+|------------|--------|-------|
+| `sma(length)` | media mobile | unità prezzo |
+| `sma_pct(length)` | (close-sma)/sma | ±0.1 frazione |
+| `rsi(length)` | oscillator | 0-100 |
+| `atr(length)` | true range | unità prezzo |
+| `atr_pct(length)` | atr/close | 0-0.1 frazione |
+| `realized_vol(window)` | std returns | 0-0.1 frazione |
+| `macd(fast,slow,signal)` | momentum | unità prezzo |
+| `macd_pct(...)` | macd/close | ±0.02 frazione |
+
+**7 stili cognitive** (in `prompts.json`, editable): physicist, biologist, historian, meteorologist, engineer, military_strategist, psychologist. Ognuno con directive 800-950 char, ASCII-strict, archetipo dominante + lookback consigliato + 4 focus_metrics.

 ## Stato del progetto

-**Phase 3 (paper-trading forward-test) in corso** dal 13 maggio 2026 su VPS. Runner `scripts/run_paper_trading.py` long-running in Docker, dashboard NiceGUI su `https://swarm.tielogic.xyz/strategy_crypto_gui/`. Due strategie freezate:
+**Phase 3 (paper-trading forward-test) in corso** dal 13 maggio 2026 su VPS. Runner `scripts/run_paper_trading.py` long-running in Docker, dashboard NiceGUI su `https://swarm.tielogic.xyz/strategy_crypto_gui/`. Due strategie freezate in `src/strategy_crypto/strategy_crypto/strategies/`:

- `strategy_crypto/strategies/btc_fb63e851.json` — BTC-PERPETUAL, RSI estremi + ATR vs SMA + filtro orario 9-17 (Sharpe OOS +0,26 su 7,33 anni).
- `strategy_crypto/strategies/eth_facd6af85d5d.json` — ETH-PERPETUAL, regime-based (Sharpe OOS +0,19 su 6,75 anni).
+- `btc_fb63e851.json` — BTC-PERPETUAL, RSI estremi + ATR vs SMA + filtro orario 9-17 (Sharpe OOS +0,26 su 7,33 anni).
+- `eth_facd6af85d5d.json` — ETH-PERPETUAL, regime-based (con `atr_pct` post-fix bug unità).

-Phase 1 → 2.7 chiuse (30 run GA, $3.74 cumulato LLM).
+Phase 1 → 2.7 chiuse (30 run GA, $3.74 cumulato LLM). Sessione refactor 15 maggio 2026:
+- Split repo invertito, monorepo unificato come uv workspace
+- Family `*_pct` completa (atr_pct, sma_pct, macd_pct) per fix bug unità
+- Dashboard split: core (GA) vs strategy (paper)
+- Prompt architecture compositor + prompts.json v3.2 (vedi decision log)

- [**Stato progetto e roadmap (14 maggio 2026)**](src/multi_swarm_core/docs/reports/2026-05-14-stato-progetto-e-roadmap.md) — riepilogo fasi, decisioni, caveat, roadmap.
- Decision log: [`src/multi_swarm_core/docs/decisions/`](src/multi_swarm_core/docs/decisions/) (gate Phase 1, scelta nemotron).
- Design docs concettuali: [`src/multi_swarm_core/docs/design/`](src/multi_swarm_core/docs/design/).
+Documenti:
+- [**Stato progetto e roadmap (14 maggio 2026)**](src/multi_swarm_core/docs/reports/2026-05-14-stato-progetto-e-roadmap.md)
+- Decision log: [`src/multi_swarm_core/docs/decisions/`](src/multi_swarm_core/docs/decisions/) (gate Phase 1, nemotron, atr_pct fix)
+- Design docs concettuali: [`src/multi_swarm_core/docs/design/`](src/multi_swarm_core/docs/design/)

 Stack: Python 3.13, uv workspace, hatchling, pytest+pytest-mock+responses, openai SDK (verso OpenRouter), requests+tenacity, pandas+numpy+scipy, sqlmodel+sqlite, nicegui+plotly, yfinance.

@@ -65,7 +111,7 @@ Stack: Python 3.13, uv workspace, hatchling, pytest+pytest-mock+responses, opena
 ```bash
 uv sync                           # installa entrambi i workspace member come editable
 cp .env.example .env              # compila CERBERO_*_TOKEN e OPENROUTER_API_KEY
-uv run pytest                     # 186 test attesi (182 core + 4 smoke strategy_crypto)
+uv run pytest                     # 252 test attesi (248 core + 4 smoke strategy_crypto)
 ```

 ### Variabili .env richieste
@@ -84,10 +130,9 @@ OPENROUTER_API_KEY=<sk-or-v1-...>
 GA_DB_PATH=./state/runs.db
 STRATEGY_CRYPTO_DB_PATH=./state/strategy_crypto.db

-# Deploy Docker
+# Deploy Docker — DASHBOARD_ROOT_PATH ora per-servizio (vedi docker-compose.yml)
 DOMAIN_NAME=tielogic.xyz
 SWARM_DASHBOARD_PORT=8080
-DASHBOARD_ROOT_PATH=/strategy_crypto_gui   # subpath traefik per la dashboard
 ```

 Backcompat: `DB_PATH` legacy continua a funzionare come alias di `GA_DB_PATH`.
@@ -105,14 +150,27 @@ uv run mypy src/ scripts/
 # Smoke run (MockLLM + OHLCV sintetico, no API calls)
 uv run python scripts/smoke_run.py

-# Run reale Phase 1/2 (Cerbero + OpenRouter, ~$0.07 per run K=20 10gen)
+# Run reale Phase 1/2 (Cerbero + OpenRouter, ~$0.15-0.25 per run K=20 10gen,
+# ~$0.40-0.55 per run esteso K=40 20gen con WFA OOS).
+# Default --start ora 2018-09-01 (7.3y, copre bear+halving+covid+ATH+winter+ETF).
 uv run python scripts/run_phase1.py \
  --name run-XXX \
  --exchange deribit --symbol BTC-PERPETUAL --timeframe 1h \
-  --start 2024-01-01T00:00:00+00:00 \
-  --end 2026-01-01T00:00:00+00:00 \
  --population-size 20 --n-generations 10 \
-  --prompt-mutation-weight 0.30 --fitness-v2
+  --prompt-mutation-weight 0.30 --fitness-v2 \
+  --llm-concurrency 8                          # 5-8x speedup wall time (default 1)
+# fitness-v2 hardened: hard-kill su {no_trades, degenerate, undertrading,
+# fees_eat_alpha, negative_net_pnl}. Override via --fitness-hard-kill CSV.
+# Default --prompt-library: importlib.resources del package strategy_crypto/prompts.json
+
+# Multi-fold validation di un run esistente (anti-overfit, 7y expanding-window)
+uv run python scripts/validate_run.py \
+  --run-id <run_id> \
+  --top-k 10 --n-folds 4 --train-ratio 0.5 \
+  --start 2018-09-01T00:00:00+00:00 --end 2026-01-01T00:00:00+00:00 \
+  --fitness-v2 \
+  --output-json state/validation-XXX.json
+# Ranking per "robust_score" = min(fitness_oos) su tutti i fold.

 # Backtest standalone di una strategia JSON
 uv run python scripts/backtest_strategy.py \
@@ -123,32 +181,50 @@ uv run python scripts/backtest_strategy.py \
 uv run python scripts/run_paper_trading.py \
  --name phase3-papertrade-XXX \
  --initial-capital 1000 --poll-seconds 300
-# Default --strategies-dir: importlib.resources del package strategy_crypto

-# Dashboard NiceGUI locale
-uv run python -m strategy_crypto.frontend.nicegui_app
-# → http://localhost:8080 (env SWARM_DASHBOARD_PORT)
+# Dashboard NiceGUI locale (2 distinte)
+uv run python -m multi_swarm_core.dashboard.nicegui_app   # GA core (/, /convergence, /genomes)
+uv run python -m strategy_crypto.frontend.nicegui_app     # Strategy crypto (/ paper)
 ```

-## Dashboard
+## Performance & Validation

-NiceGUI dashboard (dark palette) — **dual-DB reader** (GA + paper):
+**Backtest engine vettorializzato** (`backtest/engine.py`): rimosso il loop `pd.iterrows()` a favore di state machine numpy. Speedup misurati:

+| Dataset | Before (iterrows) | After (vectorized) | Speedup |
+|---------|-------------------|--------------------|---------|
+| 2 anni (17545 bar) | 470 ms | **28 ms** | **16.8×** |
+| 7 anni (64297 bar) | 1744 ms | **154 ms** | **11.3×** |
+
+Equivalenza numerica garantita: 5 parity test parametrici vs. reference implementation legacy (`test_backtest_engine_vectorized.py`).
+
+**Parallel propose LLM** (`orchestrator/run.py`): `--llm-concurrency N` lancia N chiamate `hypothesis_agent.propose()` concorrenti per generazione tramite `ThreadPoolExecutor`. OpenRouter qwen-2.5 regge 6-10 concorrenti senza rate-limit. Default 1 = backward-compat.
+
+**Multi-fold validation tool** (`scripts/validate_run.py`): qualunque run completato puo' essere rivalutato post-hoc su N fold expanding-window di un dataset esteso (tipicamente 7 anni). Vital per evitare il single-hold-out overfit: il GA puo' selezionare un genome con `fitness_is` alta che collassa OOS (osservato su `phase1-extended-001`: elite IS Sharpe 1.93, OOS Sharpe -1.00). Ranking finale per `robust_score = min(fitness_oos)`. Output JSON con per-fold breakdown + aggregati mean/min/std.
+
+## Dashboard (split core + strategy)
+
+Due NiceGUI dashboard distinte (dark palette, palette neon):
+
+**Core GA** — `multi_swarm_core.dashboard.nicegui_app` → `https://swarm.tielogic.xyz/multi_swarm_core_gui/`:
 - **Overview** (`/`): lista runs GA, costo cumulato, metriche aggregate evaluations.
 - **GA Convergence** (`/convergence`): fitness median/max/p90 per generazione + entropy.
 - **Genomes** (`/genomes`): top-K ordinati per fitness, ispezione system_prompt + JSON strategy.
- **Paper** (`/paper`): forward-test live con equity curve, posizioni aperte, trade list, tick log.

-In produzione su `https://swarm.tielogic.xyz/strategy_crypto_gui/` (subpath gestito via `DASHBOARD_ROOT_PATH` + Traefik PathPrefix). La root del dominio resta libera per future GUI di altre strategie.
+**Strategy crypto** — `strategy_crypto.frontend.nicegui_app` → `https://swarm.tielogic.xyz/strategy_crypto_gui/`:
+- **Paper** (`/`): forward-test live con equity curve, posizioni aperte, trade list, tick log.
+
+In produzione subpath gestiti via `DASHBOARD_ROOT_PATH` (per-servizio) + Traefik `replacepathregex` (NB: NON `stripprefix`, vedi sezione Deploy). La root del dominio resta libera per future GUI di altre strategie.

 ## Deploy

-`docker-compose.yml` definisce due servizi su immagine `multi-swarm-coevolutive:dev`:
+`docker-compose.yml` definisce 3 servizi su immagine condivisa `multi-swarm-coevolutive:dev`:

 - **`strategy-crypto-paper`** — runner `scripts/run_paper_trading.py` long-running.
- **`strategy-crypto-gui`** — NiceGUI dashboard dietro Traefik su `https://swarm.${DOMAIN_NAME}/strategy_crypto_gui/`.
+- **`strategy-crypto-gui`** — NiceGUI paper dashboard su `https://swarm.${DOMAIN_NAME}/strategy_crypto_gui/`.
+- **`multi-swarm-core-gui`** — NiceGUI GA dashboard su `https://swarm.${DOMAIN_NAME}/multi_swarm_core_gui/`.

-Persistenza via bind mount: `./data/`, `./series/`, `./state/`. Le strategie JSON sono bind-mounted in read-only dal package: `./src/strategy_crypto/strategy_crypto/strategies/`.
+Persistenza via bind mount: `./data/`, `./series/`, `./state/`. Strategie JSON bind-mounted in read-only dal package: `./src/strategy_crypto/strategy_crypto/strategies/`.

 ```bash
 docker compose up -d --build
@@ -158,12 +234,16 @@ docker compose ps

 Note operative:

- Le bind-mount dir devono essere `chown 1000:1000` (uid utente `app` nel container).
+- Bind-mount dir devono essere `chown 1000:1000` (uid utente `app` nel container). **Anche `src/strategy_crypto/strategy_crypto/strategies/`** (creata da `git mv`, default `root:root`).
 - Override del command paper via env (`PAPER_RUN_NAME`, `PAPER_INITIAL_CAPITAL`, ecc.).
 - `SWARM_DASHBOARD_PORT` controlla la porta interna del container (Traefik fa il TLS).
+- **Traefik subpath**: usa `replacepathregex` middleware (NON `stripprefix`) per evitare doppio root_path (uvicorn legge `X-Forwarded-Prefix` da stripprefix + applica `root_path` di NiceGUI = doppio prefix). Vedi commit `436613b`.
+- Dopo cambio label Traefik: `docker restart traefik-traefik-1` per forzare refresh discovery.

 ## Sviluppo

 Conventional commits con prefix `feat:` `fix:` `chore:` `docs:` `refactor:` `test:`. Body italiano. Footer `Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>` su ogni commit collaborativo.

 Branch attuale: `main`. Workspace single-repo, monorepo unificato dal 15 maggio 2026 (split temporaneo monorepo→figlio invertito, vedi tag `v0.1.0-pre-split` come bookmark).
+
+**Modificare il prompt LLM** senza toccare codice: edita `src/strategy_crypto/strategy_crypto/prompts.json`. Schema documentato in `_design_invariants` del JSON stesso. I 3 regression guard test (`test_strategy_crypto_directives_ascii_safe`, `..._have_archetype_marker`, `..._have_lookback_hint`) bloccano regressioni accidentali sulle invarianti di design.
@@ -23,7 +23,7 @@
 #   ./state   contiene runs.db (GA) + strategy_crypto.db (paper) + WAL/SHM
 #   ./src/strategy_crypto/strategy_crypto/strategies  JSON freezate (ro)
 #
-# Secrets (token Cerbero + OpenRouter): caricati da .env via env_file.
+# Secrets (token Cerbero + OpusAgent): caricati da .env via env_file.

 networks:
  traefik:
@@ -88,8 +88,8 @@ services:
      <<: *swarm-env
      DASHBOARD_ROOT_PATH: /strategy_crypto_gui
    volumes:
-      # Dashboard legge solo strategy_crypto.db: state/ in read-only (WAL: vedi nota)
-      - ./state:/app/state:ro
+      # RW richiesto: SQLite WAL mode richiede write-access dal reader per SHM.
+      - ./state:/app/state
    entrypoint:
      - python
      - -m
@@ -0,0 +1,343 @@
+# strategy_pythagoras — Design
+
+**Data:** 2026-05-19
+**Autore design:** Adriano Dal Pastro (con Claude/brainstorming)
+**Stato:** Design approvato per parti 1-4, in attesa di review utente sulla spec consolidata
+**Audience:** implementatori (Claude executor o umano)
+
+## 0. Riassunto esecutivo
+
+Nuovo workspace member `strategy_pythagoras` parallelo a `strategy_crypto`. Replica il pattern coevolutivo GA del monorepo applicato a un dominio diverso: la **scoperta di pattern frattali ricorrenti** sui mercati crypto secondo il framework Pythagoras-Malanga (Fourier + frattali + geometria Evideon), descritto nei due PDF in `src/strategy_pythagoras/Pythagoras/`.
+
+**Non cambia il genoma del core:** gli agenti restano `HypothesisAgentGenome` e producono `Strategy JSON` nella stessa DSL S-expression. Cambiano:
+- 7 stili cognitivi Pythagoras-themed (in `prompts.json`)
+- 3 nuovi indicatori candle (`candle_pattern`, `pythagorean_ratio`, `fractal_mirror`)
+- Fitness con bonus di asset-invariance BTC↔ETH
+- Output: strategie JSON freezate + dashboard NiceGUI + report markdown di analisi cross-book
+
+**Deliverable di chiusura task:** scaffolding completo + 1 GA run short (smoke test) + `docs/analysis_first_run.md` con sintesi numerica e top winners.
+
+---
+
+## 1. Architettura
+
+### Layout package
+
+```
+src/strategy_pythagoras/
+├── pyproject.toml                     # workspace member, dipende da multi-swarm-core
+├── README.md
+├── tests/
+│   └── test_indicators.py             # unit-test per i 3 nuovi indicatori
+└── strategy_pythagoras/
+    ├── __init__.py
+    ├── prompts.json                   # 7 stili Pythagoras-aligned (schema v3.2 di strategy_crypto)
+    ├── indicators.py                  # candle_pattern, pythagorean_ratio, fractal_mirror
+    ├── backend/
+    │   ├── __init__.py
+    │   ├── schema.py                  # tabelle paper_trading_*
+    │   ├── executor.py                # PaperExecutor (port da strategy_crypto)
+    │   ├── portfolio.py               # Portfolio (port da strategy_crypto)
+    │   └── persistence.py             # PaperRepository
+    ├── frontend/
+    │   ├── __init__.py
+    │   ├── nicegui_app.py             # /strategy_pythagoras_gui
+    │   └── data.py                    # dual-reader: GA db + paper db + invariance metrics
+    └── strategies/                    # JSON winners shippati col package
+        └── (vuoto al t0)
+```
+
+### Workspace registration
+
+In `pyproject.toml` (root):
+```toml
+[tool.uv.workspace]
+members = ["src/multi_swarm_core", "src/strategy_crypto", "src/strategy_pythagoras"]
+
+[tool.uv.sources]
+strategy-pythagoras = { workspace = true }
+```
+
+Aggiungere `strategy-pythagoras` a `[project].dependencies` per il deployable root.
+
+### Riuso
+
+| Componente | Source | Note |
+|---|---|---|
+| GA loop, mutation, crossover | `multi_swarm_core.ga` | invariato |
+| Protocol parser/validator/compiler | `multi_swarm_core.protocol` | esteso con 3 indicatori (vedi §2) |
+| Backtest engine | `multi_swarm_core.backtest` | invariato |
+| LLM / OpenRouter / Anthropic clients | `multi_swarm_core.llm` | invariato |
+| PaperExecutor + Portfolio | `strategy_crypto.backend` | **port** (non import), per isolamento DB |
+| NiceGUI dashboard shell | `strategy_crypto.frontend` | **port** + adatta tabs |
+
+### Persistence
+
+```
+state/strategy_pythagoras.db          # GA: genomi, generazioni, fitness history
+state/strategy_pythagoras_paper.db    # paper-trading post-deploy
+strategy_pythagoras/strategies/       # JSON winners shippati
+docs/analysis_first_run.md            # report cross-book
+docs/analysis_runs/<run-id>/          # per-run dump
+```
+
+Env vars:
+- `STRATEGY_PYTHAGORAS_DB_PATH` (default `state/strategy_pythagoras.db`)
+- `STRATEGY_PYTHAGORAS_PAPER_DB_PATH` (default `state/strategy_pythagoras_paper.db`)
+- `GA_INVARIANCE_ALPHA` (default 0.3)
+- `GA_INVARIANCE_TOLERANCE_BARS` (default 36 = ±3h su 5m TF)
+
+### Routing GUI
+
+`/strategy_pythagoras_gui` (allinea a [[gui_subpath_routing]] in memory: ogni asset GUI su subpath, root dominio libera).
+
+### Docker (Phase 2, fuori scope di questa spec)
+
+Servizi paralleli a quelli di strategy_crypto:
+- `strategy-pythagoras-paper` (runner)
+- `strategy-pythagoras-gui` (dashboard)
+Rete Traefik external, bind mount uid 1000 (vedi [[production_deployment]]).
+
+---
+
+## 2. Genoma e DSL
+
+### Il genoma non cambia
+
+`HypothesisAgentGenome` di `multi_swarm_core.genome.hypothesis` resta identico:
+- `system_prompt: str`
+- `feature_access: list[str]`
+- `temperature: float`
+- `top_p: float`
+- `model_tier: ModelTier`
+- `lookback_window: int` — vincolo 12 ≤ lw ≤ 200
+- `cognitive_style: str` — uno dei 7 nuovi stili
+- `parent_ids, generation, id` — invariati
+
+### 3 nuovi indicatori (`strategy_pythagoras/indicators.py`)
+
+| Nome | Params (JSON) | Output | Semantica operativa |
+|---|---|---|---|
+| `candle_pattern` | `[seq_str]` es. `"UDU"`, `"UUD0U"` | 1.0 se le ultime k=len(seq_str) candele matchano la sequenza, 0.0 altrimenti | `"U"` = close>open; `"D"` = close<open; `"0"` = `abs(close-open)/open < 0.001` |
+| `pythagorean_ratio` | `[lookback: int]` | float = `max(close[-lookback:]) / min(close[-lookback:])` | Ratio prezzo, da confrontare con literal vicini a φ=1.618, π=3.1416, √2=1.414, e=2.718 |
+| `fractal_mirror` | `[k: int, axis: str]` axis ∈ `{"h","v"}` | float ∈ [-1, +1] | Correlazione di Pearson tra ultime k candele e loro mirror: `"h"` = mirror tempo (inversione sequenza); `"v"` = mirror prezzo (1 - close/max) |
+
+Vincoli del compiler:
+- `candle_pattern`: `len(seq_str)` ∈ [3, 12], simboli ∈ `{U,D,0}`
+- `pythagorean_ratio`: `lookback` ∈ [12, 200]
+- `fractal_mirror`: `k` ∈ [3, 12], `axis` ∈ `{"h","v"}`
+
+Tutti e 3 vanno aggiunti a `KNOWN_INDICATORS` in `multi_swarm_core.protocol.grammar`.
+
+### Esempio strategy JSON tipica
+
+```json
+{
+  "rules": [
+    {
+      "condition": {
+        "op": "and",
+        "args": [
+          {"op": "eq", "args": [
+            {"kind": "indicator", "name": "candle_pattern", "params": ["UDDU"]},
+            {"kind": "literal", "value": 1.0}
+          ]},
+          {"op": "gt", "args": [
+            {"kind": "indicator", "name": "pythagorean_ratio", "params": [89]},
+            {"kind": "literal", "value": 1.618}
+          ]}
+        ]
+      },
+      "action": "entry-long"
+    },
+    {
+      "condition": {
+        "op": "or", "args": [
+          {"op": "gt", "args": [
+            {"kind": "indicator", "name": "pythagorean_ratio", "params": [89]},
+            {"kind": "literal", "value": 2.618}
+          ]},
+          {"op": "crossunder", "args": [
+            {"kind": "indicator", "name": "fractal_mirror", "params": [12, "h"]},
+            {"kind": "literal", "value": 0.0}
+          ]}
+        ]
+      },
+      "action": "exit"
+    }
+  ]
+}
+```
+
+### Vincoli protettivi (anti-overfitting)
+
+- `lookback_window ≤ 200`
+- `candle_pattern` seq length ∈ [3, 12] (range 1+2 del Libro dei Frattali)
+- `time_in_market` monitorato come metric (red flag [[selectivity_red_flag]]); non hard-gate al primo run
+- Letterali con `pythagorean_ratio`: max 4 decimali (no `1.6180339`)
+- Max 4 condizioni in AND per regola (eredita da prompts.json)
+
+---
+
+## 3. Stili cognitivi (`strategy_pythagoras/prompts.json`)
+
+### Schema
+
+Schema v3.2 identico a `strategy_crypto/prompts.json` (campi `_schema`, `_changelog`, `_design_invariants`, `agent_role`, `pattern_guidance`, `instruction`, `domain_warnings`, `anti_patterns`, `output_priorities`, `styles`).
+
+### `agent_role`
+
+> Sei un agente di un sistema swarm coevolutivo che cerca pattern frattali ricorrenti sui mercati crypto secondo il framework Pythagoras-Malanga (frattali H-C, trasformata di Fourier, geometria Evideon). Sei parte di una popolazione che esplora collettivamente lo spazio dei pattern: la diversità delle ipotesi è un asset critico. Preferisci esplorare strutture meno ovvie per la tua lente cognitiva. La strategia che produci deve essere riconoscibile come emanata dal tuo stile.
+
+### `pattern_guidance` (specifico Pythagoras)
+
+- Sequenze candle 3-12 lunghezza in alfabeto `{U,D,0}` via `candle_pattern`
+- Mirror H/V come operatori di proiezione via `fractal_mirror`
+- Ratios di prezzo vicini a φ=1.618, 1/φ=0.618, √2=1.414, π/2=1.571, e/2=1.359 entro tolleranza 0.5%
+- Pattern composti: pattern lunghi (6-12) come concatenazione di pattern corti (Pattern 13 = Pattern 3 + Pattern 11, p. 53 del paper)
+- Cicli ricorrenti: stesso pattern firato a distanze regolari (Poincaré)
+
+### `domain_warnings`
+
+- Crypto 24/7, no CME gap
+- I "numeri sacri" (Solfeggio 396-852 Hz, 137.0359, etc.) sono prior teorici, NON leggi: usali come scale candidate, non come dogma
+- Il paper Pythagoras è esplicitamente non-falsificabile (cita "consapevolezza del trader" come jolly per il fallimento): il backtest è l'unico arbitro
+- `time_in_market > 80%` red flag (leveraged B&H camuffato)
+- Tolleranza ±3h del paper → su 5m TF = ±36 barre
+
+### `anti_patterns`
+
+- Sequenza `candle_pattern` con `len > 7` simboli vincolati → overfitting
+- `pythagorean_ratio` con tolleranza > 2% sui literal → numerologia spuriosa
+- `fractal_mirror` con `k == lookback_window` → tautologico
+- Letterali con più di 4 decimali
+- Più di 4 condizioni in AND
+- Crossover tra indicatori dello stesso tipo con lookback vicini → chattering
+
+### `output_priorities`
+
+1. **Coerenza con lente cognitiva** (es. `pythagorean` usa ratios, `candle_grammarian` usa sequenze esplicite)
+2. **Asset-invariance** (segnali che attivano sia su BTC che su ETH entro ±36 barre)
+3. **Selettività** (poche entry forti)
+4. **Composizionalità** (pattern lunghi come somma di corti)
+5. **Robustezza vs random baseline** (3σ richiesto da `skeptic_quant`)
+
+### I 7 stili (`styles`)
+
+Ogni stile mantiene shape v3.2 (directive 800-950 char ASCII-safe, `focus_metrics` 4 voci, ultimo periodo = "Archetipo dominante: X.", lookback range esplicito):
+
+| `cognitive_style` | Archetipo / Metafora ancorante | `focus_metrics` | Lookback consigliato |
+|---|---|---|---|
+| `pythagorean` | Armonia di ratios sacri (φ, π, √2) | `pythagorean_ratio`, `candle_pattern`, `sma_pct`, `realized_vol` | 89-144 |
+| `fractal_geometer` | Autosimilarità: pattern di 3 candele si ripetono dilatati a 6, 12 | `candle_pattern`, `fractal_mirror`, `atr_pct`, `pythagorean_ratio` | 48-144 |
+| `fourier_analyst` | Somma di seni: frequenze ricorrenti dominanti | `sma_pct`, `realized_vol`, `candle_pattern`, `atr` | 60-200 |
+| `evideonic_projector` | Presente = passato proiettato via mirror H+V e scale | `fractal_mirror`, `pythagorean_ratio`, `candle_pattern`, `sma_pct` | 24-96 |
+| `candle_grammarian` | Lingua di 3 simboli (U,D,0); parole 3-12 lettere | `candle_pattern`, `volume`, `atr`, `realized_vol` | 12-48 |
+| `recurrence_theorist` | Per Poincaré, eventi tornano: cerca pattern di oggi che firarono ieri | `candle_pattern`, `fractal_mirror`, `pythagorean_ratio`, `sma_pct` | 100-200 |
+| `skeptic_quant` | Anticorpo all'unfalsifiability: solo edge 3σ vs random | `realized_vol`, `atr_pct`, `sma_pct`, `candle_pattern` | 60-150 |
+
+Lo `skeptic_quant` è importante: la sua directive richiede esplicitamente che la strategia sia testabile e che il fitness sia confrontato contro random baseline.
+
+### `_design_invariants`
+
+Stessa filosofia di v3.2 di strategy_crypto:
+- ASCII-safe (no Unicode oltre U+007F nelle directive)
+- Ogni directive chiude con `Archetipo dominante: <metafora>.`
+- Ogni directive ha range lookback numerico esplicito
+- Prima frase: `Il mercato e ...`
+- Lunghezza 800-950 char
+
+---
+
+## 4. Fitness, run GA short, deliverable analisi
+
+### Fitness con bonus invariance
+
+```
+fitness(genome) = mean(sharpe_BTC, sharpe_ETH) × (1 + α × invariance_score)
+
+invariance_score = corr_signal(entries_BTC, entries_ETH, tolerance_bars=GA_INVARIANCE_TOLERANCE_BARS)
+                 ∈ [0, 1]
+α = GA_INVARIANCE_ALPHA (default 0.3)
+```
+
+`corr_signal` = frazione di entries su BTC che hanno una entry corrispondente su ETH entro ±36 barre (=±3h su 5m TF).
+
+Implementazione come callback al GA in `multi_swarm_core.ga`, registrata da `strategy_pythagoras` al startup. Il core non sa nulla di Pythagoras: riceve solo la callback.
+
+### GA run short (smoke test)
+
+| Parametro | Valore | Note |
+|---|---|---|
+| `population_size` | 20 | minimo per 7 stili (~3 per stile) |
+| `generations` | 5 | smoke test, non training |
+| `elite_fraction` | 0.2 | top-4 sopravvivono |
+| `mutation_rate` | 0.3 | invariato vs strategy_crypto |
+| `crossover_rate` | 0.5 | invariato |
+| `model_tier` distribuzione | 70% C (qwen-2.5-72b), 30% B (sonnet) | rispetta [[model_qwen_dependency]] |
+| `dataset` | BTC 5m + ETH 5m da `strategy_crypto/series/` | riusa serie esistenti |
+| `train_window` | 2024-07 → 2024-12 | copre le date Pythagoras (lug-ago 2024) |
+| `test_window` | 2025-01 (1 mese) | hold-out per validare invariance |
+| `name` | `pythagoras-smoke-001` | run id |
+
+Lo smoke test verifica:
+1. Workspace member installato in venv (`uv sync` + `uv run python -c "import strategy_pythagoras"`)
+2. I 3 nuovi indicatori registrati nel grammar e compilabili
+3. `prompts.json` caricato, 7 stili producono genomi distinti (no collisioni di id)
+4. Bonus invariance impatta fitness (verifica via log)
+5. JSON winners atterrano in `strategy_pythagoras/strategies/`
+6. Dashboard NiceGUI si avvia e legge i due DB
+
+### Deliverable analisi cross-book
+
+`docs/analysis_first_run.md` con:
+
+1. **Sintesi numerica dei riassunti** — riferimenti a `src/strategy_pythagoras/Pythagoras/Pythagoras_Trading_Prediction.summary.md` e `Libro_frattali.summary.md`
+2. **Top-5 winners** — id, cognitive_style, fitness, sharpe_BTC, sharpe_ETH, invariance_score
+3. **Pattern frattali emersi** — dump dei `candle_pattern` seq usate, conteggio per stile, % sovrapposizione con spazio teorico dei 57 pattern del Libro
+4. **Ratios di prezzo emersi** — distribuzione literal usati con `pythagorean_ratio`, distanza dai numeri universali (φ/π/√2/Solfeggio)
+5. **Cross-asset invariance osservata** — istogramma di `corr_signal` per top genomi
+6. **Conclusione onesta** — confronto vs random baseline, quanti winners superano sharpe>1.0 su test + invariance>0.3, cosa il framework Pythagoras predice e cosa NON regge al backtest
+
+Niente "consapevolezza" come jolly. Solo numeri.
+
+### Dashboard NiceGUI
+
+`/strategy_pythagoras_gui`:
+- Tab **Genomes** — winners con stile/sharpe/invariance, click per drill su rules
+- Tab **Patterns** — heatmap delle sequenze `candle_pattern` emerse, frequenza per stile
+- Tab **Ratios** — istogramma literal vicini a costanti universali, bins centrati su φ, π, √2, ecc.
+- Tab **Invariance** — scatter sharpe_BTC vs sharpe_ETH per ogni winner
+
+---
+
+## 5. Out-of-scope (esplicito)
+
+- **Asset oltre BTC/ETH** (Oro/Argento del paper): non in primo run. Estensione futura.
+- **Range candele oltre 12**: range 3-5 del Libro (12-56 candele). Estensione futura quando lo smoke test conferma stabilità.
+- **Live trading reale**: solo paper-trading via stesso pattern di strategy_crypto.
+- **OCR/Vision sulle figure del Libro dei Frattali**: esplicitamente ESCLUSO da request utente ("senza passare alle immagini").
+- **Modifica del genoma del core**: nessuna modifica a `HypothesisAgentGenome`. Solo extension del grammar (3 indicatori).
+- **Riferimenti pseudoscientifici operativizzati come legge**: i numeri sacri/Solfeggio sono prior teorici per literal candidati, non vincoli rigidi.
+
+---
+
+## 6. Criteri di accettazione
+
+- [ ] `uv sync` riesce dalla root con `strategy_pythagoras` come member
+- [ ] `uv run python -c "from strategy_pythagoras.indicators import candle_pattern, pythagorean_ratio, fractal_mirror"` non solleva
+- [ ] `pytest src/strategy_pythagoras/tests/` verde (almeno unit-test per i 3 indicatori)
+- [ ] GA short run `pythagoras-smoke-001` completa 5 generazioni senza errori
+- [ ] Almeno 1 winner con fitness > 0 e `cognitive_style` ∈ {7 stili Pythagoras}
+- [ ] Dashboard avvia su `http://localhost:PORT/strategy_pythagoras_gui` e mostra winners
+- [ ] `docs/analysis_first_run.md` esiste e contiene tutte le sezioni elencate in §4
+
+---
+
+## 7. Riferimenti
+
+- [Pythagoras Trading Prediction — riassunto](../../../src/strategy_pythagoras/Pythagoras/Pythagoras_Trading_Prediction.summary.md)
+- [Libro dei Frattali — riassunto](../../../src/strategy_pythagoras/Pythagoras/Libro_frattali.summary.md)
+- Memory: [[monorepo_uv_workspace]], [[gui_subpath_routing]], [[ownership_per_modulo]], [[production_deployment]], [[model_qwen_dependency]], [[selectivity_red_flag]]
+- Template: `src/strategy_crypto/` (paper-trading + GUI), `src/strategy_crypto/strategy_crypto/prompts.json` (schema v3.2)
+- Core: `src/multi_swarm_core/multi_swarm_core/genome/hypothesis.py`, `multi_swarm_core/protocol/grammar.py`
@@ -10,14 +10,16 @@ requires-python = ">=3.13"
 dependencies = [
    "multi-swarm-core",
    "strategy-crypto",
+    "strategy-pythagoras",
 ]

 [tool.uv.workspace]
-members = ["src/multi_swarm_core", "src/strategy_crypto"]
+members = ["src/multi_swarm_core", "src/strategy_crypto", "src/strategy_pythagoras"]

 [tool.uv.sources]
 multi-swarm-core = { workspace = true }
 strategy-crypto = { workspace = true }
+strategy-pythagoras = { workspace = true }

 [dependency-groups]
 dev = [
@@ -42,7 +44,7 @@ python_version = "3.13"
 strict = true

 [tool.pytest.ini_options]
-testpaths = ["src/multi_swarm_core/tests", "src/strategy_crypto/tests"]
+testpaths = ["src/multi_swarm_core/tests", "src/strategy_crypto/tests", "src/strategy_pythagoras/tests"]
 addopts = "-v --tb=short --import-mode=importlib"
 markers = [
    "integration: tests that require external services (Cerbero, LLM API)",
@@ -0,0 +1,125 @@
+"""Analisi per-trade dei top-K candidate del run BTC.
+
+Per ciascun genome top-K, ri-esegue il backtest su ogni fold WFA e raccoglie:
+- n_trades, n_wins, n_losses, win_rate
+- max_drawdown
+- return, sharpe
+- list trade pnl summary
+
+Output stampato a stdout, non scrive su disco.
+"""
+from __future__ import annotations
+
+import argparse
+from datetime import datetime
+
+import pandas as pd  # type: ignore[import-untyped]
+
+from multi_swarm_core.agents.hypothesis import _try_parse
+from multi_swarm_core.backtest.engine import BacktestEngine
+from multi_swarm_core.cerbero.client import CerberoClient
+from multi_swarm_core.config import load_settings
+from multi_swarm_core.data.cerbero_ohlcv import CerberoOHLCVLoader, OHLCVRequest
+from multi_swarm_core.data.splits import expanding_walk_forward
+from multi_swarm_core.metrics.basic import max_drawdown, sharpe_ratio, total_return
+from multi_swarm_core.persistence.repository import Repository
+from multi_swarm_core.protocol.compiler import compile_strategy
+
+
+def main() -> None:
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--run-id", required=True)
+    ap.add_argument("--top-k", type=int, default=10)
+    ap.add_argument("--n-folds", type=int, default=4)
+    ap.add_argument("--train-ratio", type=float, default=0.5)
+    ap.add_argument("--symbol", default="BTC-PERPETUAL")
+    ap.add_argument("--timeframe", default="1h")
+    ap.add_argument("--start", default="2018-09-01T00:00:00+00:00")
+    ap.add_argument("--end", default="2026-01-01T00:00:00+00:00")
+    ap.add_argument("--fees-bp", type=float, default=5.0)
+    args = ap.parse_args()
+
+    settings = load_settings()
+    repo = Repository(settings.ga_db_path)
+    repo.init_schema()
+    all_evals = repo.list_evaluations(args.run_id)
+    parseable = [
+        e for e in all_evals
+        if e.get("raw_text") and not e.get("parse_error") and e["fitness"] > 0
+    ]
+    parseable.sort(key=lambda e: e["fitness"], reverse=True)
+    seen: set[str] = set()
+    top: list[dict] = []
+    for e in parseable:
+        if e["genome_id"] in seen:
+            continue
+        seen.add(e["genome_id"])
+        top.append(e)
+        if len(top) >= args.top_k:
+            break
+
+    token = (
+        settings.cerbero_mainnet_token.get_secret_value()
+        if settings.cerbero_mainnet_token
+        else settings.cerbero_testnet_token.get_secret_value()
+    )
+    cerbero = CerberoClient(
+        base_url=settings.cerbero_base_url,
+        token=token,
+        bot_tag=settings.cerbero_bot_tag,
+    )
+    loader = CerberoOHLCVLoader(client=cerbero, cache_dir=settings.series_dir)
+    ohlcv = loader.load(OHLCVRequest(
+        symbol=args.symbol,
+        timeframe=args.timeframe,
+        start=datetime.fromisoformat(args.start),
+        end=datetime.fromisoformat(args.end),
+    ))
+    splits = expanding_walk_forward(ohlcv.index, train_ratio=args.train_ratio, n_folds=args.n_folds)
+    engine = BacktestEngine(fees_bp=args.fees_bp)
+
+    print(f"\n{'=' * 110}")
+    print(f"PER-TRADE ANALYSIS — top-{len(top)} genomes × {len(splits)} folds")
+    print(f"{'=' * 110}")
+
+    for ev in top:
+        strat, err = _try_parse(ev["raw_text"] or "")
+        if strat is None:
+            print(f"\n>>> {ev['genome_id'][:16]} — parse error: {err}")
+            continue
+
+        print(f"\n>>> {ev['genome_id']} (fit_IS={ev['fitness']:.4f}, sharpe_IS={ev['sharpe']:.3f})")
+        print(f"{'fold':<5} {'period':<26} {'trades':>7} {'wins':>5} {'losses':>7} {'win%':>6} {'avg_w':>9} {'avg_l':>9} {'ret':>7} {'maxDD':>7} {'sharpe':>7}")
+        for s in splits:
+            test_df = ohlcv.loc[s.test_idx]
+            try:
+                signal_fn = compile_strategy(strat)
+                signals = signal_fn(test_df)
+                bt = engine.run(test_df, signals)
+            except Exception as e:
+                print(f"  fold {s.fold}: error {e}")
+                continue
+            trades = bt.trades
+            n_trades = len(trades)
+            wins = [t.net_pnl for t in trades if t.net_pnl > 0]
+            losses = [t.net_pnl for t in trades if t.net_pnl <= 0]
+            n_wins = len(wins)
+            n_losses = len(losses)
+            win_rate = (n_wins / n_trades * 100) if n_trades else 0.0
+            avg_w = (sum(wins) / n_wins) if n_wins else 0.0
+            avg_l = (sum(losses) / n_losses) if n_losses else 0.0
+            # Normalize equity for DD/return
+            if n_trades > 0:
+                notional = float(test_df["close"].iloc[0])
+                equity_pos = (bt.equity_curve / notional) + 1.0
+                ret_pct = total_return(equity_pos)
+                dd = max_drawdown(equity_pos)
+                sr = sharpe_ratio(bt.returns, periods_per_year=8760)
+            else:
+                ret_pct = dd = sr = 0.0
+            period = f"{str(s.test_idx[0])[:10]}..{str(s.test_idx[-1])[:10]}"
+            print(f"{s.fold:<5} {period:<26} {n_trades:>7} {n_wins:>5} {n_losses:>7} {win_rate:>5.1f}% {avg_w:>9.1f} {avg_l:>9.1f} {ret_pct:>6.2%} {dd:>6.2%} {sr:>7.3f}")
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1,139 @@
+"""Confronto per-trade dei 4 winner cross-run (BTC/ETH × 1h/5m).
+
+Per ogni winner: ri-esegue il backtest su 4 fold WFA expanding-window e raccoglie
+trade buoni/non buoni, win-rate, avg PnL, return, max DD, Sharpe.
+"""
+from __future__ import annotations
+
+import argparse
+from datetime import datetime
+
+import pandas as pd  # type: ignore[import-untyped]
+
+from multi_swarm_core.agents.hypothesis import _try_parse
+from multi_swarm_core.backtest.engine import BacktestEngine
+from multi_swarm_core.cerbero.client import CerberoClient
+from multi_swarm_core.config import load_settings
+from multi_swarm_core.data.cerbero_ohlcv import CerberoOHLCVLoader, OHLCVRequest
+from multi_swarm_core.data.splits import expanding_walk_forward
+from multi_swarm_core.metrics.basic import max_drawdown, sharpe_ratio, total_return
+from multi_swarm_core.persistence.repository import Repository
+from multi_swarm_core.protocol.compiler import compile_strategy
+
+
+# (run_name, genome_id, symbol, timeframe, label)
+WINNERS = [
+    ("phase1-btc-100-001", "238e481262c1594c", "BTC-PERPETUAL", "1h", "BTC 1h sharpshooter (Gen 7)"),
+    ("phase1-btc-100-001", "23a24989e2ed0f84", "BTC-PERPETUAL", "1h", "BTC 1h robust (Gen 0 elite)"),
+    ("phase1-eth-100-001", "4b45a72c13acf1d5", "ETH-PERPETUAL", "1h", "ETH 1h best-by-sharpe (killed)"),
+    ("phase1-btc-100-5m-001", "f8ca6642adf7e0cd", "BTC-PERPETUAL", "5m", "BTC 5m robust winner"),
+    ("phase1-eth-100-5m-001", "c04dff7086bb9588", "ETH-PERPETUAL", "5m", "ETH 5m OOS winner"),
+]
+
+
+def analyze_genome(run_id: str, genome_id: str, symbol: str, timeframe: str, label: str,
+                   settings, cerbero, loader) -> None:
+    repo = Repository(settings.ga_db_path)
+    repo.init_schema()
+    evs = [e for e in repo.list_evaluations(run_id) if e["genome_id"] == genome_id]
+    if not evs:
+        print(f"  no eval for {genome_id} in {run_id}")
+        return
+    ev = evs[0]
+    strat, err = _try_parse(ev.get("raw_text") or "")
+    if strat is None:
+        print(f"  parse error: {err}")
+        return
+
+    req = OHLCVRequest(
+        symbol=symbol, timeframe=timeframe,
+        start=datetime.fromisoformat("2018-09-01T00:00:00+00:00"),
+        end=datetime.fromisoformat("2026-01-01T00:00:00+00:00"),
+    )
+    ohlcv = loader.load(req)
+    splits = expanding_walk_forward(ohlcv.index, train_ratio=0.5, n_folds=4)
+    engine = BacktestEngine(fees_bp=5.0)
+
+    print(f"\n>>> {label}")
+    print(f"    {genome_id} | fit_IS={ev['fitness']:.4f} sharpe_IS={ev['sharpe']:.3f} trades_IS={ev['n_trades']}")
+    print(f"    {'fold':<5} {'period':<26} {'trades':>7} {'wins':>5} {'losses':>7} {'win%':>6} {'avg_w':>10} {'avg_l':>10} {'ret':>8} {'maxDD':>7} {'sharpe':>7}")
+
+    sum_ret = 0.0
+    sum_trades = 0
+    sum_wins = 0
+    for s in splits:
+        test_df = ohlcv.loc[s.test_idx]
+        try:
+            signal_fn = compile_strategy(strat)
+            signals = signal_fn(test_df)
+            bt = engine.run(test_df, signals)
+        except Exception as e:
+            print(f"  fold {s.fold}: error {e}")
+            continue
+
+        trades = bt.trades
+        n = len(trades)
+        wins = [t.net_pnl for t in trades if t.net_pnl > 0]
+        losses = [t.net_pnl for t in trades if t.net_pnl <= 0]
+        nw, nl = len(wins), len(losses)
+        wr = (nw / n * 100) if n else 0.0
+        aw = (sum(wins) / nw) if nw else 0.0
+        al = (sum(losses) / nl) if nl else 0.0
+        if n > 0:
+            notional = float(test_df["close"].iloc[0])
+            eq = (bt.equity_curve / notional) + 1.0
+            ret = total_return(eq)
+            dd = max_drawdown(eq)
+            sr = sharpe_ratio(bt.returns, periods_per_year=8760)
+        else:
+            ret = dd = sr = 0.0
+        period = f"{str(s.test_idx[0])[:10]}..{str(s.test_idx[-1])[:10]}"
+        print(f"    {s.fold:<5} {period:<26} {n:>7} {nw:>5} {nl:>7} {wr:>5.1f}% {aw:>10.1f} {al:>10.1f} {ret:>7.2%} {dd:>6.2%} {sr:>7.3f}")
+        sum_ret += ret
+        sum_trades += n
+        sum_wins += nw
+    overall_wr = (sum_wins / sum_trades * 100) if sum_trades else 0.0
+    print(f"    {'='*5} TOTALS:                {sum_trades:>7} {sum_wins:>5} {sum_trades-sum_wins:>7} {overall_wr:>5.1f}%  cum_ret={sum_ret:+.2%}")
+
+
+def main() -> None:
+    settings = load_settings()
+    repo = Repository(settings.ga_db_path)
+    repo.init_schema()
+    name_to_id: dict[str, str] = {}
+    for w in WINNERS:
+        run_name = w[0]
+        if run_name in name_to_id:
+            continue
+        runs = repo.list_runs()
+        for r in runs:
+            if r["name"] == run_name:
+                name_to_id[run_name] = r["id"]
+                break
+
+    token = (
+        settings.cerbero_mainnet_token.get_secret_value()
+        if settings.cerbero_mainnet_token
+        else settings.cerbero_testnet_token.get_secret_value()
+    )
+    cerbero = CerberoClient(
+        base_url=settings.cerbero_base_url,
+        token=token,
+        bot_tag=settings.cerbero_bot_tag,
+    )
+    loader = CerberoOHLCVLoader(client=cerbero, cache_dir=settings.series_dir)
+
+    print(f"{'='*120}")
+    print(f"PER-TRADE COMPARISON — {len(WINNERS)} winner candidates × 4 folds WFA")
+    print(f"{'='*120}")
+
+    for run_name, genome_id, symbol, timeframe, label in WINNERS:
+        run_id = name_to_id.get(run_name)
+        if not run_id:
+            print(f"!!! run not found: {run_name}")
+            continue
+        analyze_genome(run_id, genome_id, symbol, timeframe, label, settings, cerbero, loader)
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1,112 @@
+"""2 winner cross-tick: BTC 238e4812 + ETH c04dff7086 su 5m / 15m / 1h.
+
+Per ogni combinazione strategy × timeframe: backtest year-by-year (2019-2025)
+con metriche per-anno e totale 7y.
+"""
+from __future__ import annotations
+
+from datetime import datetime
+from pathlib import Path
+
+from multi_swarm_core.backtest.engine import BacktestEngine
+from multi_swarm_core.cerbero.client import CerberoClient
+from multi_swarm_core.config import load_settings
+from multi_swarm_core.data.cerbero_ohlcv import CerberoOHLCVLoader, OHLCVRequest
+from multi_swarm_core.metrics.basic import max_drawdown, sharpe_ratio, total_return
+from multi_swarm_core.protocol.compiler import compile_strategy
+from multi_swarm_core.protocol.parser import parse_strategy
+
+
+WINNERS = [
+    # (label, path, symbol)
+    ("BTC NEW (238e4812, native=1h)", "btc_238e4812.json", "BTC-PERPETUAL"),
+    ("ETH NEW (c04dff7086, native=5m)", "eth_c04dff7086.json", "ETH-PERPETUAL"),
+]
+TIMEFRAMES = ["5m", "15m", "1h"]
+
+YEARS = [
+    ("2019", "2019-01-01T00:00:00+00:00", "2020-01-01T00:00:00+00:00"),
+    ("2020", "2020-01-01T00:00:00+00:00", "2021-01-01T00:00:00+00:00"),
+    ("2021", "2021-01-01T00:00:00+00:00", "2022-01-01T00:00:00+00:00"),
+    ("2022", "2022-01-01T00:00:00+00:00", "2023-01-01T00:00:00+00:00"),
+    ("2023", "2023-01-01T00:00:00+00:00", "2024-01-01T00:00:00+00:00"),
+    ("2024", "2024-01-01T00:00:00+00:00", "2025-01-01T00:00:00+00:00"),
+    ("2025", "2025-01-01T00:00:00+00:00", "2026-01-01T00:00:00+00:00"),
+]
+
+
+def evaluate(strat, ohlcv, engine, label, tf) -> None:
+    print(f"\n  >>> tick={tf} | {len(ohlcv)} bars")
+    print(f"  {'year':<6} {'trades':>7} {'wins':>5} {'losses':>7} {'win%':>6} {'ret':>8} {'maxDD':>7} {'sharpe':>7}")
+    sum_ret = 0.0
+    sum_trades = 0
+    sum_wins = 0
+    for year_label, start, end in YEARS:
+        mask = (ohlcv.index >= datetime.fromisoformat(start)) & (ohlcv.index < datetime.fromisoformat(end))
+        slice_df = ohlcv[mask]
+        if len(slice_df) == 0:
+            continue
+        try:
+            signal_fn = compile_strategy(strat)
+            signals = signal_fn(slice_df)
+            bt = engine.run(slice_df, signals)
+        except Exception as e:
+            print(f"  {year_label:<6} ERROR: {e}")
+            continue
+        trades = bt.trades
+        n = len(trades)
+        wins = [t.net_pnl for t in trades if t.net_pnl > 0]
+        losses = [t.net_pnl for t in trades if t.net_pnl <= 0]
+        nw, nl = len(wins), len(losses)
+        wr = (nw / n * 100) if n else 0.0
+        if n > 0:
+            notional = float(slice_df["close"].iloc[0])
+            eq = (bt.equity_curve / notional) + 1.0
+            ret = total_return(eq)
+            dd = max_drawdown(eq)
+            sr = sharpe_ratio(bt.returns, periods_per_year=8760)
+        else:
+            ret = dd = sr = 0.0
+        print(f"  {year_label:<6} {n:>7} {nw:>5} {nl:>7} {wr:>5.1f}% {ret:>7.2%} {dd:>6.2%} {sr:>7.3f}")
+        sum_ret += ret
+        sum_trades += n
+        sum_wins += nw
+    overall_wr = (sum_wins / sum_trades * 100) if sum_trades else 0.0
+    print(f"  ===== 7y TOT: {sum_trades:>7} {sum_wins:>5} {sum_trades-sum_wins:>7} {overall_wr:>5.1f}%  cum_ret={sum_ret:+.2%}")
+
+
+def main() -> None:
+    settings = load_settings()
+    token = (
+        settings.cerbero_mainnet_token.get_secret_value()
+        if settings.cerbero_mainnet_token
+        else settings.cerbero_testnet_token.get_secret_value()
+    )
+    cerbero = CerberoClient(
+        base_url=settings.cerbero_base_url,
+        token=token,
+        bot_tag=settings.cerbero_bot_tag,
+    )
+    loader = CerberoOHLCVLoader(client=cerbero, cache_dir=settings.series_dir)
+    engine = BacktestEngine(fees_bp=5.0)
+    strategies_dir = Path("/app/strategies")
+
+    for label, fname, symbol in WINNERS:
+        path = strategies_dir / fname
+        strat = parse_strategy(path.read_text())
+        print(f"\n{'='*100}")
+        print(f">>> {label} — symbol={symbol}")
+        for tf in TIMEFRAMES:
+            try:
+                ohlcv = loader.load(OHLCVRequest(
+                    symbol=symbol, timeframe=tf,
+                    start=datetime.fromisoformat("2018-09-01T00:00:00+00:00"),
+                    end=datetime.fromisoformat("2026-01-01T00:00:00+00:00"),
+                ))
+                evaluate(strat, ohlcv, engine, label, tf)
+            except Exception as e:
+                print(f"\n  >>> tick={tf} FAILED TO LOAD: {e}")
+
+
+if __name__ == "__main__":
+    main()
@@ -33,6 +33,36 @@ from strategy_crypto.backend import PaperExecutor, PaperRepository, Portfolio

 PROJECT_ROOT = Path(__file__).resolve().parent.parent

+# Mapping timeframe stringa Cerbero -> minuti del bar. Le strategie tradano
+# sul "bar appena chiuso", quindi end deve essere snappato al boundary del
+# loro timeframe (NON sempre al top dell'ora) per evitare la regressione in
+# cui ETH 5m veniva valutato una volta sola ogni 60 min.
+_TIMEFRAME_MINUTES: dict[str, int] = {
+    "1m": 1,
+    "5m": 5,
+    "15m": 15,
+    "30m": 30,
+    "1h": 60,
+    "4h": 240,
+    "1d": 1440,
+}
+
+
+def _align_end_to_timeframe(now: datetime, timeframe: str) -> datetime:
+    """Snap ``now`` al boundary del bar timeframe (UTC, naive seconds).
+
+    Es.: now=14:37:42, tf="5m"  -> 14:35:00
+         now=14:37:42, tf="1h"  -> 14:00:00
+         now=14:00:00, tf="1h"  -> 14:00:00
+    """
+    bar_min = _TIMEFRAME_MINUTES[timeframe]
+    aligned = now.replace(second=0, microsecond=0)
+    if bar_min >= 1440:
+        return aligned.replace(hour=0, minute=0)
+    total_min = aligned.hour * 60 + aligned.minute
+    snapped = (total_min // bar_min) * bar_min
+    return aligned.replace(hour=snapped // 60, minute=snapped % 60)
+

 def _default_strategies_dir() -> Path:
    """Cartella JSON shippata col package strategy_crypto."""
@@ -70,9 +100,11 @@ def load_assets(strategies_dir: Path) -> list[AssetConfig]:
        raise FileNotFoundError(
            f"Expected btc_*.json and eth_*.json in {strategies_dir}"
        )
+    # ETH winner c04dff7086 e' tunato su 5m: a 1h la strategia perde (cum_ret -33% 7y).
+    # BTC winner 238e4812 e' tunato su 1h: tick native = paper tick.
    return [
-        AssetConfig(symbol="BTC-PERPETUAL", strategy_file=btc_files[0]),
-        AssetConfig(symbol="ETH-PERPETUAL", strategy_file=eth_files[0]),
+        AssetConfig(symbol="BTC-PERPETUAL", strategy_file=btc_files[0], timeframe="1h"),
+        AssetConfig(symbol="ETH-PERPETUAL", strategy_file=eth_files[0], timeframe="5m"),
    ]


@@ -129,9 +161,12 @@ def main() -> None:
            now = datetime.now(UTC)
            last_prices: dict[str, float] = {}
            for asset, executor in zip(assets, executors, strict=True):
-                # fetch OHLCV most recent lookback bars
-                end = now.replace(minute=0, second=0, microsecond=0)
-                start = end - timedelta(hours=args.lookback_bars + 1)
+                # fetch OHLCV most recent lookback bars: end snappato al timeframe
+                # dell'asset, non sempre all'ora (altrimenti ETH 5m veniva valutato
+                # solo ogni 60 min, regressione vs backtest tunato 5m).
+                bar_min = _TIMEFRAME_MINUTES[asset.timeframe]
+                end = _align_end_to_timeframe(now, asset.timeframe)
+                start = end - timedelta(minutes=bar_min * (args.lookback_bars + 1))
                req = OHLCVRequest(
                    symbol=asset.symbol,
                    timeframe=asset.timeframe,
@@ -19,6 +19,30 @@ def _default_prompt_library_path() -> Path:
    return Path(str(importlib.resources.files("strategy_crypto") / "prompts.json"))


+# Default v2 hard-kill list: oltre ai degenerate originali, fees_eat_alpha e
+# negative_net_pnl sono deal-breaker non recuperabili via soft penalty (vedi
+# 7y-overfit incident 2026-05-16: elite IS Sharpe 1.93 -> net -5% su 7y per fees).
+_DEFAULT_V2_HARD_KILL: tuple[str, ...] = (
+    "no_trades",
+    "degenerate",
+    "undertrading",
+    "fees_eat_alpha",
+    "negative_net_pnl",
+)
+
+
+def _resolve_hard_kill(args) -> tuple[str, ...] | None:
+    """Resolve la lista hard-kill da CLI args.
+
+    Priority: ``--fitness-hard-kill`` esplicito > default v2 > ``None`` (v1).
+    """
+    if args.fitness_hard_kill:
+        return tuple(s.strip() for s in args.fitness_hard_kill.split(",") if s.strip())
+    if args.fitness_v2:
+        return _DEFAULT_V2_HARD_KILL
+    return None
+
+
 def parse_args() -> argparse.Namespace:
    p = argparse.ArgumentParser(description="Multi-Swarm Phase 1 runner")
    p.add_argument("--name", default="phase1-spike-001")
@@ -35,7 +59,10 @@ def parse_args() -> argparse.Namespace:
    )
    p.add_argument("--symbol", default="BTC-PERPETUAL")
    p.add_argument("--timeframe", default="1h")
-    p.add_argument("--start", default="2024-01-01T00:00:00+00:00")
+    # Default esteso a 7.3 anni: copre bear 2018-19, halving 2020, COVID,
+    # ATH 2021, winter 2022, ETF rally 2024, regime corrente. Una finestra
+    # corta lascia il GA libero di overfit a un singolo regime.
+    p.add_argument("--start", default="2018-09-01T00:00:00+00:00")
    p.add_argument("--end", default="2026-01-01T00:00:00+00:00")
    p.add_argument("--fees-bp", type=float, default=5.0)
    p.add_argument("--n-trials-dsr", type=int, default=50)
@@ -67,8 +94,10 @@ def parse_args() -> argparse.Namespace:
        "--fitness-v2",
        action="store_true",
        help=(
-            "Attiva fitness v2: solo {no_trades, degenerate, undertrading} azzerano; "
-            "gli altri HIGH applicano soft penalty multiplicativa"
+            "Attiva fitness v2: hard-kill su {no_trades, degenerate, undertrading, "
+            "fees_eat_alpha, negative_net_pnl}; gli altri HIGH applicano soft penalty "
+            "multiplicativa. Versione hardened post 7y-overfit incident: fees + "
+            "net negativo non sono recuperabili."
        ),
    )
    p.add_argument(
@@ -77,6 +106,16 @@ def parse_args() -> argparse.Namespace:
        default=0.4,
        help="v2: fattore soft penalty per HIGH non-hard (default 0.4 → 1 HIGH → 0.71x)",
    )
+    p.add_argument(
+        "--fitness-hard-kill",
+        type=str,
+        default=None,
+        help=(
+            "Override comma-separated della lista di finding name che azzerano la "
+            "fitness in modalita' v2. Es: 'no_trades,degenerate'. Default v2: "
+            "no_trades,degenerate,undertrading,fees_eat_alpha,negative_net_pnl."
+        ),
+    )
    p.add_argument(
        "--wfa-train-split",
        type=float,
@@ -114,6 +153,16 @@ def parse_args() -> argparse.Namespace:
            "Schema: {styles: {<name>: {directive: <testo>}}}"
        ),
    )
+    p.add_argument(
+        "--llm-concurrency",
+        type=int,
+        default=1,
+        help=(
+            "Numero di propose() LLM concorrenti per generazione (default 1 = "
+            "serial). OpusAgent processa in coda FIFO; concurrency > 1 accoda "
+            "piu' richieste in parallelo."
+        ),
+    )
    return p.parse_args()


@@ -151,13 +200,13 @@ def main() -> None:
    print(f"OHLCV loaded: {len(ohlcv)} bars from {ohlcv.index[0]} to {ohlcv.index[-1]}")

    llm = LLMClient(
-        openrouter_api_key=settings.openrouter_api_key.get_secret_value(),
+        opus_agent_api_key=settings.opus_agent_api_key.get_secret_value(),
+        opus_agent_base_url=settings.opus_agent_base_url,
        model_tier_s=settings.llm_model_tier_s,
        model_tier_a=settings.llm_model_tier_a,
        model_tier_b=settings.llm_model_tier_b,
        model_tier_c=settings.llm_model_tier_c,
        model_tier_d=settings.llm_model_tier_d,
-        openrouter_base_url=settings.openrouter_base_url,
    )

    cfg = RunConfig(
@@ -178,15 +227,14 @@ def main() -> None:
        fees_eat_alpha_threshold=args.fees_eat_alpha_threshold,
        flat_too_long_threshold=args.flat_too_long_threshold,
        undertrading_threshold=args.undertrading_threshold,
-        fitness_hard_kill_findings=(
-            ("no_trades", "degenerate", "undertrading") if args.fitness_v2 else None
-        ),
+        fitness_hard_kill_findings=_resolve_hard_kill(args),
        fitness_adversarial_soft_penalty=args.fitness_soft_penalty,
        wfa_train_split=args.wfa_train_split,
        wfa_top_k=args.wfa_top_k,
        eval_oos_during_loop=args.eval_oos_during_loop,
        fitness_combined_alpha=args.fitness_combined_alpha,
        prompt_library=prompt_library,
+        llm_concurrency=args.llm_concurrency,
    )

    run_id = run_phase1(cfg, ohlcv=ohlcv, llm=llm)
@@ -0,0 +1,475 @@
+"""Smoke test del GA per strategy_pythagoras.
+
+Esegue 2 run di Phase 1 (BTC 5m + ETH 5m), poi cross-rank i genomi
+comuni applicando il bonus di asset-invariance (corr_signal sui pattern
+di entry entro +/-36 barre = +/-3h su 5m TF, vedi paper Pythagoras p.43).
+
+Configurazione (per spec §4):
+- Population 20, 5 generations
+- Asset: BTC-PERPETUAL 5m + ETH-PERPETUAL 5m (Cerbero deribit)
+- Train window: 2024-07-01 -> 2024-12-31
+- Test window:  2025-01-01 -> 2025-01-31 (caricato come coda dello stesso
+  range; non usato dal GA ma necessario per dataset continuo se in futuro
+  si attiva WFA)
+- Stili cognitivi: 7 da strategy_pythagoras/prompts.json
+- Indicatori Pythagoras: candle_pattern, pythagorean_ratio, fractal_mirror
+  (registrati nel compiler tramite import side-effect di strategy_pythagoras.indicators)
+- Fitness post-processing cross-asset: apply_invariance_bonus
+- Output: top 50 winners persisted in state/strategy_pythagoras.db
+  (tabella pythagoras_winners)
+
+Adattamento all'API reale di run_phase1 (Task 4.1 findings + verifica diretta):
+
+- ``run_phase1(cfg: RunConfig, ohlcv: pd.DataFrame, llm: LLMClient) -> str``
+  ritorna un ``run_id``. Non c'e' un fitness hook esterno: il GA loop
+  invoca ``compute_fitness`` inline e persiste via
+  ``repo.save_evaluation``. Per il bonus invariance dobbiamo:
+    1. lanciare due ``run_phase1`` indipendenti, uno per asset;
+    2. caricare le evaluations via ``repo.list_evaluations(run_id)``;
+    3. ricompilare la strategia (``_try_parse`` + ``compile_strategy``) sui
+       segnali di ciascun OHLCV per estrarre gli entry index;
+    4. calcolare ``corr_signal`` sugli entry binari (Series int-indexed)
+       e applicare ``apply_invariance_bonus``.
+
+- Le serie OHLCV NON sono shippate in repo come ``src/strategy_crypto/series/``:
+  il default loader Cerbero le cachea in ``./series/{cache_key}.parquet``
+  (cache key = sha1 di ``exchange|symbol|timeframe|start|end``). Riusiamo
+  quel meccanismo: caricamento via ``CerberoOHLCVLoader``, identico a
+  ``scripts/run_phase1.py``.
+
+Shape effettivo del dict ritornato da ``repo.list_evaluations(run_id)``
+(vedi ``persistence/repository.py:213`` e schema in ``schema.py``):
+
+    {
+      'run_id', 'genome_id', 'fitness', 'dsr', 'dsr_pvalue', 'sharpe',
+      'max_dd', 'total_return', 'n_trades', 'parse_error', 'raw_text',
+      'eval_ts', 'fitness_oos', 'sharpe_oos', 'return_oos',
+      'max_dd_oos', 'n_trades_oos'
+    }
+
+Nota: ``cognitive_style`` e ``generation`` NON sono nelle evaluations;
+vanno presi via ``repo.list_genomes(run_id)`` (payload_json del genoma).
+``raw_text`` contiene il completion grezzo del LLM, da cui si estrae
+nuovamente lo ``Strategy`` AST via ``_try_parse``.
+"""
+from __future__ import annotations
+
+import json
+import logging
+import os
+import sqlite3
+from datetime import datetime
+from pathlib import Path
+from typing import Any
+
+import pandas as pd  # type: ignore[import-untyped]
+
+# Side-effect import: registra candle_pattern, pythagorean_ratio, fractal_mirror
+# in compiler.INDICATOR_FNS prima che il GA inizi a compilare strategie.
+# (Il compiler in protocol/compiler.py importa gia' i 3 simboli dal package
+# strategy_pythagoras.indicators, ma facciamo l'import esplicito qui per
+# rendere la dipendenza chiara e indipendente dall'ordine di import.)
+import strategy_pythagoras.indicators  # noqa: F401
+from multi_swarm_core.agents.hypothesis import _try_parse
+from multi_swarm_core.backtest.orders import Side
+from multi_swarm_core.cerbero.client import CerberoClient
+from multi_swarm_core.config import load_settings
+from multi_swarm_core.data.cerbero_ohlcv import CerberoOHLCVLoader, OHLCVRequest
+from multi_swarm_core.genome.hypothesis import ModelTier
+from multi_swarm_core.genome.prompt_library import PromptLibrary
+from multi_swarm_core.llm.client import LLMClient
+from multi_swarm_core.orchestrator.run import RunConfig, run_phase1
+from multi_swarm_core.persistence.repository import Repository
+from multi_swarm_core.protocol.compiler import compile_strategy
+from strategy_pythagoras.fitness_invariance import (
+    apply_invariance_bonus,
+    corr_signal,
+)
+
+ROOT = Path(__file__).resolve().parents[1]
+DB_PATH = Path(
+    os.getenv("STRATEGY_PYTHAGORAS_DB_PATH", str(ROOT / "state" / "strategy_pythagoras.db"))
+)
+PROMPTS_PATH = ROOT / "src" / "strategy_pythagoras" / "strategy_pythagoras" / "prompts.json"
+RUN_NAME = os.getenv("PYTHAGORAS_SMOKE_RUN_NAME", "pythagoras-smoke-001")
+
+# GA configuration (smoke per spec §4)
+POPULATION = 20
+GENERATIONS = 5
+
+# Data window
+TRAIN_START = datetime.fromisoformat("2024-07-01T00:00:00+00:00")
+TRAIN_END = datetime.fromisoformat("2024-12-31T23:55:00+00:00")
+# Carichiamo anche gennaio 2025 come coda (per usi futuri: WFA OOS).
+# Il GA loop in questa fase usa l'intero range; e' compito di un eventuale
+# wfa_train_split (non attivato qui per coerenza con spec §4 smoke).
+TEST_END = datetime.fromisoformat("2025-01-31T23:55:00+00:00")
+
+ASSETS: list[tuple[str, str]] = [
+    ("BTC-PERPETUAL", "btc"),
+    ("ETH-PERPETUAL", "eth"),
+]
+TIMEFRAME = "5m"
+EXCHANGE = "deribit"
+TOP_K_PERSIST = 50
+
+logger = logging.getLogger(RUN_NAME)
+
+
+def init_winners_table(con: sqlite3.Connection) -> None:
+    """Crea ``pythagoras_winners`` se non esiste (idempotente)."""
+    con.execute(
+        """
+        CREATE TABLE IF NOT EXISTS pythagoras_winners (
+            genome_id TEXT PRIMARY KEY,
+            cognitive_style TEXT,
+            fitness REAL,
+            sharpe_btc REAL,
+            sharpe_eth REAL,
+            invariance_score REAL,
+            rules_json TEXT,
+            generation INTEGER,
+            run_name TEXT
+        )
+        """
+    )
+    con.commit()
+
+
+def _load_ohlcv(loader: CerberoOHLCVLoader, symbol: str) -> pd.DataFrame:
+    """Carica la finestra ``TRAIN_START -> TEST_END`` per ``symbol`` su 5m."""
+    req = OHLCVRequest(
+        symbol=symbol,
+        timeframe=TIMEFRAME,
+        start=TRAIN_START,
+        end=TEST_END,
+        exchange=EXCHANGE,
+    )
+    ohlcv = loader.load(req)
+    logger.info(
+        "OHLCV loaded for %s: %d bars (%s -> %s)",
+        symbol, len(ohlcv),
+        ohlcv.index[0] if len(ohlcv) else "n/a",
+        ohlcv.index[-1] if len(ohlcv) else "n/a",
+    )
+    return ohlcv
+
+
+def _build_run_config(
+    run_name: str, symbol: str, prompt_library: PromptLibrary, db_path: Path,
+) -> RunConfig:
+    """Costruisce il ``RunConfig`` per un singolo asset.
+
+    Usa lo stesso GA-core DB del progetto (``settings.ga_db_path`` se override
+    non passato): vi vengono scritte ``runs``, ``generations``, ``genomes``,
+    ``evaluations`` per la run.
+    """
+    return RunConfig(
+        run_name=run_name,
+        population_size=POPULATION,
+        n_generations=GENERATIONS,
+        elite_k=2,
+        tournament_k=3,
+        p_crossover=0.5,
+        seed=42,
+        model_tier=ModelTier.C,
+        symbol=symbol,
+        timeframe=TIMEFRAME,
+        fees_bp=5.0,
+        n_trials_dsr=50,
+        db_path=db_path,
+        prompt_library=prompt_library,
+        # Smoke: niente WFA, niente eval OOS in loop, niente prompt mutation LLM.
+        # I parametri restano sui default sicuri di RunConfig.
+    )
+
+
+def run_ga_for_asset(
+    asset_label: str,
+    symbol: str,
+    ohlcv: pd.DataFrame,
+    prompt_library: PromptLibrary,
+    llm: LLMClient,
+    ga_db_path: Path,
+) -> tuple[str, Repository]:
+    """Lancia ``run_phase1`` per un asset.
+
+    Ritorna ``(run_id, repo)`` per il caller, che usera' ``repo`` per
+    estrarre evaluations + genomes a fine del run.
+    """
+    run_name = f"{RUN_NAME}-{asset_label}"
+    cfg = _build_run_config(run_name, symbol, prompt_library, ga_db_path)
+    logger.info("Starting GA run '%s' on %s (%d bars)", run_name, symbol, len(ohlcv))
+    run_id = run_phase1(cfg, ohlcv=ohlcv, llm=llm)
+    logger.info("Run '%s' completed: run_id=%s", run_name, run_id)
+    repo = Repository(ga_db_path)
+    return run_id, repo
+
+
+def _entries_series_from_eval(
+    eval_row: dict[str, Any], ohlcv: pd.DataFrame,
+) -> pd.Series | None:
+    """Ricostruisce gli entries binari (Side.LONG/SHORT -> 1, altrimenti 0)
+    a partire dal ``raw_text`` salvato nell'eval row.
+
+    Ritorna ``None`` se il raw_text non e' parsabile (caso parse_error).
+    L'index della Series ritornata e' INTERO posizionale (0..N-1) come
+    richiesto da ``corr_signal`` (vedi tests in
+    ``strategy_pythagoras/tests/test_fitness_invariance.py``).
+    """
+    raw = eval_row.get("raw_text")
+    if not raw:
+        return None
+    strategy, parse_err = _try_parse(raw)
+    if strategy is None:
+        logger.debug(
+            "skip genome %s: parse error '%s'",
+            eval_row.get("genome_id"), parse_err,
+        )
+        return None
+    try:
+        signal_fn = compile_strategy(strategy)
+        signals = signal_fn(ohlcv)
+    except Exception as exc:
+        logger.debug(
+            "skip genome %s: compile/exec error: %s",
+            eval_row.get("genome_id"), exc,
+        )
+        return None
+    # 1 dove il signal e' LONG o SHORT (entry attiva), 0 altrove.
+    is_entry = signals.isin([Side.LONG, Side.SHORT]).fillna(False).astype(int)
+    # Riassegna integer index per il match in corr_signal (che somma delta
+    # interi all'index e fa il test ``ti + delta in b_set``).
+    return pd.Series(is_entry.values, index=range(len(is_entry)), dtype="int64")
+
+
+def _collect_evaluations(
+    repo: Repository, run_id: str, ohlcv: pd.DataFrame,
+) -> dict[str, dict[str, Any]]:
+    """Carica evaluations + genomes per ``run_id`` e li unisce per genome_id.
+
+    Returns: dict ``{genome_id: row}`` dove ``row`` contiene i campi
+    dell'eval + ``cognitive_style``, ``generation``, ``strategy_json``
+    (dict del genoma serializzato) e ``entries`` (pd.Series int-indexed).
+    """
+    evals = repo.list_evaluations(run_id)
+    genomes = repo.list_genomes(run_id)
+    genome_by_id: dict[str, dict[str, Any]] = {}
+    for grow in genomes:
+        try:
+            payload = json.loads(grow["payload_json"])
+        except (json.JSONDecodeError, TypeError):
+            payload = {}
+        genome_by_id[grow["id"]] = payload
+
+    out: dict[str, dict[str, Any]] = {}
+    for ev in evals:
+        gid = ev["genome_id"]
+        payload = genome_by_id.get(gid, {})
+        row = dict(ev)
+        row["cognitive_style"] = payload.get("cognitive_style", "")
+        row["generation"] = int(payload.get("generation", 0))
+        # ``raw_text`` e' il completion grezzo; lo ri-parsiamo in
+        # _entries_series_from_eval. Salviamo la rappresentazione canonica
+        # ``strategy_json`` per persistenza (best-effort: se il parse fallisce
+        # salviamo il raw_text come fallback).
+        strategy, _err = _try_parse(row.get("raw_text") or "")
+        if strategy is not None:
+            # Strategy non e' direttamente JSON-serializable: serializziamo
+            # la struttura nominale tramite dataclasses.asdict-like fallback.
+            try:
+                row["strategy_json"] = _strategy_to_jsonable(strategy)
+            except Exception:
+                row["strategy_json"] = {"raw_text": row.get("raw_text", "")}
+        else:
+            row["strategy_json"] = {"raw_text": row.get("raw_text", "")}
+        row["entries"] = _entries_series_from_eval(ev, ohlcv)
+        out[gid] = row
+    return out
+
+
+def _strategy_to_jsonable(strategy: Any) -> dict[str, Any]:
+    """Serializza un ``Strategy`` AST in dict JSON-friendly.
+
+    Strategy/Rule/Node sono dataclass: usiamo ``dataclasses.asdict`` quando
+    possibile, con fallback a ``str(strategy)`` se la struttura contiene
+    membri non-serializzabili (es. enum non-Str).
+    """
+    import dataclasses
+    if dataclasses.is_dataclass(strategy):
+        try:
+            return dataclasses.asdict(strategy)
+        except TypeError:
+            pass
+    return {"repr": repr(strategy)}
+
+
+def compute_invariance_for_pair(
+    btc_evals: dict[str, dict[str, Any]],
+    eth_evals: dict[str, dict[str, Any]],
+) -> list[dict[str, Any]]:
+    """Per ogni genome_id presente in entrambi i run, calcola invariance + bonus.
+
+    Lo stesso ``genome_id`` puo' apparire in entrambi i run perche' l'id e'
+    deterministico (sha1 di system_prompt+feature_access+temperature+...) e
+    il seed del GA e' fisso: il founder set e i mutanti hanno alta probabilita'
+    di collisione cross-asset. Quando il genoma compare in entrambi, le
+    metriche ``sharpe`` IS sono comparabili e ha senso valutare l'invarianza.
+    """
+    out: list[dict[str, Any]] = []
+    common_ids = set(btc_evals) & set(eth_evals)
+    logger.info(
+        "Common genomes BTC ∩ ETH: %d (BTC: %d, ETH: %d)",
+        len(common_ids), len(btc_evals), len(eth_evals),
+    )
+    for gid in common_ids:
+        b = btc_evals[gid]
+        e = eth_evals[gid]
+        entries_btc: pd.Series | None = b.get("entries")
+        entries_eth: pd.Series | None = e.get("entries")
+        if entries_btc is None or entries_eth is None:
+            inv = 0.0
+        elif len(entries_btc) == 0 or len(entries_eth) == 0:
+            inv = 0.0
+        else:
+            try:
+                # Allineiamo a lunghezza minima: i due asset possono avere
+                # un numero di bars leggermente diverso (gap nel feed Cerbero).
+                # corr_signal lavora solo sugli index 1 -> il troncamento non
+                # introduce bias asimmetrici.
+                min_len = min(len(entries_btc), len(entries_eth))
+                inv = corr_signal(
+                    entries_btc.iloc[:min_len].reset_index(drop=True),
+                    entries_eth.iloc[:min_len].reset_index(drop=True),
+                )
+            except Exception as exc:
+                logger.warning("corr_signal failed for %s: %s", gid, exc)
+                inv = 0.0
+        sharpe_btc = float(b.get("sharpe") or 0.0)
+        sharpe_eth = float(e.get("sharpe") or 0.0)
+        mean_sharpe = 0.5 * (sharpe_btc + sharpe_eth)
+        boosted = apply_invariance_bonus(mean_sharpe, inv)
+        out.append({
+            "genome_id": gid,
+            "cognitive_style": b.get("cognitive_style") or e.get("cognitive_style", ""),
+            "fitness": float(boosted),
+            "sharpe_btc": sharpe_btc,
+            "sharpe_eth": sharpe_eth,
+            "invariance_score": float(inv),
+            "rules_json": json.dumps(b.get("strategy_json") or {}, default=str),
+            "generation": int(b.get("generation", 0)),
+            "run_name": RUN_NAME,
+        })
+    return sorted(out, key=lambda r: r["fitness"], reverse=True)
+
+
+def persist_winners(con: sqlite3.Connection, winners: list[dict[str, Any]]) -> None:
+    if not winners:
+        logger.warning("No winners to persist")
+        return
+    con.executemany(
+        """
+        INSERT OR REPLACE INTO pythagoras_winners
+        (genome_id, cognitive_style, fitness, sharpe_btc, sharpe_eth,
+         invariance_score, rules_json, generation, run_name)
+        VALUES (:genome_id, :cognitive_style, :fitness, :sharpe_btc, :sharpe_eth,
+                :invariance_score, :rules_json, :generation, :run_name)
+        """,
+        winners,
+    )
+    con.commit()
+
+
+def main() -> None:
+    logging.basicConfig(
+        level=logging.INFO,
+        format="%(asctime)s %(levelname)s %(name)s %(message)s",
+    )
+    settings = load_settings()
+
+    # Prompt library Pythagoras (NON quello di strategy_crypto).
+    if not PROMPTS_PATH.exists():
+        raise FileNotFoundError(f"Prompts file not found: {PROMPTS_PATH}")
+    prompt_library = PromptLibrary.from_json(PROMPTS_PATH)
+    logger.info(
+        "PromptLibrary loaded from %s: %d styles (%s)",
+        PROMPTS_PATH, len(prompt_library.styles),
+        ", ".join(prompt_library.cognitive_styles),
+    )
+
+    # Cerbero client + OHLCV loader (riusa la cache parquet in ./series).
+    token = (
+        settings.cerbero_mainnet_token.get_secret_value()
+        if settings.cerbero_mainnet_token
+        else settings.cerbero_testnet_token.get_secret_value()
+    )
+    cerbero = CerberoClient(
+        base_url=settings.cerbero_base_url,
+        token=token,
+        bot_tag=settings.cerbero_bot_tag,
+    )
+    loader = CerberoOHLCVLoader(client=cerbero, cache_dir=settings.series_dir)
+
+    llm = LLMClient(
+        opus_agent_api_key=settings.opus_agent_api_key.get_secret_value(),
+        opus_agent_base_url=settings.opus_agent_base_url,
+        model_tier_s=settings.llm_model_tier_s,
+        model_tier_a=settings.llm_model_tier_a,
+        model_tier_b=settings.llm_model_tier_b,
+        model_tier_c=settings.llm_model_tier_c,
+        model_tier_d=settings.llm_model_tier_d,
+    )
+
+    # Setup DB winners (separato dal GA core DB).
+    DB_PATH.parent.mkdir(parents=True, exist_ok=True)
+    con = sqlite3.connect(DB_PATH)
+    try:
+        init_winners_table(con)
+        logger.info("Winners DB initialized at %s", DB_PATH)
+
+        # Carica OHLCV per entrambi gli asset PRIMA dei run GA, cosi' se la
+        # rete o Cerbero sono giu' falliamo subito senza sprecare chiamate LLM.
+        ohlcv_by_asset: dict[str, pd.DataFrame] = {}
+        for symbol, label in ASSETS:
+            ohlcv_by_asset[label] = _load_ohlcv(loader, symbol)
+
+        # Run GA per asset. Usa il GA-core DB definito in settings; ogni run
+        # crea un proprio run_id e set di evaluations isolato.
+        evals_by_asset: dict[str, dict[str, dict[str, Any]]] = {}
+        for symbol, label in ASSETS:
+            run_id, repo = run_ga_for_asset(
+                asset_label=label,
+                symbol=symbol,
+                ohlcv=ohlcv_by_asset[label],
+                prompt_library=prompt_library,
+                llm=llm,
+                ga_db_path=settings.ga_db_path,
+            )
+            evals_by_asset[label] = _collect_evaluations(
+                repo, run_id, ohlcv_by_asset[label]
+            )
+            logger.info(
+                "%s: %d evaluations collected", label.upper(),
+                len(evals_by_asset[label]),
+            )
+
+        # Cross-rank con invariance bonus.
+        winners = compute_invariance_for_pair(
+            evals_by_asset["btc"], evals_by_asset["eth"],
+        )
+        logger.info(
+            "Computed invariance bonus for %d common genomes", len(winners),
+        )
+
+        top = winners[:TOP_K_PERSIST]
+        persist_winners(con, top)
+        logger.info(
+            "Persisted top %d winners to %s (table: pythagoras_winners)",
+            len(top), DB_PATH,
+        )
+    finally:
+        con.close()
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1,271 @@
+"""Multi-fold validation di un run esistente.
+
+Prende un ``run_id`` da ``state/runs.db``, seleziona i top-K genomi per fitness IS,
+e li rivaluta su N fold expanding-window di un dataset OHLCV (tipicamente piu'
+lungo del train del GA). Output: per-fold + aggregati (mean / min / std) della
+fitness OOS.
+
+Use case: il GA puo' selezionare un "lucky-shot" overfit a uno specifico regime.
+Validare i top-K su finestre temporali diverse rivela quali strategie sono
+robuste vs overfitter.
+
+Esempio::
+
+  python scripts/validate_run.py \\
+    --run-id e263651598894da688d95fda90a34a96 \\
+    --top-k 10 --n-folds 4 \\
+    --symbol BTC-PERPETUAL --timeframe 1h \\
+    --start 2018-09-01 --end 2026-01-01
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import statistics
+from datetime import datetime
+from pathlib import Path
+
+import pandas as pd  # type: ignore[import-untyped]
+
+from multi_swarm_core.agents.adversarial import AdversarialAgent
+from multi_swarm_core.agents.falsification import FalsificationAgent
+from multi_swarm_core.agents.hypothesis import _try_parse
+from multi_swarm_core.cerbero.client import CerberoClient
+from multi_swarm_core.config import load_settings
+from multi_swarm_core.data.cerbero_ohlcv import CerberoOHLCVLoader, OHLCVRequest
+from multi_swarm_core.data.splits import expanding_walk_forward
+from multi_swarm_core.ga.fitness import compute_fitness
+from multi_swarm_core.persistence.repository import Repository
+
+
+def parse_args() -> argparse.Namespace:
+    p = argparse.ArgumentParser(description="Multi-fold validation di top-K genomi")
+    p.add_argument("--run-id", required=True, help="run_id da validare")
+    p.add_argument("--top-k", type=int, default=10, help="quanti genomi top valutare")
+    p.add_argument("--n-folds", type=int, default=4, help="numero fold expanding-window")
+    p.add_argument(
+        "--train-ratio",
+        type=float,
+        default=0.5,
+        help="frazione iniziale per il train iniziale (folds testano la coda)",
+    )
+    p.add_argument("--symbol", default="BTC-PERPETUAL")
+    p.add_argument("--timeframe", default="1h")
+    p.add_argument("--exchange", default="deribit", choices=["deribit", "bybit", "hyperliquid"])
+    p.add_argument("--start", default="2018-09-01T00:00:00+00:00")
+    p.add_argument("--end", default="2026-01-01T00:00:00+00:00")
+    p.add_argument("--fees-bp", type=float, default=5.0)
+    p.add_argument("--n-trials-dsr", type=int, default=50)
+    p.add_argument(
+        "--fees-eat-alpha-threshold", type=float, default=0.5,
+    )
+    p.add_argument(
+        "--flat-too-long-threshold", type=float, default=0.95,
+    )
+    p.add_argument(
+        "--undertrading-threshold", type=int, default=10,
+    )
+    p.add_argument(
+        "--fitness-v2", action="store_true",
+        help="Coerente con --fitness-v2 del run originale",
+    )
+    p.add_argument(
+        "--fitness-soft-penalty", type=float, default=0.4,
+    )
+    p.add_argument(
+        "--output-json",
+        type=Path,
+        default=None,
+        help="Path JSON dove salvare i risultati (default: stdout solo)",
+    )
+    return p.parse_args()
+
+
+def main() -> None:
+    args = parse_args()
+    settings = load_settings()
+
+    # Repository: top-K genomi per fitness IS, con raw_text parsable.
+    repo = Repository(settings.ga_db_path)
+    repo.init_schema()
+    run = repo.get_run(args.run_id)
+    if run is None:
+        raise SystemExit(f"run_id non trovato: {args.run_id}")
+    print(f"Validating run: {run['name']} ({args.run_id})")
+    print(f"  status: {run['status']}, cost: ${run['total_cost_usd']:.4f}")
+
+    all_evals = repo.list_evaluations(args.run_id)
+    parseable = [
+        e for e in all_evals
+        if e.get("raw_text") and not e.get("parse_error") and e["fitness"] > 0
+    ]
+    parseable.sort(key=lambda e: e["fitness"], reverse=True)
+
+    # Dedup by genome_id (gli elite vengono salvati una sola volta ma possono apparire
+    # in evaluations multiple se rivalutati con eval_oos_during_loop).
+    seen_ids: set[str] = set()
+    top_genomes: list[dict] = []
+    for e in parseable:
+        if e["genome_id"] in seen_ids:
+            continue
+        seen_ids.add(e["genome_id"])
+        top_genomes.append(e)
+        if len(top_genomes) >= args.top_k:
+            break
+    print(f"  selected top-{len(top_genomes)} genomes for validation")
+
+    # OHLCV: carica il dataset esteso.
+    token = (
+        settings.cerbero_mainnet_token.get_secret_value()
+        if settings.cerbero_mainnet_token
+        else settings.cerbero_testnet_token.get_secret_value()
+    )
+    cerbero = CerberoClient(
+        base_url=settings.cerbero_base_url,
+        token=token,
+        bot_tag=settings.cerbero_bot_tag,
+    )
+    loader = CerberoOHLCVLoader(client=cerbero, cache_dir=settings.series_dir)
+    req = OHLCVRequest(
+        symbol=args.symbol,
+        timeframe=args.timeframe,
+        start=datetime.fromisoformat(args.start),
+        end=datetime.fromisoformat(args.end),
+        exchange=args.exchange,
+    )
+    ohlcv = loader.load(req)
+    print(f"  OHLCV: {len(ohlcv)} bars from {ohlcv.index[0]} to {ohlcv.index[-1]}")
+
+    splits = expanding_walk_forward(
+        ohlcv.index, train_ratio=args.train_ratio, n_folds=args.n_folds,
+    )
+    print(f"  generated {len(splits)} folds")
+    for s in splits:
+        print(f"    fold {s.fold}: test [{s.test_idx[0]} -> {s.test_idx[-1]}] ({len(s.test_idx)} bars)")
+
+    fals_agent = FalsificationAgent(fees_bp=args.fees_bp, n_trials_dsr=args.n_trials_dsr)
+    adv_agent = AdversarialAgent(
+        fees_bp=args.fees_bp,
+        fees_eat_alpha_threshold=args.fees_eat_alpha_threshold,
+        flat_too_long_threshold=args.flat_too_long_threshold,
+        undertrading_threshold=args.undertrading_threshold,
+    )
+    hard_kill = (
+        ("no_trades", "degenerate", "undertrading") if args.fitness_v2 else None
+    )
+
+    # Itera per ogni genome + fold.
+    results: list[dict] = []
+    for gi, ev in enumerate(top_genomes):
+        strategy, parse_err = _try_parse(ev["raw_text"] or "")
+        if strategy is None:
+            print(f"  [{gi}] {ev['genome_id'][:12]} skip (parse error: {parse_err})")
+            continue
+
+        per_fold: list[dict] = []
+        for s in splits:
+            test_df = ohlcv.loc[s.test_idx]
+            try:
+                fals = fals_agent.evaluate(strategy, test_df)
+                adv = adv_agent.review(strategy, test_df)
+                fit = compute_fitness(
+                    fals, adv,
+                    hard_kill_findings=hard_kill,
+                    adversarial_soft_penalty=args.fitness_soft_penalty,
+                )
+            except Exception as e:
+                print(f"    fold {s.fold} eval failed: {e}")
+                continue
+            per_fold.append({
+                "fold": s.fold,
+                "fitness": float(fit),
+                "sharpe": float(fals.sharpe),
+                "dsr": float(fals.dsr),
+                "dsr_pvalue": float(fals.dsr_pvalue),
+                "return": float(fals.total_return),
+                "max_dd": float(fals.max_drawdown),
+                "n_trades": int(fals.n_trades),
+                "test_start": str(s.test_idx[0]),
+                "test_end": str(s.test_idx[-1]),
+            })
+
+        if not per_fold:
+            continue
+
+        fits = [pf["fitness"] for pf in per_fold]
+        sharps = [pf["sharpe"] for pf in per_fold]
+        results.append({
+            "genome_id": ev["genome_id"],
+            "fitness_is": float(ev["fitness"]),
+            "sharpe_is": float(ev["sharpe"]),
+            "folds": per_fold,
+            "fitness_oos_mean": statistics.mean(fits),
+            "fitness_oos_min": min(fits),
+            "fitness_oos_max": max(fits),
+            "fitness_oos_std": statistics.pstdev(fits) if len(fits) > 1 else 0.0,
+            "sharpe_oos_mean": statistics.mean(sharps),
+            "sharpe_oos_min": min(sharps),
+            "robust_score": min(fits),  # min across folds = pessimismo
+        })
+
+    # Ranking finale: per robust_score (min fitness) decrescente.
+    results.sort(key=lambda r: r["robust_score"], reverse=True)
+
+    print()
+    print(f"{'='*120}")
+    print(f"VALIDATION RESULTS ({len(results)} genomes, {len(splits)} folds)")
+    print(f"{'='*120}")
+    print(
+        f"{'rank':>4} {'genome':12} {'fit_is':>8} {'sh_is':>7} "
+        f"{'fit_mean':>9} {'fit_min':>8} {'fit_max':>8} {'fit_std':>8} "
+        f"{'sh_mean':>8} {'sh_min':>8} {'robust':>7}"
+    )
+    print("-" * 120)
+    for rank, r in enumerate(results, 1):
+        print(
+            f"{rank:>4} {r['genome_id'][:12]:12} "
+            f"{r['fitness_is']:>8.4f} {r['sharpe_is']:>7.3f} "
+            f"{r['fitness_oos_mean']:>9.4f} {r['fitness_oos_min']:>8.4f} "
+            f"{r['fitness_oos_max']:>8.4f} {r['fitness_oos_std']:>8.4f} "
+            f"{r['sharpe_oos_mean']:>8.3f} {r['sharpe_oos_min']:>8.3f} "
+            f"{r['robust_score']:>7.4f}"
+        )
+
+    if results:
+        winner = results[0]
+        print()
+        print(f"ROBUST WINNER: {winner['genome_id']}")
+        print(f"  fitness_is={winner['fitness_is']:.4f}, "
+              f"fitness_oos_min={winner['fitness_oos_min']:.4f}, "
+              f"fitness_oos_mean={winner['fitness_oos_mean']:.4f}")
+        print(f"  sharpe_is={winner['sharpe_is']:.3f}, "
+              f"sharpe_oos_min={winner['sharpe_oos_min']:.3f}")
+        print(f"  per-fold breakdown:")
+        for pf in winner["folds"]:
+            print(
+                f"    fold {pf['fold']} [{pf['test_start'][:10]} .. {pf['test_end'][:10]}]: "
+                f"fit={pf['fitness']:.4f} sharpe={pf['sharpe']:.3f} "
+                f"ret={pf['return']:.3f} n_trades={pf['n_trades']}"
+            )
+
+    if args.output_json:
+        payload = {
+            "run_id": args.run_id,
+            "run_name": run["name"],
+            "n_folds": len(splits),
+            "top_k_requested": args.top_k,
+            "top_k_evaluated": len(results),
+            "symbol": args.symbol,
+            "timeframe": args.timeframe,
+            "start": args.start,
+            "end": args.end,
+            "ohlcv_bars": len(ohlcv),
+            "results": results,
+        }
+        args.output_json.write_text(json.dumps(payload, indent=2, default=str))
+        print(f"\nResults saved to: {args.output_json}")
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1,112 @@
+"""Per-year breakdown delle 4 strategie: 2 NEW (BTC 238e4812 + ETH c04dff7086)
+ 2 OLD freezate (btc_9cf506b8 hardened-001 + eth_facd6af85d5d).
+
+Backtest anno-per-anno (2019-2025) sul tick di discovery di ciascuna strategia.
+Output: trade, wins/losses, win%, return%, max DD%, Sharpe per ogni anno.
+"""
+from __future__ import annotations
+
+from datetime import datetime
+from pathlib import Path
+
+from multi_swarm_core.backtest.engine import BacktestEngine
+from multi_swarm_core.cerbero.client import CerberoClient
+from multi_swarm_core.config import load_settings
+from multi_swarm_core.data.cerbero_ohlcv import CerberoOHLCVLoader, OHLCVRequest
+from multi_swarm_core.metrics.basic import max_drawdown, sharpe_ratio, total_return
+from multi_swarm_core.protocol.compiler import compile_strategy
+from multi_swarm_core.protocol.parser import parse_strategy
+
+
+STRATEGIES = [
+    # (label, path, symbol, timeframe)
+    ("BTC NEW (238e4812, paper attuale)", "btc_238e4812.json", "BTC-PERPETUAL", "1h"),
+    ("BTC OLD (9cf506b8, hardened-001 prev paper)", "archive/btc_9cf506b8.json", "BTC-PERPETUAL", "1h"),
+    ("ETH NEW (c04dff7086, paper attuale)", "eth_c04dff7086.json", "ETH-PERPETUAL", "5m"),
+    ("ETH OLD (facd6af85d5d, prev paper)", "archive/eth_facd6af85d5d.json", "ETH-PERPETUAL", "1h"),
+]
+
+YEARS = [
+    ("2019", "2019-01-01T00:00:00+00:00", "2020-01-01T00:00:00+00:00"),
+    ("2020", "2020-01-01T00:00:00+00:00", "2021-01-01T00:00:00+00:00"),
+    ("2021", "2021-01-01T00:00:00+00:00", "2022-01-01T00:00:00+00:00"),
+    ("2022", "2022-01-01T00:00:00+00:00", "2023-01-01T00:00:00+00:00"),
+    ("2023", "2023-01-01T00:00:00+00:00", "2024-01-01T00:00:00+00:00"),
+    ("2024", "2024-01-01T00:00:00+00:00", "2025-01-01T00:00:00+00:00"),
+    ("2025", "2025-01-01T00:00:00+00:00", "2026-01-01T00:00:00+00:00"),
+]
+
+
+def main() -> None:
+    settings = load_settings()
+    token = (
+        settings.cerbero_mainnet_token.get_secret_value()
+        if settings.cerbero_mainnet_token
+        else settings.cerbero_testnet_token.get_secret_value()
+    )
+    cerbero = CerberoClient(
+        base_url=settings.cerbero_base_url,
+        token=token,
+        bot_tag=settings.cerbero_bot_tag,
+    )
+    loader = CerberoOHLCVLoader(client=cerbero, cache_dir=settings.series_dir)
+    engine = BacktestEngine(fees_bp=5.0)
+    strategies_dir = Path("/app/strategies")
+
+    for label, fname, symbol, timeframe in STRATEGIES:
+        path = strategies_dir / fname
+        strat = parse_strategy(path.read_text())
+
+        # Carica intero range una volta sola
+        ohlcv = loader.load(OHLCVRequest(
+            symbol=symbol, timeframe=timeframe,
+            start=datetime.fromisoformat("2018-09-01T00:00:00+00:00"),
+            end=datetime.fromisoformat("2026-01-01T00:00:00+00:00"),
+        ))
+
+        print(f"\n{'=' * 110}")
+        print(f">>> {label}")
+        print(f"    symbol={symbol} timeframe={timeframe} | {len(ohlcv)} bars total")
+        print(f"    {'year':<6} {'bars':>6} {'trades':>7} {'wins':>5} {'losses':>7} {'win%':>6} {'avg_w':>10} {'avg_l':>10} {'ret':>8} {'maxDD':>7} {'sharpe':>7}")
+
+        sum_ret = 0.0
+        sum_trades = 0
+        sum_wins = 0
+        for year_label, start, end in YEARS:
+            mask = (ohlcv.index >= datetime.fromisoformat(start)) & (ohlcv.index < datetime.fromisoformat(end))
+            slice_df = ohlcv[mask]
+            if len(slice_df) == 0:
+                continue
+            try:
+                signal_fn = compile_strategy(strat)
+                signals = signal_fn(slice_df)
+                bt = engine.run(slice_df, signals)
+            except Exception as e:
+                print(f"    {year_label:<6} ERROR: {e}")
+                continue
+            trades = bt.trades
+            n = len(trades)
+            wins = [t.net_pnl for t in trades if t.net_pnl > 0]
+            losses = [t.net_pnl for t in trades if t.net_pnl <= 0]
+            nw, nl = len(wins), len(losses)
+            wr = (nw / n * 100) if n else 0.0
+            aw = (sum(wins) / nw) if nw else 0.0
+            al = (sum(losses) / nl) if nl else 0.0
+            if n > 0:
+                notional = float(slice_df["close"].iloc[0])
+                eq = (bt.equity_curve / notional) + 1.0
+                ret = total_return(eq)
+                dd = max_drawdown(eq)
+                sr = sharpe_ratio(bt.returns, periods_per_year=8760)
+            else:
+                ret = dd = sr = 0.0
+            print(f"    {year_label:<6} {len(slice_df):>6} {n:>7} {nw:>5} {nl:>7} {wr:>5.1f}% {aw:>10.1f} {al:>10.1f} {ret:>7.2%} {dd:>6.2%} {sr:>7.3f}")
+            sum_ret += ret
+            sum_trades += n
+            sum_wins += nw
+        overall_wr = (sum_wins / sum_trades * 100) if sum_trades else 0.0
+        print(f"    {'='*5} TOTALS 7y:                 {sum_trades:>7} {sum_wins:>5} {sum_trades-sum_wins:>7} {overall_wr:>5.1f}%                       cum_ret={sum_ret:+.2%}")
+
+
+if __name__ == "__main__":
+    main()
@@ -172,10 +172,11 @@ class AdversarialAgent:
                )
            )

-        # Fees-eat-alpha: gross_pnl > 0 ma fees > 50% del lordo.
+        # Fees-eat-alpha: gross_pnl > 0 ma fees > soglia del lordo.
        # La strategia ha edge teorico ma il margine viene mangiato dai
        # costi di transazione: non sostenibile in produzione.
-        # Se gross_pnl <= 0 il check non si applica (gia' perdente).
+        # Se gross_pnl <= 0 il check non si applica (la condizione e' coperta
+        # da ``negative_net_pnl`` sotto).
        gross_pnl = sum(t.gross_pnl for t in result.trades)
        total_fees = sum(t.fees for t in result.trades)
        if gross_pnl > 0 and total_fees / gross_pnl > self._fees_eat_alpha_threshold:
@@ -190,4 +191,22 @@ class AdversarialAgent:
                )
            )

+        # Negative-net-pnl: somma di ``trade.net_pnl`` < 0 sul training.
+        # Cattura sia il caso "gross negativo" (no edge direzionale) sia il
+        # caso "gross positivo ma fees superiori a gross" (edge insufficiente).
+        # Sintesi del net-after-fees su finestra continua: deal-breaker, non
+        # negoziabile via soft penalty.
+        net_pnl = gross_pnl - total_fees
+        if net_pnl < 0:
+            report.findings.append(
+                Finding(
+                    name="negative_net_pnl",
+                    severity=Severity.HIGH,
+                    detail=(
+                        f"Net PnL ${net_pnl:.2f} < 0 after fees over {n_bars} bars; "
+                        f"gross ${gross_pnl:.2f}, fees ${total_fees:.2f}"
+                    ),
+                )
+            )
+
        return report
@@ -4,11 +4,9 @@ import re
 from dataclasses import dataclass, field
 from typing import Any

-import openai
-
 from ..genome.hypothesis import HypothesisAgentGenome
 from ..genome.prompt_library import PromptLibrary
-from ..llm.client import CompletionResult, EmptyCompletionError, LLMClient
+from ..llm.client import CompletionResult, EmptyCompletionError, LLMClient, OpusAgentError, OpusAgentTransientError
 from ..protocol.parser import ParseError, Strategy, parse_strategy
 from ..protocol.validator import ValidationError, validate_strategy

@@ -181,6 +179,12 @@ def _build_system_prompt(lib: PromptLibrary, genome: HypothesisAgentGenome) -> s
    parts.append("")
    # 3. Grammar spec (core scaffold)
    parts.append(_SYSTEM_GRAMMAR_SPEC)
+    # 3b. Custom indicators spec (da prompts.json, opzionale) - estende la lista
+    # "Leaf - indicatori" con firme strategy-specific non note al core.
+    if lib.custom_indicators_spec:
+        parts.append("\nINDICATORI CUSTOM (firme aggiuntive, applica le stesse regole grammaticali):\n")
+        parts.append(lib.custom_indicators_spec)
+        parts.append("")
    # 4. Pattern guidance (da prompts.json, opzionale)
    if lib.pattern_guidance:
        parts.append(
@@ -284,19 +288,12 @@ def _render_focus_block(keys: list[str], market: MarketSummary) -> str:


 _RETRY_TEMPLATE = """\
-{original_user}
+Il JSON che hai generato contiene un errore: {previous_error}

--- TENTATIVO PRECEDENTE FALLITO ---
-Output: {previous_raw}
-Errore: {previous_error}
---
-Correggi l'errore e rispondi di nuovo con un singolo oggetto JSON valido
-dentro fence ```json...```, seguendo strettamente lo schema fornito nel
-SYSTEM message.
+Correggi e rispondi di nuovo con un singolo oggetto JSON valido
+dentro fence ```json...```, seguendo strettamente lo schema fornito.
 """

-_RETRY_RAW_TRUNCATE = 800
-

 _JSON_FENCE_RE = re.compile(
    r"```(?:json)?\s*(\{[\s\S]*\})\s*```",
@@ -425,34 +422,39 @@ class HypothesisAgent:
        errors: list[str] = []
        last_raw = ""
        max_attempts = 1 + self._max_retries
+        session_id: str | None = None

+        try:
            for attempt in range(max_attempts):
                if attempt == 0:
                    user = original_user
+                    req_session_id = "new"
                else:
-                truncated = last_raw[:_RETRY_RAW_TRUNCATE]
-                user = _RETRY_TEMPLATE.format(
-                    original_user=original_user,
-                    previous_raw=truncated,
-                    previous_error=errors[-1],
-                )
+                    user = _RETRY_TEMPLATE.format(previous_error=errors[-1])
+                    req_session_id = session_id or "new"

                try:
-                completion = self._llm.complete(genome, system=system, user=user)
+                    completion = self._llm.complete(
+                        genome, system=system, user=user,
+                        session_id=req_session_id,
+                        summarize=True,
+                    )
                except EmptyCompletionError as e:
-                # LLM esaurito retry tenacity senza una risposta. Tratta come
-                # parse-fail "empty" e ritenta nel loop esterno (max_attempts).
                    errors.append(f"empty_completion: {e}")
                    last_raw = ""
                    continue
-            except openai.RateLimitError as e:
-                # Provider upstream rate limited oltre i retry tenacity.
-                # Marca genome come fallito senza propagare l'eccezione al run.
-                errors.append(f"rate_limit: {e}")
+                except OpusAgentTransientError as e:
+                    errors.append(f"transient_error: {e}")
+                    last_raw = ""
+                    continue
+                except OpusAgentError as e:
+                    errors.append(f"opus_agent_error: {e}")
                    last_raw = ""
                    continue
                completions.append(completion)
                last_raw = completion.text
+                if completion.session_id:
+                    session_id = completion.session_id

                strategy, err = _try_parse(completion.text)
                if strategy is not None:
@@ -465,6 +467,9 @@ class HypothesisAgent:
                    )
                assert err is not None
                errors.append(err)
+        finally:
+            if session_id:
+                self._llm.close_session(session_id)

        chained = " | ".join(
            f"attempt {i + 1}: {e}" for i, e in enumerate(errors)
@@ -2,6 +2,7 @@ from __future__ import annotations

 from dataclasses import dataclass

+import numpy as np
 import pandas as pd  # type: ignore[import-untyped]

 from .orders import Position, Side, Trade
@@ -28,74 +29,110 @@ class BacktestEngine:
        self.fees_bp = fees_bp

    def run(self, ohlcv: pd.DataFrame, signals: pd.Series) -> BacktestResult:
+        n = len(ohlcv)
+        if n == 0:
+            empty = pd.Series([], dtype=float)
+            return BacktestResult(equity_curve=empty, returns=empty, trades=[])
+
        signals = signals.reindex(ohlcv.index).ffill().fillna(Side.FLAT)
-
        # Esecuzione con delay 1: segnale a t-1 esegue a open di t.
-        shifted = [Side.FLAT, *list(signals.iloc[:-1])]
-        executed_side = pd.Series(shifted, index=ohlcv.index, dtype=object)
+        executed = pd.Series(
+            [Side.FLAT, *list(signals.iloc[:-1])],
+            index=ohlcv.index,
+            dtype=object,
+        )
+
+        # Codifica side in int per vectorizzazione: 0=FLAT, +1=LONG, -1=SHORT.
+        side_code = np.where(
+            executed.values == Side.LONG, 1,
+            np.where(executed.values == Side.SHORT, -1, 0),
+        ).astype(np.int8)
+        opens = ohlcv["open"].to_numpy(dtype=np.float64)
+        closes = ohlcv["close"].to_numpy(dtype=np.float64)
+        ts_index = ohlcv.index
+
+        # Identifica transizioni: punto in cui side[i] != side[i-1] (con side[-1]=0).
+        prev = np.concatenate(([0], side_code[:-1]))
+        change = side_code != prev
+
+        # Indici di entry (cambio verso side != 0).
+        entry_idxs = np.flatnonzero(change & (side_code != 0))
+        # Indici di chiusura: per ogni entry, il prossimo indice dove side[i] != side_entry.
+        # Vectorized: per ogni entry_idx, cerca change & side != side_entry oltre l'entry.

-        position: Position | None = None
-        position_entry_ts: pd.Timestamp | None = None
        trades: list[Trade] = []
-        equity = 0.0
-        equity_history: list[float] = []
-        returns_history: list[float] = []
-        prev_equity = 0.0
+        # realized_pnl[t]: PnL netto cumulato dopo le chiusure avvenute a OPEN di t.
+        realized_pnl = np.zeros(n, dtype=np.float64)
+        fees_rate = self.fees_bp / 10000.0
+        size = 1.0

-        for ts, row in ohlcv.iterrows():
-            target_side = executed_side.loc[ts]
-            current_side = position.side if position else Side.FLAT
+        # Posizione corrente all'inizio di ogni indice t (prima di applicare il transitorio):
+        # used per MtM computation. open_side_at_t / open_entry_at_t.
+        open_side = np.zeros(n, dtype=np.int8)
+        open_entry = np.zeros(n, dtype=np.float64)

-            if target_side != current_side:
-                if position is not None:
-                    assert position_entry_ts is not None
-                    trade = Trade(
-                        entry_ts=position_entry_ts,
-                        exit_ts=ts,
-                        side=position.side,
-                        size=position.size,
-                        entry_price=position.entry_price,
-                        exit_price=float(row["open"]),
+        for entry_idx in entry_idxs:
+            entry_side = int(side_code[entry_idx])
+            entry_price = opens[entry_idx]
+            # Cerca exit: primo indice > entry_idx dove side differisce.
+            after = side_code[entry_idx + 1:]
+            rel = np.flatnonzero(after != entry_side)
+            if rel.size > 0:
+                exit_idx = entry_idx + 1 + int(rel[0])
+                exit_price = opens[exit_idx]
+                exit_ts = ts_index[exit_idx]
+                gross = entry_side * (exit_price - entry_price) * size
+                fees = fees_rate * size * (entry_price + exit_price)
+                net = gross - fees
+                # La chiusura avviene a open[exit_idx]: dal bar exit_idx in poi il
+                # PnL e' realizzato (non piu' MtM).
+                realized_pnl[exit_idx:] += net
+                # Posizione aperta in [entry_idx, exit_idx-1].
+                open_side[entry_idx:exit_idx] = entry_side
+                open_entry[entry_idx:exit_idx] = entry_price
+                trades.append(Trade(
+                    entry_ts=ts_index[entry_idx],
+                    exit_ts=exit_ts,
+                    side=Side.LONG if entry_side == 1 else Side.SHORT,
+                    size=size,
+                    entry_price=entry_price,
+                    exit_price=exit_price,
                    fees_bp=self.fees_bp,
-                    )
-                    trades.append(trade)
-                    equity += trade.net_pnl
-                    position = None
-                    position_entry_ts = None
-                if target_side in (Side.LONG, Side.SHORT):
-                    position = Position(
-                        side=target_side, entry_price=float(row["open"]), size=1.0
-                    )
-                    position_entry_ts = ts
-
-            mark = float(row["close"])
-            mtm = position.unrealized_pnl(mark) if position else 0.0
-            current_equity = equity + mtm
-            equity_history.append(current_equity)
-            returns_history.append(current_equity - prev_equity)
-            prev_equity = current_equity
-
-        if position is not None:
-            assert position_entry_ts is not None
-            last_ts = ohlcv.index[-1]
-            last_close = float(ohlcv["close"].iloc[-1])
-            trade = Trade(
-                entry_ts=position_entry_ts,
-                exit_ts=last_ts,
-                side=position.side,
-                size=position.size,
-                entry_price=position.entry_price,
+                ))
+            else:
+                # Ultima posizione ancora aperta: chiusura forced a close[-1].
+                # Parita' col loop legacy: MtM su [entry_idx, n-1), realized totale
+                # SOLO al bar n-1 (legacy fa equity_history[-1] = equity).
+                last_close = closes[-1]
+                gross = entry_side * (last_close - entry_price) * size
+                fees = fees_rate * size * (entry_price + last_close)
+                net = gross - fees
+                if entry_idx < n - 1:
+                    open_side[entry_idx:n - 1] = entry_side
+                    open_entry[entry_idx:n - 1] = entry_price
+                realized_pnl[-1] += net
+                trades.append(Trade(
+                    entry_ts=ts_index[entry_idx],
+                    exit_ts=ts_index[-1],
+                    side=Side.LONG if entry_side == 1 else Side.SHORT,
+                    size=size,
+                    entry_price=entry_price,
                    exit_price=last_close,
                    fees_bp=self.fees_bp,
-            )
-            trades.append(trade)
-            equity += trade.net_pnl
-            equity_history[-1] = equity
-            if len(returns_history) >= 2:
-                returns_history[-1] = equity - equity_history[-2]
+                ))
+
+        # MtM unrealized per ogni bar in cui c'e' una posizione aperta.
+        mtm = open_side.astype(np.float64) * (closes - open_entry) * size
+        equity_arr = realized_pnl + mtm
+        # Returns = first diff dell'equity (col loop legacy il primo bar e' equity[0]-0).
+        returns_arr = np.concatenate(([equity_arr[0]], np.diff(equity_arr)))

        return BacktestResult(
-            equity_curve=pd.Series(equity_history, index=ohlcv.index, name="equity"),
-            returns=pd.Series(returns_history, index=ohlcv.index, name="returns"),
+            equity_curve=pd.Series(equity_arr, index=ts_index, name="equity"),
+            returns=pd.Series(returns_arr, index=ts_index, name="returns"),
            trades=trades,
        )
+
+
+# Lo facade Position re-export e' tenuto per backward-compat con import legacy.
+__all__ = ["BacktestEngine", "BacktestResult", "Position", "Side", "Signal", "Trade"]
@@ -23,14 +23,14 @@ class Settings(BaseSettings):
    cerbero_mainnet_token: SecretStr | None = None
    cerbero_bot_tag: str = "swarm-poc-phase1"

-    openrouter_api_key: SecretStr
+    opus_agent_api_key: SecretStr
+    opus_agent_base_url: str = "https://opus-agent.tielogic.xyz"

-    llm_model_tier_s: str = "google/gemini-3-flash-preview"
-    llm_model_tier_a: str = "deepseek/deepseek-v4-flash"
-    llm_model_tier_b: str = "deepseek/deepseek-v4-flash"
-    llm_model_tier_c: str = "qwen/qwen-2.5-72b-instruct"
-    llm_model_tier_d: str = "openai/gpt-oss-20b"
-    openrouter_base_url: str = "https://openrouter.ai/api/v1"
+    llm_model_tier_s: str = "claude-opus-4-7"
+    llm_model_tier_a: str = "claude-opus-4-7"
+    llm_model_tier_b: str = "claude-sonnet-4-6"
+    llm_model_tier_c: str = "claude-sonnet-4-6"
+    llm_model_tier_d: str = "claude-haiku-4-5-20251001"

    run_name: str = "phase1-spike-001"
    data_dir: Path = Field(default=Path("./data"))
@@ -263,7 +263,8 @@ def index() -> None:
        refresh()

    select.on_value_change(on_select_change)
-    ui.timer(REFRESH_INTERVAL_S, refresh)
+    _timer = ui.timer(REFRESH_INTERVAL_S, refresh)
+    ui.context.client.on_disconnect(_timer.deactivate)
    refresh()


@@ -353,7 +354,8 @@ def convergence() -> None:
        refresh()

    select.on_value_change(on_select_change)
-    ui.timer(REFRESH_INTERVAL_S, refresh)
+    _timer = ui.timer(REFRESH_INTERVAL_S, refresh)
+    ui.context.client.on_disconnect(_timer.deactivate)
    refresh()


@@ -535,7 +537,8 @@ def genomes() -> None:
    select.on_value_change(on_select_change)
    top_k_select.on_value_change(lambda _: refresh())
    top_table.on("selection", on_row_selected)
-    ui.timer(REFRESH_INTERVAL_S, refresh)
+    _timer = ui.timer(REFRESH_INTERVAL_S, refresh)
+    ui.context.client.on_disconnect(_timer.deactivate)
    refresh()


@@ -8,11 +8,11 @@ from typing import Any


 class ModelTier(StrEnum):
-    S = "S"  # top-tier reasoning (Opus / equivalent) via Anthropic
-    A = "A"  # premium override via Anthropic
-    B = "B"  # Sonnet 4.6 via Anthropic
-    C = "C"  # Qwen 2.5 72B via OpenRouter
-    D = "D"  # ultra-economic (Llama / cheap models) via OpenRouter
+    S = "S"  # top-tier reasoning → opus via OpusAgent
+    A = "A"  # premium → opus via OpusAgent
+    B = "B"  # standard → sonnet via OpusAgent
+    C = "C"  # default GA → sonnet via OpusAgent
+    D = "D"  # economic → haiku via OpusAgent


@dataclass
@@ -101,6 +101,10 @@ class PromptLibrary:
    domain_warnings: str = field(default="")  # opzionale: warning di dominio (es. crypto 24/7)
    anti_patterns: str = field(default="")       # NEW v3.1: lista esplicita di pattern da evitare
    output_priorities: str = field(default="")   # NEW v3.1: trade-off espliciti di output
+    # NEW v3.2: firme formali (arity + range) di indicatori strategy-specific
+    # non noti al core. Iniettato in _SYSTEM_GRAMMAR_SPEC come estensione della
+    # lista "Leaf - indicatori" per consentire alla LLM di chiamarli correttamente.
+    custom_indicators_spec: str = field(default="")

    def __post_init__(self) -> None:
        if not self.styles:
@@ -120,6 +124,7 @@ class PromptLibrary:
            ("domain_warnings", self.domain_warnings),
            ("anti_patterns", self.anti_patterns),
            ("output_priorities", self.output_priorities),
+            ("custom_indicators_spec", self.custom_indicators_spec),
        ):
            if not isinstance(value, str):
                raise PromptLibraryError(
@@ -142,6 +147,7 @@ class PromptLibrary:
            domain_warnings="",
            anti_patterns="",
            output_priorities="",
+            custom_indicators_spec="",
        )

    @classmethod
@@ -204,6 +210,12 @@ class PromptLibrary:
            raise PromptLibraryError(f"anti_patterns deve essere stringa, non {type(anti_patterns_raw)}")
        if not isinstance(output_priorities_raw, str):
            raise PromptLibraryError(f"output_priorities deve essere stringa, non {type(output_priorities_raw)}")
+        # Parse new optional top-level field (v3.2)
+        custom_indicators_spec_raw = data.get("custom_indicators_spec", "")
+        if not isinstance(custom_indicators_spec_raw, str):
+            raise PromptLibraryError(
+                f"custom_indicators_spec deve essere stringa, non {type(custom_indicators_spec_raw)}"
+            )

        return cls(
            styles=styles,
@@ -214,6 +226,7 @@ class PromptLibrary:
            domain_warnings=domain_warnings,
            anti_patterns=anti_patterns_raw,
            output_priorities=output_priorities_raw,
+            custom_indicators_spec=custom_indicators_spec_raw,
        )

    @property
@@ -1,10 +1,12 @@
 from __future__ import annotations

+import hashlib
+import logging
+import threading
+import time
 from dataclasses import dataclass
-from typing import Any

-import openai
-from openai import OpenAI
+import httpx
 from tenacity import (
    retry,
    retry_if_exception_type,
@@ -14,26 +16,33 @@ from tenacity import (

 from ..genome.hypothesis import HypothesisAgentGenome, ModelTier

-# Modelli configurati per Phase 1 — tutti via OpenRouter
-MODEL_TIER_S = "google/gemini-3-flash-preview"
-MODEL_TIER_A = "deepseek/deepseek-v4-flash"
-MODEL_TIER_B = "deepseek/deepseek-v4-flash"
-MODEL_TIER_C = "qwen/qwen-2.5-72b-instruct"
-MODEL_TIER_D = "openai/gpt-oss-20b"
-OPENROUTER_BASE_URL = "https://openrouter.ai/api/v1"
+logger = logging.getLogger(__name__)
+
+MODEL_TIER_MAP: dict[ModelTier, str] = {
+    ModelTier.S: "claude-opus-4-7",
+    ModelTier.A: "claude-opus-4-7",
+    ModelTier.B: "claude-sonnet-4-6",
+    ModelTier.C: "claude-sonnet-4-6",
+    ModelTier.D: "claude-haiku-4-5-20251001",
+}
+

 class EmptyCompletionError(RuntimeError):
    pass


-# Errori transient: retry. Auth/InvalidRequest: NO retry.
-# RateLimitError (HTTP 429) ora retryable: provider OpenRouter come DeepInfra
-# applicano rate limit upstream temporaneo, recuperabile con backoff.
+class OpusAgentError(RuntimeError):
+    pass
+
+
+class OpusAgentTransientError(RuntimeError):
+    pass
+
+
 _RETRYABLE_EXCEPTIONS: tuple[type[BaseException], ...] = (
-    openai.APIConnectionError,
-    openai.APITimeoutError,
-    openai.InternalServerError,
-    openai.RateLimitError,
+    httpx.ConnectError,
+    httpx.TimeoutException,
+    OpusAgentTransientError,
    EmptyCompletionError,
 )

@@ -45,50 +54,99 @@ class CompletionResult:
    output_tokens: int
    tier: ModelTier
    model: str
+    session_id: str | None = None


 class LLMClient:
-    # Provider OpenRouter da escludere di default. Novita rifiuta /completions
-    # endpoint per modelli Qwen 2.x — vedi bug 2026-05-12.
-    DEFAULT_PROVIDER_IGNORE: tuple[str, ...] = ("Novita",)
-
    def __init__(
        self,
-        openrouter_api_key: str,
-        model_tier_s: str = MODEL_TIER_S,
-        model_tier_a: str = MODEL_TIER_A,
-        model_tier_b: str = MODEL_TIER_B,
-        model_tier_c: str = MODEL_TIER_C,
-        model_tier_d: str = MODEL_TIER_D,
-        openrouter_base_url: str = OPENROUTER_BASE_URL,
-        provider_ignore: tuple[str, ...] | None = None,
+        opus_agent_api_key: str,
+        opus_agent_base_url: str = "https://opus-agent.tielogic.xyz",
+        model_tier_s: str = "claude-opus-4-7",
+        model_tier_a: str = "claude-opus-4-7",
+        model_tier_b: str = "claude-sonnet-4-6",
+        model_tier_c: str = "claude-sonnet-4-6",
+        model_tier_d: str = "claude-haiku-4-5-20251001",
+        poll_interval: float = 3.0,
+        poll_timeout: float = 180.0,
    ) -> None:
-        self.model_tier_s = model_tier_s
-        self.model_tier_a = model_tier_a
-        self.model_tier_b = model_tier_b
-        self.model_tier_c = model_tier_c
-        self.model_tier_d = model_tier_d
-        self.openrouter_base_url = openrouter_base_url
-        self._provider_ignore = (
-            provider_ignore if provider_ignore is not None else self.DEFAULT_PROVIDER_IGNORE
-        )
-        self._tier_models: dict[ModelTier, str] = {
+        self._base_url = opus_agent_base_url.rstrip("/")
+        self._api_key = opus_agent_api_key
+        self._tier_map: dict[ModelTier, str] = {
            ModelTier.S: model_tier_s,
            ModelTier.A: model_tier_a,
            ModelTier.B: model_tier_b,
            ModelTier.C: model_tier_c,
            ModelTier.D: model_tier_d,
        }
-        # Timeout esplicito (60s) per prevenire hang infinito su connessioni
-        # stallate. Tenacity retry su APITimeoutError gestisce il recovery.
-        self._client = OpenAI(
-            api_key=openrouter_api_key,
-            base_url=openrouter_base_url,
-            timeout=60.0,
+        self._poll_interval = poll_interval
+        self._poll_timeout = poll_timeout
+        self._topic_cache: dict[str, str] = {}
+        self._topic_lock = threading.Lock()
+        self._client = httpx.Client(
+            base_url=self._base_url,
+            headers={"X-Api-Key": self._api_key, "Content-Type": "application/json"},
+            timeout=30.0,
        )

+    def _get_or_create_topic(self, system_prompt: str) -> str:
+        prompt_hash = hashlib.sha256(system_prompt.encode()).hexdigest()[:16]
+        if prompt_hash in self._topic_cache:
+            return self._topic_cache[prompt_hash]
+
+        with self._topic_lock:
+            if prompt_hash in self._topic_cache:
+                return self._topic_cache[prompt_hash]
+
+            topic_name = f"swarm-{prompt_hash}"
+            resp = self._client.post("/api/topics", json={
+                "name": topic_name,
+                "system_prompt": system_prompt,
+            })
+
+            if resp.status_code == 409:
+                list_resp = self._client.get("/api/topics")
+                list_resp.raise_for_status()
+                for topic in list_resp.json()["data"]:
+                    if topic["name"] == topic_name:
+                        self._topic_cache[prompt_hash] = topic["id"]
+                        return topic["id"]
+                raise OpusAgentError(f"Topic {topic_name} conflict but not found")
+
+            if resp.status_code >= 500:
+                raise OpusAgentTransientError(f"Server error {resp.status_code}")
+            resp.raise_for_status()
+
+            topic_id = resp.json()["data"]["id"]
+            self._topic_cache[prompt_hash] = topic_id
+            logger.debug("Created topic %s -> %s", topic_name, topic_id)
+            return topic_id
+
+    def _poll_result(self, request_id: str) -> dict:
+        deadline = time.monotonic() + self._poll_timeout
+        while time.monotonic() < deadline:
+            resp = self._client.get(f"/api/requests/{request_id}")
+            resp.raise_for_status()
+            data = resp.json()["data"]
+            status = data["status"]
+            if status == "completed":
+                return data
+            if status == "failed":
+                error = data.get("error") or "unknown error"
+                raise OpusAgentError(f"Request {request_id} failed: {error}")
+            time.sleep(self._poll_interval)
+        raise OpusAgentTransientError(
+            f"Request {request_id} timed out after {self._poll_timeout}s"
+        )
+
+    def close_session(self, session_id: str) -> None:
+        try:
+            self._client.delete(f"/api/sessions/{session_id}")
+        except httpx.HTTPError:
+            logger.debug("Failed to close session %s", session_id)
+
    @retry(
-        stop=stop_after_attempt(5),
+        stop=stop_after_attempt(3),
        wait=wait_exponential(multiplier=2.0, min=2.0, max=30.0),
        retry=retry_if_exception_type(_RETRYABLE_EXCEPTIONS),
        reraise=True,
@@ -99,29 +157,45 @@ class LLMClient:
        system: str,
        user: str,
        max_tokens: int = 2000,
+        session_id: str | None = None,
+        summarize: bool = False,
    ) -> CompletionResult:
-        model = self._tier_models[genome.model_tier]
-        extra_body: dict[str, Any] = {}
-        if self._provider_ignore:
-            extra_body["provider"] = {"ignore": list(self._provider_ignore)}
-        resp = self._client.chat.completions.create(
-            model=model,
-            messages=[
-                {"role": "system", "content": system},
-                {"role": "user", "content": user},
-            ],
-            temperature=genome.temperature,
-            top_p=genome.top_p,
-            max_tokens=max_tokens,
-            extra_body=extra_body or None,
-        )
-        if not resp.choices or resp.choices[0].message.content is None:
-            raise EmptyCompletionError(f"empty response from {model}")
-        usage = resp.usage
+        model = self._tier_map[genome.model_tier]
+        topic_id = self._get_or_create_topic(system)
+
+        prompt = f"[SYSTEM]\n{system}\n\n[USER]\n{user}" if session_id in (None, "new") else user
+
+        body: dict = {
+            "topic_id": topic_id,
+            "prompt": prompt,
+            "model": model,
+        }
+        if session_id is not None:
+            body["session_id"] = session_id
+        if summarize:
+            body["summarize"] = True
+
+        resp = self._client.post("/api/requests", json=body)
+
+        if resp.status_code == 429:
+            raise OpusAgentTransientError("Rate limited")
+        if resp.status_code >= 500:
+            raise OpusAgentTransientError(f"Server error {resp.status_code}")
+        if resp.status_code != 202:
+            raise OpusAgentError(f"Unexpected status {resp.status_code}: {resp.text}")
+
+        request_id = resp.json()["data"]["id"]
+        result = self._poll_result(request_id)
+
+        text = result.get("result") or ""
+        if not text:
+            raise EmptyCompletionError(f"empty response from OpusAgent ({model})")
+
        return CompletionResult(
-            text=resp.choices[0].message.content,
-            input_tokens=usage.prompt_tokens if usage else 0,
-            output_tokens=usage.completion_tokens if usage else 0,
+            text=text,
+            input_tokens=0,
+            output_tokens=0,
            tier=genome.model_tier,
            model=model,
+            session_id=result.get("session_id"),
        )
@@ -13,6 +13,7 @@ possa leggere lo stato a run terminato (o in corso).
 from __future__ import annotations

 import random
+from concurrent.futures import ThreadPoolExecutor
 from dataclasses import dataclass, field
 from pathlib import Path

@@ -20,13 +21,13 @@ import pandas as pd  # type: ignore[import-untyped]

 from ..agents.adversarial import AdversarialAgent
 from ..agents.falsification import FalsificationAgent
-from ..agents.hypothesis import HypothesisAgent
+from ..agents.hypothesis import HypothesisAgent, HypothesisProposal, MarketSummary
 from ..agents.market_summary import build_market_summary
 from ..ga.fitness import compute_fitness
 from ..ga.initial import build_initial_population
 from ..ga.loop import GAConfig, next_generation
 from ..ga.summary import generation_summary
-from ..genome.hypothesis import ModelTier
+from ..genome.hypothesis import HypothesisAgentGenome, ModelTier
 from ..genome.mutation import set_cognitive_styles
 from ..genome.prompt_library import PromptLibrary
 from ..llm.client import LLMClient
@@ -73,6 +74,29 @@ class RunConfig:
    # i 6 builtin (PromptLibrary.default()). Tipicamente caricata da
    # strategy_crypto/prompts.json via PromptLibrary.from_json().
    prompt_library: PromptLibrary | None = None
+    # Numero di propose() LLM concorrenti per generazione. 1 = sequenziale (default,
+    # backward compat). OpusAgent processa in coda FIFO; concurrency > 1
+    # accoda piu' richieste in parallelo ma il throughput dipende dal server.
+    llm_concurrency: int = 1
+
+
+def _parallel_propose(
+    agent: HypothesisAgent,
+    genomes: list[HypothesisAgentGenome],
+    market: MarketSummary,
+    n_workers: int,
+) -> list[HypothesisProposal]:
+    """Esegue ``agent.propose()`` su una lista di genomi, opzionalmente in parallelo.
+
+    ``n_workers <= 1`` mantiene il comportamento serial originale (ordine fisso,
+    determinismo data un seed). ``n_workers > 1`` usa un thread pool: l'order
+    dei risultati e' preservato (1:1 con ``genomes``). LLMClient (OpusAgent)
+    e' thread-safe; ``PromptLibrary`` e ``HypothesisAgent`` non hanno stato mutabile.
+    """
+    if n_workers <= 1 or len(genomes) <= 1:
+        return [agent.propose(g, market) for g in genomes]
+    with ThreadPoolExecutor(max_workers=n_workers) as pool:
+        return list(pool.map(lambda g: agent.propose(g, market), genomes))


 def run_phase1(
@@ -88,10 +112,16 @@ def run_phase1(

    repo = Repository(cfg.db_path)
    repo.init_schema()
+    # Escludi prompt_library (PromptLibrary dataclass non e' JSON-serializable);
+    # salva solo i nomi degli stili per reproducibility.
    config_dict = {
-        **cfg.__dict__,
+        **{k: v for k, v in cfg.__dict__.items() if k != "prompt_library"},
        "db_path": str(cfg.db_path),
        "model_tier": cfg.model_tier.value,
+        "prompt_library_styles": (
+            list(cfg.prompt_library.cognitive_styles)
+            if cfg.prompt_library is not None else None
+        ),
    }
    run_id = repo.create_run(name=cfg.run_name, config=config_dict)

@@ -142,11 +172,20 @@ def run_phase1(

    try:
        for gen in range(cfg.n_generations):
+            # Step 1: raccogli i genomi da valutare in questa generazione (esclude
+            # elite gia' presenti nella cache fitnesses) e lancia propose() in
+            # parallelo. La sezione DB-write resta serial sotto.
+            uncached = [g for g in population if g.id not in fitnesses]
+            proposals = _parallel_propose(
+                hypothesis_agent, uncached, market, cfg.llm_concurrency
+            )
+            proposal_by_id = {g.id: p for g, p in zip(uncached, proposals, strict=True)}
+
            for genome in population:
                if genome.id in fitnesses:
                    continue  # elite gia' valutata in generazione precedente
                repo.save_genome(run_id=run_id, generation_idx=gen, genome=genome)
-                proposal = hypothesis_agent.propose(genome, market)
+                proposal = proposal_by_id[genome.id]
                # Registra costo per OGNI completion (incluse retry).
                for completion in proposal.completions:
                    cost_record = cost_tracker.record(
@@ -220,7 +259,7 @@ def run_phase1(
                            cfg.fitness_combined_alpha * fit
                            + (1.0 - cfg.fitness_combined_alpha) * fit_oos_inloop
                        )
-                    except Exception:  # noqa: BLE001
+                    except Exception:
                        pass  # fallback: usa solo IS
                repo.save_evaluation(
                    run_id=run_id,
@@ -261,7 +300,7 @@ def run_phase1(
        # WFA re-eval: i top_k genomi (by fitness in-sample > 0) vengono rivalutati
        # sul test_ohlcv. Le metriche OOS finiscono in evaluations.fitness_oos etc.
        if test_ohlcv is not None and len(test_ohlcv) >= 100:
-            from ..agents.hypothesis import _try_parse  # noqa: PLC0415
+            from ..agents.hypothesis import _try_parse

            all_evals = repo.list_evaluations(run_id)
            top_evals = sorted(
@@ -276,7 +315,7 @@ def run_phase1(
                try:
                    fals_oos = falsification_agent.evaluate(strategy, test_ohlcv)
                    adv_oos = adversarial_agent.review(strategy, test_ohlcv)
-                except Exception:  # noqa: BLE001
+                except Exception:
                    continue
                fit_oos = compute_fitness(
                    fals_oos, adv_oos,
@@ -25,6 +25,12 @@ from typing import Any
 import numpy as np
 import pandas as pd  # type: ignore[import-untyped]

+from strategy_pythagoras.indicators import (
+    candle_pattern as _public_candle_pattern,
+    fractal_mirror as _public_fractal_mirror,
+    pythagorean_ratio as _public_pythagorean_ratio,
+)
+
 from ..backtest.orders import Side
 from .parser import (
    FeatureNode,
@@ -126,6 +132,22 @@ def _ind_macd(
    return macd_line - signal_line


+def _ind_candle_pattern(df: pd.DataFrame, *params: float) -> pd.Series:
+    # Adapter: il dispatch in _eval_node fa ``fn(df, *node.params)``, ma la
+    # public API in strategy_pythagoras.indicators accetta ``params: list[float]``
+    # come singolo argomento. Re-pack qui per mantenere indicators.py testabile
+    # in isolamento.
+    return _public_candle_pattern(df, list(params))
+
+
+def _ind_pythagorean_ratio(df: pd.DataFrame, lookback: float) -> pd.Series:
+    return _public_pythagorean_ratio(df, [lookback])
+
+
+def _ind_fractal_mirror(df: pd.DataFrame, k: float, axis_int: float) -> pd.Series:
+    return _public_fractal_mirror(df, [k, axis_int])
+
+
 # Annotated as ``dict[str, Any]`` deliberately: each indicator has its own
 # arity and parameter names, so a single ``Callable`` signature would be a
 # lie. Dispatch happens in :func:`_eval_node`, which validates the verb name
@@ -139,6 +161,9 @@ INDICATOR_FNS: dict[str, Any] = {
    "realized_vol": _ind_realized_vol,
    "macd": _ind_macd,
    "macd_pct": _ind_macd_pct,
+    "candle_pattern": _ind_candle_pattern,
+    "pythagorean_ratio": _ind_pythagorean_ratio,
+    "fractal_mirror": _ind_fractal_mirror,
 }

 _TIME_FEATURE_FNS: dict[str, Callable[[pd.DatetimeIndex], pd.Series]] = {
@@ -17,7 +17,8 @@ ACTION_VALUES: frozenset[str] = frozenset(
 KIND_VALUES: frozenset[str] = frozenset({"indicator", "feature", "literal"})

 KNOWN_INDICATORS: frozenset[str] = frozenset(
-    {"sma", "sma_pct", "rsi", "atr", "atr_pct", "macd", "macd_pct", "realized_vol"}
+    {"sma", "sma_pct", "rsi", "atr", "atr_pct", "macd", "macd_pct", "realized_vol",
+     "candle_pattern", "pythagorean_ratio", "fractal_mirror"}
 )
 KNOWN_FEATURES: frozenset[str] = frozenset(
    {"open", "high", "low", "close", "volume",
@@ -39,6 +39,10 @@ INDICATOR_ARITY: dict[str, tuple[int, int]] = {
    "realized_vol": (1, 1),  # window
    "macd": (0, 3),          # fast, slow, signal (tutti opzionali)
    "macd_pct": (0, 3),      # macd/close, frazionale (per confronti con literal)
+    # Pythagoras indicators (params encoded as floats)
+    "candle_pattern": (4, 13),       # [length, sym0, ..., sym_{length-1}]
+    "pythagorean_ratio": (1, 1),     # lookback in [12,200]
+    "fractal_mirror": (2, 2),        # k in [3,12], axis_int in {0=h,1=v}
 }


@@ -110,3 +114,37 @@ def _validate_indicator(node: IndicatorNode) -> None:
        raise ValidationError(
            f"indicator '{node.name}' arity {n_params} out of [{min_p},{max_p}]"
        )
+
+    # Pythagoras-specific param semantics
+    name = node.name
+    if name == "candle_pattern":
+        length = int(node.params[0])
+        if not (3 <= length <= 12):
+            raise ValidationError(
+                f"candle_pattern length must be in [3,12], got {length}"
+            )
+        if n_params != 1 + length:
+            raise ValidationError(
+                f"candle_pattern: expected 1+length={1 + length} params, got {n_params}"
+            )
+        for i, sym in enumerate(node.params[1:], start=1):
+            sym_int = int(sym)
+            if sym_int not in (0, 1, 2):
+                raise ValidationError(
+                    f"candle_pattern sym[{i - 1}] must be 0(U)/1(D)/2(doji), got {sym}"
+                )
+    elif name == "pythagorean_ratio":
+        lookback = int(node.params[0])
+        if not (12 <= lookback <= 200):
+            raise ValidationError(
+                f"pythagorean_ratio lookback in [12,200], got {lookback}"
+            )
+    elif name == "fractal_mirror":
+        k = int(node.params[0])
+        if not (3 <= k <= 12):
+            raise ValidationError(f"fractal_mirror k must be in [3,12], got {k}")
+        axis_int = int(node.params[1])
+        if axis_int not in (0, 1):
+            raise ValidationError(
+                f"fractal_mirror axis must be 0(h)/1(v), got {axis_int}"
+            )
@@ -108,7 +108,13 @@ def test_e2e_wfa_populates_fitness_oos(
    fake_llm,
    mocker,
 ):
-    """WFA: train_split=0.7 → top genomi devono avere fitness_oos popolato."""
+    """WFA: train_split=0.7 → top genomi devono avere fitness_oos popolato.
+
+    Usa fitness v2 con hard-kill minimale (solo no_trades): il fixture sintetico
+    non produce strategie profittevoli, quindi i check aggressivi
+    fees_eat_alpha/negative_net_pnl azzererebbero tutti i genomi rendendo
+    inverificabile il wiring WFA.
+    """
    cfg = RunConfig(
        run_name="e2e-wfa-test",
        population_size=5,
@@ -125,6 +131,7 @@ def test_e2e_wfa_populates_fitness_oos(
        db_path=tmp_path / "runs.db",
        wfa_train_split=0.7,
        wfa_top_k=3,
+        fitness_hard_kill_findings=("no_trades",),
    )
    run_id = run_phase1(cfg, ohlcv=synthetic_ohlcv, llm=fake_llm)
    repo = Repository(db_path=tmp_path / "runs.db")
@@ -3,7 +3,6 @@ import json
 import numpy as np
 import pandas as pd
 import pytest
-
 from multi_swarm_core.agents.adversarial import (
    AdversarialAgent,
    AdversarialReport,
@@ -54,7 +53,10 @@ def test_degenerate_always_long_flagged(ohlcv: pd.DataFrame) -> None:
    assert any(f.name == "degenerate" and f.severity == Severity.HIGH for f in report.findings)


-def test_no_findings_on_reasonable_strategy(ohlcv: pd.DataFrame) -> None:
+def test_rsi_mean_reversion_loses_money_on_synthetic_data(ohlcv: pd.DataFrame) -> None:
+    """RSI mean-reversion sul fixture sintetico ha net negativo: deve firare
+    negative_net_pnl (deal-breaker). Conferma che il check cattura strategie
+    che perdono sul training, indipendentemente dal motivo (no edge / fees)."""
    src = json.dumps(
        {
            "rules": [
@@ -84,8 +86,59 @@ def test_no_findings_on_reasonable_strategy(ohlcv: pd.DataFrame) -> None:
    ast = parse_strategy(src)
    agent = AdversarialAgent()
    report = agent.review(ast, ohlcv)
+    assert any(
+        f.name == "negative_net_pnl" and f.severity == Severity.HIGH
+        for f in report.findings
+    )
+
+
+def test_profitable_strategy_no_high_findings(
+    monkeypatch: pytest.MonkeyPatch, ohlcv: pd.DataFrame
+) -> None:
+    """Sanity test: una strategia con gross > 0 e fees << gross + n_trades ragionevole
+    + signal misto non deve produrre nessun finding HIGH."""
+    n = 15
+    # entry=100 exit=110 gross=10 per trade, fees a 5bp -> 0.105 per trade
+    # totali: gross=150, fees=1.575 -> net=+148.4
+    fake_trades = [
+        _make_trade(
+            ohlcv.index[i * 30],
+            ohlcv.index[i * 30 + 1],
+            entry_price=100.0,
+            exit_price=110.0,
+        )
+        for i in range(n)
+    ]
+    # 50/50 LONG/FLAT per evitare degenerate/flat_too_long/time_in_market.
+    fake_signals = pd.Series(
+        [Side.LONG if i % 2 == 0 else Side.FLAT for i in range(len(ohlcv))],
+        index=ohlcv.index,
+        dtype=object,
+    )
+
+    def fake_run(self, ohlcv: pd.DataFrame, signals: pd.Series) -> BacktestResult:  # type: ignore[no-untyped-def]
+        return BacktestResult(
+            equity_curve=pd.Series([0.0] * len(ohlcv), index=ohlcv.index, name="equity"),
+            returns=pd.Series([0.0] * len(ohlcv), index=ohlcv.index, name="returns"),
+            trades=fake_trades,
+        )
+
+    def fake_compile(strategy):  # type: ignore[no-untyped-def]
+        return lambda df: fake_signals
+
+    monkeypatch.setattr(
+        "multi_swarm_core.agents.adversarial.BacktestEngine.run", fake_run
+    )
+    monkeypatch.setattr(
+        "multi_swarm_core.agents.adversarial.compile_strategy", fake_compile
+    )
+
+    ast = parse_strategy(_MINIMAL_STRATEGY_SRC)
+    report = AdversarialAgent().review(ast, ohlcv)
    high_findings = [f for f in report.findings if f.severity == Severity.HIGH]
-    assert len(high_findings) == 0
+    assert high_findings == [], (
+        f"expected no HIGH findings, got: {[f.name for f in high_findings]}"
+    )


 def test_zero_trade_strategy_flagged(ohlcv: pd.DataFrame) -> None:
@@ -383,6 +436,55 @@ def test_fees_eat_alpha_flagged(monkeypatch: pytest.MonkeyPatch,
    )


+def test_negative_net_pnl_fires_on_negative_gross(
+    monkeypatch: pytest.MonkeyPatch, ohlcv: pd.DataFrame
+) -> None:
+    """gross_pnl < 0 (perdente direzionale) -> HIGH negative_net_pnl.
+    fees_eat_alpha NON deve firare perche' la sua condizione richiede gross > 0.
+    """
+    n = 15
+    # entry=100 exit=95 gross=-5 per trade (LONG perdente)
+    fake_trades = [
+        _make_trade(
+            ohlcv.index[i * 30],
+            ohlcv.index[i * 30 + 1],
+            entry_price=100.0,
+            exit_price=95.0,
+        )
+        for i in range(n)
+    ]
+    fake_signals = pd.Series(
+        [Side.LONG if i % 2 == 0 else Side.FLAT for i in range(len(ohlcv))],
+        index=ohlcv.index,
+        dtype=object,
+    )
+
+    def fake_run(self, ohlcv, signals):  # type: ignore[no-untyped-def]
+        return BacktestResult(
+            equity_curve=pd.Series([0.0] * len(ohlcv), index=ohlcv.index, name="equity"),
+            returns=pd.Series([0.0] * len(ohlcv), index=ohlcv.index, name="returns"),
+            trades=fake_trades,
+        )
+
+    def fake_compile(strategy):  # type: ignore[no-untyped-def]
+        return lambda df: fake_signals
+
+    monkeypatch.setattr(
+        "multi_swarm_core.agents.adversarial.BacktestEngine.run", fake_run
+    )
+    monkeypatch.setattr(
+        "multi_swarm_core.agents.adversarial.compile_strategy", fake_compile
+    )
+
+    ast = parse_strategy(_MINIMAL_STRATEGY_SRC)
+    report = AdversarialAgent().review(ast, ohlcv)
+    assert any(
+        f.name == "negative_net_pnl" and f.severity == Severity.HIGH
+        for f in report.findings
+    )
+    assert not any(f.name == "fees_eat_alpha" for f in report.findings)
+
+
 def test_time_in_market_too_high_flagged(monkeypatch: pytest.MonkeyPatch,
                                          ohlcv: pd.DataFrame) -> None:
    """Signal LONG per >80% delle bar -> HIGH time_in_market_too_high."""
@@ -0,0 +1,160 @@
+"""Parity check: engine vettorializzato vs reference iterrows implementation.
+
+Mantiene una copia inline del loop ``iterrows`` come reference per garantire
+che la vettorizzazione produca esattamente gli stessi trades, equity_curve e
+returns su input pseudocasuali, indipendentemente dal regime di prezzo.
+"""
+
+from __future__ import annotations
+
+import numpy as np
+import pandas as pd
+import pytest
+from multi_swarm_core.backtest.engine import BacktestEngine, BacktestResult
+from multi_swarm_core.backtest.orders import Position, Side, Trade
+
+
+def _legacy_run(
+    ohlcv: pd.DataFrame, signals: pd.Series, fees_bp: float = 5.0
+) -> BacktestResult:
+    """Reference implementation: il loop iterrows originale (pre-vectorize).
+
+    Mantenuto qui esclusivamente come oracolo per i test di parità.
+    """
+    signals = signals.reindex(ohlcv.index).ffill().fillna(Side.FLAT)
+    shifted = [Side.FLAT, *list(signals.iloc[:-1])]
+    executed_side = pd.Series(shifted, index=ohlcv.index, dtype=object)
+
+    position: Position | None = None
+    position_entry_ts: pd.Timestamp | None = None
+    trades: list[Trade] = []
+    equity = 0.0
+    equity_history: list[float] = []
+    returns_history: list[float] = []
+    prev_equity = 0.0
+
+    for ts, row in ohlcv.iterrows():
+        target_side = executed_side.loc[ts]
+        current_side = position.side if position else Side.FLAT
+        if target_side != current_side:
+            if position is not None:
+                assert position_entry_ts is not None
+                trade = Trade(
+                    entry_ts=position_entry_ts,
+                    exit_ts=ts,
+                    side=position.side,
+                    size=position.size,
+                    entry_price=position.entry_price,
+                    exit_price=float(row["open"]),
+                    fees_bp=fees_bp,
+                )
+                trades.append(trade)
+                equity += trade.net_pnl
+                position = None
+                position_entry_ts = None
+            if target_side in (Side.LONG, Side.SHORT):
+                position = Position(
+                    side=target_side, entry_price=float(row["open"]), size=1.0
+                )
+                position_entry_ts = ts
+        mark = float(row["close"])
+        mtm = position.unrealized_pnl(mark) if position else 0.0
+        current_equity = equity + mtm
+        equity_history.append(current_equity)
+        returns_history.append(current_equity - prev_equity)
+        prev_equity = current_equity
+    if position is not None:
+        assert position_entry_ts is not None
+        last_ts = ohlcv.index[-1]
+        last_close = float(ohlcv["close"].iloc[-1])
+        trade = Trade(
+            entry_ts=position_entry_ts,
+            exit_ts=last_ts,
+            side=position.side,
+            size=position.size,
+            entry_price=position.entry_price,
+            exit_price=last_close,
+            fees_bp=fees_bp,
+        )
+        trades.append(trade)
+        equity += trade.net_pnl
+        equity_history[-1] = equity
+        if len(returns_history) >= 2:
+            returns_history[-1] = equity - equity_history[-2]
+
+    return BacktestResult(
+        equity_curve=pd.Series(equity_history, index=ohlcv.index, name="equity"),
+        returns=pd.Series(returns_history, index=ohlcv.index, name="returns"),
+        trades=trades,
+    )
+
+
+def _random_ohlcv(n: int, seed: int) -> pd.DataFrame:
+    rng = np.random.default_rng(seed)
+    rets = rng.normal(0.0, 0.01, size=n)
+    close = 100.0 * np.exp(np.cumsum(rets))
+    idx = pd.date_range("2024-01-01", periods=n, freq="1h", tz="UTC")
+    return pd.DataFrame(
+        {
+            "open": close * (1 + rng.normal(0, 0.001, n)),
+            "high": close * 1.005,
+            "low": close * 0.995,
+            "close": close,
+            "volume": rng.uniform(1.0, 100.0, n),
+        },
+        index=idx,
+    )
+
+
+def _random_signals(n: int, seed: int, p_change: float = 0.1) -> pd.Series:
+    """Segnali con persistenza: ad ogni bar con prob p_change cambia stato."""
+    rng = np.random.default_rng(seed + 999)
+    states = [Side.LONG, Side.SHORT, Side.FLAT]
+    out: list[Side] = [rng.choice(states)]
+    for _ in range(1, n):
+        out.append(rng.choice(states) if rng.random() < p_change else out[-1])
+    idx = pd.date_range("2024-01-01", periods=n, freq="1h", tz="UTC")
+    return pd.Series(out, index=idx, dtype=object)
+
+
+@pytest.mark.parametrize("seed", [0, 1, 42, 123, 999])
+def test_vectorized_equals_legacy(seed: int) -> None:
+    df = _random_ohlcv(500, seed)
+    signals = _random_signals(500, seed)
+    engine = BacktestEngine(fees_bp=5.0)
+    new = engine.run(df, signals)
+    ref = _legacy_run(df, signals, fees_bp=5.0)
+
+    pd.testing.assert_series_equal(
+        new.equity_curve, ref.equity_curve, rtol=1e-9, atol=1e-9
+    )
+    pd.testing.assert_series_equal(
+        new.returns, ref.returns, rtol=1e-9, atol=1e-9
+    )
+    assert len(new.trades) == len(ref.trades)
+    for nt, rt in zip(new.trades, ref.trades, strict=True):
+        assert nt.entry_ts == rt.entry_ts
+        assert nt.exit_ts == rt.exit_ts
+        assert nt.side == rt.side
+        assert nt.entry_price == pytest.approx(rt.entry_price, abs=1e-12)
+        assert nt.exit_price == pytest.approx(rt.exit_price, abs=1e-12)
+        assert nt.net_pnl == pytest.approx(rt.net_pnl, abs=1e-12)
+
+
+def test_vectorized_handles_position_still_open_at_end() -> None:
+    """Edge case: signal LONG fino all'ultimo bar (exit a close[-1] forced)."""
+    df = _random_ohlcv(100, seed=7)
+    signals = pd.Series([Side.LONG] * 100, index=df.index)
+    new = BacktestEngine(fees_bp=10.0).run(df, signals)
+    ref = _legacy_run(df, signals, fees_bp=10.0)
+    pd.testing.assert_series_equal(new.equity_curve, ref.equity_curve, atol=1e-9)
+    assert len(new.trades) == 1
+    assert new.trades[0].side == Side.LONG
+
+
+def test_vectorized_zero_signals_no_trades() -> None:
+    df = _random_ohlcv(50, seed=3)
+    signals = pd.Series([Side.FLAT] * 50, index=df.index)
+    new = BacktestEngine().run(df, signals)
+    assert len(new.trades) == 0
+    assert (new.equity_curve == 0).all()
@@ -18,13 +18,14 @@ def test_settings_loads_from_env(monkeypatch: pytest.MonkeyPatch) -> None:
    monkeypatch.setenv("CERBERO_TESTNET_TOKEN", "tok-test")
    monkeypatch.setenv("CERBERO_MAINNET_TOKEN", "tok-main")
    monkeypatch.setenv("CERBERO_BOT_TAG", "swarm-poc-phase1")
-    monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
+    monkeypatch.setenv("OPUS_AGENT_API_KEY", "oa-key")
    monkeypatch.setenv("RUN_NAME", "test-run")

    s = Settings()  # type: ignore[call-arg]

    assert s.cerbero_base_url == "http://test:9000"
    assert s.cerbero_testnet_token.get_secret_value() == "tok-test"
+    assert s.opus_agent_api_key.get_secret_value() == "oa-key"
    assert s.run_name == "test-run"
    assert s.data_dir.name == "data"
    assert s.db_path.name == "runs.db"
@@ -32,50 +33,33 @@ def test_settings_loads_from_env(monkeypatch: pytest.MonkeyPatch) -> None:

 def test_settings_requires_tokens(monkeypatch: pytest.MonkeyPatch) -> None:
    monkeypatch.delenv("CERBERO_TESTNET_TOKEN", raising=False)
-    monkeypatch.delenv("OPENROUTER_API_KEY", raising=False)
+    monkeypatch.delenv("OPUS_AGENT_API_KEY", raising=False)
    from pydantic import ValidationError

    with pytest.raises(ValidationError):
-        # Disable .env loading to keep the test deterministic regardless of
-        # whether a developer's local .env exists and is populated.
        Settings(_env_file=None)  # type: ignore[call-arg]


-def test_settings_loads_llm_model_overrides(monkeypatch: pytest.MonkeyPatch) -> None:
+def test_settings_opus_agent_defaults(monkeypatch: pytest.MonkeyPatch) -> None:
    monkeypatch.setenv("CERBERO_TESTNET_TOKEN", "tok-test")
-    monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
-    monkeypatch.setenv("LLM_MODEL_TIER_S", "claude-mega-x")
-    monkeypatch.setenv("LLM_MODEL_TIER_A", "claude-premium-y")
-    monkeypatch.setenv("LLM_MODEL_TIER_B", "claude-opus-4-7")
-    monkeypatch.setenv("LLM_MODEL_TIER_C", "deepseek/deepseek-chat")
-    monkeypatch.setenv("LLM_MODEL_TIER_D", "mistralai/mistral-7b")
-    monkeypatch.setenv("OPENROUTER_BASE_URL", "https://example.com/api/v1")
+    monkeypatch.setenv("OPUS_AGENT_API_KEY", "oa-key")
+    monkeypatch.delenv("OPUS_AGENT_BASE_URL", raising=False)

    s = Settings(_env_file=None)  # type: ignore[call-arg]

-    assert s.llm_model_tier_s == "claude-mega-x"
-    assert s.llm_model_tier_a == "claude-premium-y"
-    assert s.llm_model_tier_b == "claude-opus-4-7"
-    assert s.llm_model_tier_c == "deepseek/deepseek-chat"
-    assert s.llm_model_tier_d == "mistralai/mistral-7b"
-    assert s.openrouter_base_url == "https://example.com/api/v1"
+    assert s.opus_agent_base_url == "https://opus-agent.tielogic.xyz"
+    assert s.llm_model_tier_s == "claude-opus-4-7"
+    assert s.llm_model_tier_a == "claude-opus-4-7"
+    assert s.llm_model_tier_b == "claude-sonnet-4-6"
+    assert s.llm_model_tier_c == "claude-sonnet-4-6"
+    assert s.llm_model_tier_d == "claude-haiku-4-5-20251001"


-def test_settings_llm_model_defaults(monkeypatch: pytest.MonkeyPatch) -> None:
+def test_settings_opus_agent_base_url_override(monkeypatch: pytest.MonkeyPatch) -> None:
    monkeypatch.setenv("CERBERO_TESTNET_TOKEN", "tok-test")
-    monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
-    monkeypatch.delenv("LLM_MODEL_TIER_S", raising=False)
-    monkeypatch.delenv("LLM_MODEL_TIER_A", raising=False)
-    monkeypatch.delenv("LLM_MODEL_TIER_B", raising=False)
-    monkeypatch.delenv("LLM_MODEL_TIER_C", raising=False)
-    monkeypatch.delenv("LLM_MODEL_TIER_D", raising=False)
-    monkeypatch.delenv("OPENROUTER_BASE_URL", raising=False)
+    monkeypatch.setenv("OPUS_AGENT_API_KEY", "oa-key")
+    monkeypatch.setenv("OPUS_AGENT_BASE_URL", "https://custom.example.com")

    s = Settings(_env_file=None)  # type: ignore[call-arg]

-    assert s.llm_model_tier_s == "google/gemini-3-flash-preview"
-    assert s.llm_model_tier_a == "deepseek/deepseek-v4-flash"
-    assert s.llm_model_tier_b == "deepseek/deepseek-v4-flash"
-    assert s.llm_model_tier_c == "qwen/qwen-2.5-72b-instruct"
-    assert s.llm_model_tier_d == "openai/gpt-oss-20b"
-    assert s.openrouter_base_url == "https://openrouter.ai/api/v1"
+    assert s.opus_agent_base_url == "https://custom.example.com"
@@ -420,3 +420,48 @@ def test_build_system_prompt_skips_anti_patterns_and_priorities_when_empty(mocke
    system_msg = call_kwargs["system"]
    assert "ANTI-PATTERN" not in system_msg
    assert "PRIORITA' DI OUTPUT" not in system_msg
+
+
+def test_build_system_prompt_includes_custom_indicators_spec(mocker):  # type: ignore[no-untyped-def]
+    """custom_indicators_spec da PromptLibrary appare nel SYSTEM prompt con header."""
+    fake_llm = mocker.MagicMock()
+    fake_llm.complete.return_value = CompletionResult(
+        text=VALID_STRATEGY_JSON,
+        input_tokens=200,
+        output_tokens=80,
+        tier=ModelTier.C,
+        model="qwen",
+    )
+    lib = PromptLibrary(
+        styles={"physicist": "Pensa come un fisico."},
+        focus={},
+        custom_indicators_spec="CUSTOM_IND_X",
+    )
+    agent = HypothesisAgent(llm=fake_llm, prompt_library=lib)
+    agent.propose(make_genome(), make_summary())
+    call_kwargs = fake_llm.complete.call_args.kwargs
+    system_msg = call_kwargs["system"]
+    assert "INDICATORI CUSTOM" in system_msg
+    assert "CUSTOM_IND_X" in system_msg
+
+
+def test_build_system_prompt_skips_custom_indicators_spec_when_empty(mocker):  # type: ignore[no-untyped-def]
+    """custom_indicators_spec='' -> sezione assente nel SYSTEM prompt."""
+    fake_llm = mocker.MagicMock()
+    fake_llm.complete.return_value = CompletionResult(
+        text=VALID_STRATEGY_JSON,
+        input_tokens=200,
+        output_tokens=80,
+        tier=ModelTier.C,
+        model="qwen",
+    )
+    lib = PromptLibrary(
+        styles={"physicist": "Pensa come un fisico."},
+        focus={},
+        custom_indicators_spec="",
+    )
+    agent = HypothesisAgent(llm=fake_llm, prompt_library=lib)
+    agent.propose(make_genome(), make_summary())
+    call_kwargs = fake_llm.complete.call_args.kwargs
+    system_msg = call_kwargs["system"]
+    assert "INDICATORI CUSTOM" not in system_msg
@@ -1,7 +1,14 @@
+import httpx
 import pytest

 from multi_swarm_core.genome.hypothesis import HypothesisAgentGenome, ModelTier
-from multi_swarm_core.llm.client import CompletionResult, LLMClient
+from multi_swarm_core.llm.client import (
+    CompletionResult,
+    EmptyCompletionError,
+    LLMClient,
+    OpusAgentError,
+    OpusAgentTransientError,
+)


 def make_genome(tier: ModelTier) -> HypothesisAgentGenome:
@@ -16,217 +23,248 @@ def make_genome(tier: ModelTier) -> HypothesisAgentGenome:
    )


-def test_completion_tier_c_uses_openrouter(mocker):
-    fake_openai = mocker.MagicMock()
-    fake_response = mocker.MagicMock()
-    fake_response.choices = [mocker.MagicMock(message=mocker.MagicMock(content="(strategy ...)"))]
-    fake_response.usage = mocker.MagicMock(prompt_tokens=100, completion_tokens=200)
-    fake_openai.chat.completions.create.return_value = fake_response
+TOPIC_RESPONSE = {
+    "success": True,
+    "data": {"id": "topic-123", "name": "swarm-test", "system_prompt": "sys"},
+}

-    mocker.patch("multi_swarm_core.llm.client.OpenAI", return_value=fake_openai)
+REQUEST_ACCEPTED = {
+    "success": True,
+    "data": {"id": "req-456", "session_id": None, "status": "pending"},
+}

-    client = LLMClient(openrouter_api_key="or-x")
-    g = make_genome(ModelTier.C)
-    out = client.complete(g, system="sys", user="usr")
+
+def _completed_response(
+    text: str = "(strategy ...)", session_id: str = "sess-789",
+) -> dict:
+    return {
+        "success": True,
+        "data": {
+            "id": "req-456",
+            "status": "completed",
+            "result": text,
+            "session_id": session_id,
+        },
+    }
+
+
+def _mock_transport(responses: list[httpx.Response]) -> httpx.MockTransport:
+    call_idx = {"i": 0}
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        idx = call_idx["i"]
+        call_idx["i"] += 1
+        if idx < len(responses):
+            return responses[idx]
+        return responses[-1]
+
+    return httpx.MockTransport(handler)
+
+
+def _make_client(transport: httpx.MockTransport) -> LLMClient:
+    client = LLMClient(opus_agent_api_key="test-key", poll_interval=0.01, poll_timeout=5.0)
+    client._client = httpx.Client(
+        base_url="https://opus-agent.tielogic.xyz",
+        headers={"X-Api-Key": "test-key", "Content-Type": "application/json"},
+        transport=transport,
+    )
+    return client
+
+
+def test_complete_tier_c_uses_sonnet():
+    transport = _mock_transport([
+        httpx.Response(201, json=TOPIC_RESPONSE),
+        httpx.Response(202, json=REQUEST_ACCEPTED),
+        httpx.Response(200, json=_completed_response()),
+    ])
+    client = _make_client(transport)
+    out = client.complete(make_genome(ModelTier.C), system="sys", user="usr")

    assert isinstance(out, CompletionResult)
    assert out.text == "(strategy ...)"
-    assert out.input_tokens == 100
-    assert out.output_tokens == 200
+    assert out.model == "claude-sonnet-4-6"
    assert out.tier == ModelTier.C
-    fake_openai.chat.completions.create.assert_called_once()


-def test_completion_tier_b_uses_openrouter_with_anthropic_model(mocker):
-    fake_openai = mocker.MagicMock()
-    fake_response = mocker.MagicMock()
-    fake_response.choices = [mocker.MagicMock(message=mocker.MagicMock(content="(strategy ...)"))]
-    fake_response.usage = mocker.MagicMock(prompt_tokens=80, completion_tokens=150)
-    fake_openai.chat.completions.create.return_value = fake_response
-    mocker.patch("multi_swarm_core.llm.client.OpenAI", return_value=fake_openai)
+def test_complete_tier_s_uses_opus():
+    transport = _mock_transport([
+        httpx.Response(201, json=TOPIC_RESPONSE),
+        httpx.Response(202, json=REQUEST_ACCEPTED),
+        httpx.Response(200, json=_completed_response("(strategy s)")),
+    ])
+    client = _make_client(transport)
+    out = client.complete(make_genome(ModelTier.S), system="sys", user="usr")

-    client = LLMClient(openrouter_api_key="or-x")
-    g = make_genome(ModelTier.B)
-    out = client.complete(g, system="sys", user="usr")
-
-    assert out.text == "(strategy ...)"
-    assert out.input_tokens == 80
-    assert out.output_tokens == 150
-    assert out.tier == ModelTier.B
-    call_kwargs = fake_openai.chat.completions.create.call_args.kwargs
-    assert call_kwargs["model"] == "deepseek/deepseek-v4-flash"
-    assert out.model == "deepseek/deepseek-v4-flash"
-
-
-@pytest.mark.slow
-def test_completion_retries_on_connection_error(mocker):
-    """Retry esegue 3 tentativi su APIConnectionError, poi rilancia."""
-    import openai
-
-    fake_openai = mocker.MagicMock()
-    fake_openai.chat.completions.create.side_effect = openai.APIConnectionError(
-        request=mocker.MagicMock()
-    )
-    mocker.patch("multi_swarm_core.llm.client.OpenAI", return_value=fake_openai)
-
-    client = LLMClient(openrouter_api_key="or-x")
-    g = make_genome(ModelTier.C)
-
-    with pytest.raises(openai.APIConnectionError):
-        client.complete(g, system="sys", user="usr")
-
-    assert fake_openai.chat.completions.create.call_count == 5
-
-
-def test_completion_uses_custom_model_tier_c(mocker):
-    fake_openai = mocker.MagicMock()
-    fake_response = mocker.MagicMock()
-    fake_response.choices = [
-        mocker.MagicMock(message=mocker.MagicMock(content="(strategy ...)"))
-    ]
-    fake_response.usage = mocker.MagicMock(prompt_tokens=10, completion_tokens=20)
-    fake_openai.chat.completions.create.return_value = fake_response
-    mocker.patch("multi_swarm_core.llm.client.OpenAI", return_value=fake_openai)
-
-    client = LLMClient(
-        openrouter_api_key="or-x",
-        model_tier_c="deepseek/deepseek-chat",
-    )
-    g = make_genome(ModelTier.C)
-    out = client.complete(g, system="sys", user="usr")
-
-    fake_openai.chat.completions.create.assert_called_once()
-    call_kwargs = fake_openai.chat.completions.create.call_args.kwargs
-    assert call_kwargs["model"] == "deepseek/deepseek-chat"
-    assert out.model == "deepseek/deepseek-chat"
-
-
-def test_completion_uses_custom_model_tier_b(mocker):
-    fake_openai = mocker.MagicMock()
-    fake_response = mocker.MagicMock()
-    fake_response.choices = [
-        mocker.MagicMock(message=mocker.MagicMock(content="(strategy ...)"))
-    ]
-    fake_response.usage = mocker.MagicMock(prompt_tokens=10, completion_tokens=20)
-    fake_openai.chat.completions.create.return_value = fake_response
-    mocker.patch("multi_swarm_core.llm.client.OpenAI", return_value=fake_openai)
-
-    client = LLMClient(
-        openrouter_api_key="or-x",
-        model_tier_b="anthropic/claude-opus-4-7",
-    )
-    g = make_genome(ModelTier.B)
-    out = client.complete(g, system="sys", user="usr")
-
-    fake_openai.chat.completions.create.assert_called_once()
-    call_kwargs = fake_openai.chat.completions.create.call_args.kwargs
-    assert call_kwargs["model"] == "anthropic/claude-opus-4-7"
-    assert out.model == "anthropic/claude-opus-4-7"
-
-
-def test_completion_tier_s_uses_openrouter_with_anthropic_model(mocker):
-    fake_openai = mocker.MagicMock()
-    fake_response = mocker.MagicMock()
-    fake_response.choices = [mocker.MagicMock(message=mocker.MagicMock(content="(strategy s)"))]
-    fake_response.usage = mocker.MagicMock(prompt_tokens=50, completion_tokens=100)
-    fake_openai.chat.completions.create.return_value = fake_response
-    mocker.patch("multi_swarm_core.llm.client.OpenAI", return_value=fake_openai)
-
-    client = LLMClient(openrouter_api_key="or-x")
-    g = make_genome(ModelTier.S)
-    out = client.complete(g, system="sys", user="usr")
-
-    fake_openai.chat.completions.create.assert_called_once()
-    call_kwargs = fake_openai.chat.completions.create.call_args.kwargs
-    assert call_kwargs["model"] == "google/gemini-3-flash-preview"
+    assert out.model == "claude-opus-4-7"
    assert out.tier == ModelTier.S
-    assert out.model == "google/gemini-3-flash-preview"
+    assert out.text == "(strategy s)"


-def test_completion_tier_a_uses_openrouter_with_anthropic_model(mocker):
-    fake_openai = mocker.MagicMock()
-    fake_response = mocker.MagicMock()
-    fake_response.choices = [mocker.MagicMock(message=mocker.MagicMock(content="(strategy a)"))]
-    fake_response.usage = mocker.MagicMock(prompt_tokens=40, completion_tokens=80)
-    fake_openai.chat.completions.create.return_value = fake_response
-    mocker.patch("multi_swarm_core.llm.client.OpenAI", return_value=fake_openai)
+def test_complete_tier_d_uses_haiku():
+    transport = _mock_transport([
+        httpx.Response(201, json=TOPIC_RESPONSE),
+        httpx.Response(202, json=REQUEST_ACCEPTED),
+        httpx.Response(200, json=_completed_response("(strategy d)")),
+    ])
+    client = _make_client(transport)
+    out = client.complete(make_genome(ModelTier.D), system="sys", user="usr")

-    client = LLMClient(openrouter_api_key="or-x")
-    g = make_genome(ModelTier.A)
-    out = client.complete(g, system="sys", user="usr")
-
-    fake_openai.chat.completions.create.assert_called_once()
-    call_kwargs = fake_openai.chat.completions.create.call_args.kwargs
-    assert call_kwargs["model"] == "deepseek/deepseek-v4-flash"
-    assert out.tier == ModelTier.A
-    assert out.model == "deepseek/deepseek-v4-flash"
-
-
-def test_completion_tier_d_uses_openrouter_with_llama(mocker):
-    fake_openai = mocker.MagicMock()
-    fake_response = mocker.MagicMock()
-    fake_response.choices = [
-        mocker.MagicMock(message=mocker.MagicMock(content="(strategy d)"))
-    ]
-    fake_response.usage = mocker.MagicMock(prompt_tokens=30, completion_tokens=70)
-    fake_openai.chat.completions.create.return_value = fake_response
-    mocker.patch("multi_swarm_core.llm.client.OpenAI", return_value=fake_openai)
-
-    client = LLMClient(openrouter_api_key="or-x")
-    g = make_genome(ModelTier.D)
-    out = client.complete(g, system="sys", user="usr")
-
-    fake_openai.chat.completions.create.assert_called_once()
-    call_kwargs = fake_openai.chat.completions.create.call_args.kwargs
-    assert call_kwargs["model"] == "openai/gpt-oss-20b"
+    assert out.model == "claude-haiku-4-5-20251001"
    assert out.tier == ModelTier.D
-    assert out.model == "openai/gpt-oss-20b"


-def test_completion_uses_custom_model_tier_s(mocker):
-    fake_openai = mocker.MagicMock()
-    fake_response = mocker.MagicMock()
-    fake_response.choices = [
-        mocker.MagicMock(message=mocker.MagicMock(content="(strategy custom-s)"))
-    ]
-    fake_response.usage = mocker.MagicMock(prompt_tokens=10, completion_tokens=20)
-    fake_openai.chat.completions.create.return_value = fake_response
-    mocker.patch("multi_swarm_core.llm.client.OpenAI", return_value=fake_openai)
+def test_topic_cached_on_second_call():
+    transport = _mock_transport([
+        httpx.Response(201, json=TOPIC_RESPONSE),
+        httpx.Response(202, json=REQUEST_ACCEPTED),
+        httpx.Response(200, json=_completed_response()),
+        httpx.Response(202, json=REQUEST_ACCEPTED),
+        httpx.Response(200, json=_completed_response("second")),
+    ])
+    client = _make_client(transport)
+    client.complete(make_genome(ModelTier.C), system="sys", user="usr1")
+    out2 = client.complete(make_genome(ModelTier.C), system="sys", user="usr2")

-    client = LLMClient(
-        openrouter_api_key="or-x",
-        model_tier_s="anthropic/claude-future-mega",
+    assert out2.text == "second"
+
+
+def test_topic_conflict_409_recovers():
+    transport = _mock_transport([
+        httpx.Response(409, json={"success": False, "error": {"code": "CONFLICT"}}),
+        httpx.Response(200, json={"success": True, "data": [
+            {"id": "topic-existing", "name": "swarm-wrong"},
+        ]}),
+    ])
+    client = _make_client(transport)
+
+    with pytest.raises(OpusAgentError, match="conflict but not found"):
+        client.complete(make_genome(ModelTier.C), system="sys", user="usr")
+
+
+def test_empty_response_raises():
+    transport = _mock_transport([
+        httpx.Response(201, json=TOPIC_RESPONSE),
+        httpx.Response(202, json=REQUEST_ACCEPTED),
+        httpx.Response(200, json={
+            "success": True,
+            "data": {"id": "req-456", "status": "completed", "result": ""},
+        }),
+    ])
+    client = _make_client(transport)
+
+    with pytest.raises(EmptyCompletionError):
+        client.complete(make_genome(ModelTier.C), system="sys", user="usr")
+
+
+def test_request_failed_raises_opus_agent_error():
+    transport = _mock_transport([
+        httpx.Response(201, json=TOPIC_RESPONSE),
+        httpx.Response(202, json=REQUEST_ACCEPTED),
+        httpx.Response(200, json={
+            "success": True,
+            "data": {"id": "req-456", "status": "failed", "error": "model overloaded"},
+        }),
+    ])
+    client = _make_client(transport)
+
+    with pytest.raises(OpusAgentError, match="failed"):
+        client.complete(make_genome(ModelTier.C), system="sys", user="usr")
+
+
+def test_rate_limit_429_is_retryable():
+    transport = _mock_transport([
+        httpx.Response(201, json=TOPIC_RESPONSE),
+        httpx.Response(429, json={"success": False}),
+        httpx.Response(202, json=REQUEST_ACCEPTED),
+        httpx.Response(200, json=_completed_response("after retry")),
+    ])
+    client = _make_client(transport)
+    out = client.complete(make_genome(ModelTier.C), system="sys", user="usr")
+
+    assert out.text == "after retry"
+
+
+def test_polling_waits_for_completion():
+    transport = _mock_transport([
+        httpx.Response(201, json=TOPIC_RESPONSE),
+        httpx.Response(202, json=REQUEST_ACCEPTED),
+        httpx.Response(200, json={
+            "success": True,
+            "data": {"id": "req-456", "status": "processing"},
+        }),
+        httpx.Response(200, json={
+            "success": True,
+            "data": {"id": "req-456", "status": "processing"},
+        }),
+        httpx.Response(200, json=_completed_response("done")),
+    ])
+    client = _make_client(transport)
+    out = client.complete(make_genome(ModelTier.C), system="sys", user="usr")
+
+    assert out.text == "done"
+
+
+def test_tokens_are_zero():
+    transport = _mock_transport([
+        httpx.Response(201, json=TOPIC_RESPONSE),
+        httpx.Response(202, json=REQUEST_ACCEPTED),
+        httpx.Response(200, json=_completed_response()),
+    ])
+    client = _make_client(transport)
+    out = client.complete(make_genome(ModelTier.C), system="sys", user="usr")
+
+    assert out.input_tokens == 0
+    assert out.output_tokens == 0
+
+
+def test_session_id_returned_from_completion():
+    transport = _mock_transport([
+        httpx.Response(201, json=TOPIC_RESPONSE),
+        httpx.Response(202, json=REQUEST_ACCEPTED),
+        httpx.Response(200, json=_completed_response(session_id="sess-abc")),
+    ])
+    client = _make_client(transport)
+    out = client.complete(make_genome(ModelTier.C), system="sys", user="usr", session_id="new")
+
+    assert out.session_id == "sess-abc"
+
+
+def test_session_id_and_summarize_sent_in_request():
+    requests_seen: list[dict] = []
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        if request.method == "POST" and "/requests" in str(request.url):
+            import json
+            requests_seen.append(json.loads(request.content))
+            return httpx.Response(202, json=REQUEST_ACCEPTED)
+        if request.method == "POST" and "/topics" in str(request.url):
+            return httpx.Response(201, json=TOPIC_RESPONSE)
+        return httpx.Response(200, json=_completed_response())
+
+    transport = httpx.MockTransport(handler)
+    client = _make_client(transport)
+    client.complete(
+        make_genome(ModelTier.C), system="sys", user="usr",
+        session_id="sess-existing", summarize=True,
    )
-    g = make_genome(ModelTier.S)
-    out = client.complete(g, system="sys", user="usr")

-    call_kwargs = fake_openai.chat.completions.create.call_args.kwargs
-    assert call_kwargs["model"] == "anthropic/claude-future-mega"
-    assert out.model == "anthropic/claude-future-mega"
+    assert len(requests_seen) == 1
+    assert requests_seen[0]["session_id"] == "sess-existing"
+    assert requests_seen[0]["summarize"] is True


-@pytest.mark.slow
-def test_completion_succeeds_after_one_retry(mocker):
-    """Dopo 1 fallimento transient, il retry riesce al 2 tentativo."""
-    import openai
+def test_close_session():
+    deleted: list[str] = []

-    fake_response = mocker.MagicMock()
-    fake_response.choices = [
-        mocker.MagicMock(message=mocker.MagicMock(content="(strategy ...)"))
-    ]
-    fake_response.usage = mocker.MagicMock(prompt_tokens=100, completion_tokens=200)
+    def handler(request: httpx.Request) -> httpx.Response:
+        if request.method == "DELETE":
+            deleted.append(str(request.url))
+            return httpx.Response(200, json={"success": True})
+        return httpx.Response(200, json={"success": True})

-    fake_openai = mocker.MagicMock()
-    fake_openai.chat.completions.create.side_effect = [
-        openai.APITimeoutError(request=mocker.MagicMock()),
-        fake_response,
-    ]
-    mocker.patch("multi_swarm_core.llm.client.OpenAI", return_value=fake_openai)
+    transport = httpx.MockTransport(handler)
+    client = _make_client(transport)
+    client.close_session("sess-to-close")

-    client = LLMClient(openrouter_api_key="or-x")
-    g = make_genome(ModelTier.C)
-    out = client.complete(g, system="sys", user="usr")
-
-    assert isinstance(out, CompletionResult)
-    assert out.text == "(strategy ...)"
-    assert fake_openai.chat.completions.create.call_count == 2
+    assert len(deleted) == 1
+    assert "sess-to-close" in deleted[0]
@@ -0,0 +1,78 @@
+"""Test che `_parallel_propose` preservi l'ordine dei risultati e funzioni
+sia in modalita' sequenziale (workers=1) che concorrente (workers>1).
+
+Non vogliamo testare il vero `HypothesisAgent.propose()` (che fa chiamate
+LLM); usiamo un dummy con una latenza simulata per validare ordine e parallelismo.
+"""
+
+from __future__ import annotations
+
+import time
+from dataclasses import dataclass
+from typing import Any
+
+from multi_swarm_core.orchestrator.run import _parallel_propose
+
+
+@dataclass
+class _FakeGenome:
+    id: str
+
+
+@dataclass
+class _FakeProposal:
+    genome_id: str
+
+
+class _FakeAgent:
+    """Agent dummy: propose() dorme 50ms e ritorna un proposal con l'id del genome."""
+
+    def __init__(self, delay_s: float = 0.05) -> None:
+        self._delay = delay_s
+        self.call_count = 0
+
+    def propose(self, genome: _FakeGenome, market: Any) -> _FakeProposal:
+        time.sleep(self._delay)
+        self.call_count += 1
+        return _FakeProposal(genome_id=genome.id)
+
+
+def test_parallel_propose_preserves_order_serial() -> None:
+    agent = _FakeAgent(delay_s=0.01)
+    genomes = [_FakeGenome(id=f"g{i}") for i in range(5)]
+    results = _parallel_propose(agent, genomes, market=None, n_workers=1)
+    assert [r.genome_id for r in results] == ["g0", "g1", "g2", "g3", "g4"]
+    assert agent.call_count == 5
+
+
+def test_parallel_propose_preserves_order_concurrent() -> None:
+    agent = _FakeAgent(delay_s=0.05)
+    genomes = [_FakeGenome(id=f"g{i}") for i in range(8)]
+    results = _parallel_propose(agent, genomes, market=None, n_workers=4)
+    assert [r.genome_id for r in results] == [f"g{i}" for i in range(8)]
+    assert agent.call_count == 8
+
+
+def test_parallel_propose_actually_parallelizes() -> None:
+    """Wall time con 4 worker su 4 task da 100ms deve essere ~100ms, non ~400ms."""
+    agent = _FakeAgent(delay_s=0.1)
+    genomes = [_FakeGenome(id=f"g{i}") for i in range(4)]
+    t0 = time.time()
+    _parallel_propose(agent, genomes, market=None, n_workers=4)
+    elapsed = time.time() - t0
+    # serial sarebbe 0.4s; con 4 worker scendiamo a ~0.1s (max 0.2 per overhead).
+    assert elapsed < 0.2, f"expected <200ms with 4 workers, got {elapsed * 1000:.0f}ms"
+
+
+def test_parallel_propose_handles_single_genome() -> None:
+    agent = _FakeAgent()
+    results = _parallel_propose(agent, [_FakeGenome(id="solo")], market=None, n_workers=8)
+    assert len(results) == 1
+    assert results[0].genome_id == "solo"
+
+
+def test_parallel_propose_empty_input() -> None:
+    agent = _FakeAgent()
+    results = _parallel_propose(agent, [], market=None, n_workers=4)
+    assert results == []
+    assert agent.call_count == 0
@@ -88,3 +88,81 @@ def test_from_json_loads_anti_patterns_and_output_priorities(tmp_path: Path) ->
    lib = PromptLibrary.from_json(_write_json(data, tmp_path))
    assert lib.anti_patterns == "Evita overfitting."
    assert lib.output_priorities == "Robustezza > ottimalita."
+
+
+def test_from_json_loads_custom_indicators_spec(tmp_path: Path) -> None:
+    """from_json() legge custom_indicators_spec (v3.2): firme di indicatori strategy-specific."""
+    data = {
+        "styles": {"physicist": {"directive": "Cerca leggi conservative."}},
+        "custom_indicators_spec": (
+            "candle_pattern: params=[length, sym0, sym1, ...], length in [3,12], "
+            "sym in {0=U,1=D,2=doji}\n"
+            "pythagorean_ratio: params=[lookback], lookback in [12,200]"
+        ),
+    }
+    lib = PromptLibrary.from_json(_write_json(data, tmp_path))
+    assert "candle_pattern" in lib.custom_indicators_spec
+    assert "lookback in [12,200]" in lib.custom_indicators_spec
+
+
+def test_from_json_defaults_custom_indicators_spec_when_absent(tmp_path: Path) -> None:
+    """custom_indicators_spec assente -> default stringa vuota."""
+    data = {"styles": {"physicist": {"directive": "Cerca leggi conservative."}}}
+    lib = PromptLibrary.from_json(_write_json(data, tmp_path))
+    assert lib.custom_indicators_spec == ""
+
+
+def test_strategy_crypto_directives_ascii_safe() -> None:
+    """REGRESSION GUARD: nessuna directive contiene caratteri > U+007F.
+
+    v3.1 aveva regredito introducendo il carattere circa-uguale (U+2248) in 3 stili.
+    v3.2 ripristina ASCII-strict come invariante permanente.
+    """
+    import importlib.resources
+
+    path = importlib.resources.files("strategy_crypto") / "prompts.json"
+    lib = PromptLibrary.from_json(str(path))
+
+    for style, directive in lib.styles.items():
+        non_ascii = [c for c in directive if ord(c) > 127]
+        assert not non_ascii, (
+            f"directive di {style!r} contiene caratteri non-ASCII: "
+            f"{non_ascii} (codepoints: {[hex(ord(c)) for c in non_ascii]})"
+        )
+
+
+def test_strategy_crypto_directives_have_archetype_marker() -> None:
+    """REGRESSION GUARD: ogni directive chiude con 'Archetipo dominante: ...'.
+
+    L'archetipo e' l'ancora semantica identitaria della lente; deve essere
+    presente per resistere alle riscritture di mutate_prompt_llm.
+    """
+    import importlib.resources
+
+    path = importlib.resources.files("strategy_crypto") / "prompts.json"
+    lib = PromptLibrary.from_json(str(path))
+
+    for style, directive in lib.styles.items():
+        assert "Archetipo dominante:" in directive, (
+            f"directive di {style!r} manca del marker 'Archetipo dominante:'"
+        )
+
+
+def test_strategy_crypto_directives_have_lookback_hint() -> None:
+    """REGRESSION GUARD: ogni directive contiene un hint 'Lookback consigliato: X-Y barre'.
+
+    Il range numerico orienta il parametro evoluto lookback_window del genoma;
+    differenziato per stile per favorire diversita di scala temporale nella
+    popolazione iniziale.
+    """
+    import re
+    import importlib.resources
+
+    path = importlib.resources.files("strategy_crypto") / "prompts.json"
+    lib = PromptLibrary.from_json(str(path))
+
+    pattern = re.compile(r"[Ll]ookback consigliato:\s*\d+\s*-\s*\d+", re.IGNORECASE)
+    for style, directive in lib.styles.items():
+        assert pattern.search(directive), (
+            f"directive di {style!r} manca dell'hint 'Lookback consigliato: X-Y'"
+        )
@@ -7,6 +7,7 @@ from __future__ import annotations

 import json
 import sqlite3
+import time
 from pathlib import Path
 from typing import Any

@@ -14,9 +15,18 @@ import pandas as pd  # type: ignore[import-untyped]


 def _paper_conn(db_path: str | Path) -> sqlite3.Connection:
-    conn = sqlite3.connect(str(db_path))
+    # Cold-start race: GUI può avviarsi prima che il paper writer crei il file.
+    db_path_str = str(db_path)
+    deadline = time.monotonic() + 5.0
+    while True:
+        try:
+            conn = sqlite3.connect(db_path_str, timeout=5.0)
            conn.row_factory = sqlite3.Row
            return conn
+        except sqlite3.OperationalError:
+            if time.monotonic() >= deadline:
+                raise
+            time.sleep(1.0)


 def paper_runs_df(db_path: str | Path) -> pd.DataFrame:
@@ -1,54 +1,49 @@
 {
  "_comment": "Stili cognitivi e direttive del system_prompt per il GA di strategy_crypto. Modifica liberamente: cambia 'directive' di uno stile esistente o aggiungi nuove voci a 'styles'. Il nome dello stile (key) viene usato come 'cognitive_style' del genoma; la 'directive' diventa il system_prompt iniziale.",
-  "_schema": "3.1",
-  "_changelog": "v3.1 - Refactor contenuto post-diagnosi: pattern_guidance astratto (nessun riferimento a indicatori specifici, lascia il GA scoprire il mapping); domain_warnings riformulato in 'NON assumere' (rimossa inferenza su funding rate); agent_role con swarm awareness; NEW anti_patterns + output_priorities; directive ridotte sotto 900 char; focus_metrics standardizzati a 4 per stile, rimosse ridondanze (autocorr_baseline da historian, kurt/skew da psychologist sostituiti con autocorr_recent + spectral_entropy). v3.0 - Refactor compositore: prompts.json controlla agent_role/pattern_guidance/instruction/domain_warnings top-level; core fornisce solo lo SCAFFOLD universale. v2.2 - Aggiunte 5 metriche geometrico-frattali con focus_metrics per stile. v2.1 - directive estese con interpretazione dei 4 input statistici. v2.0 - Riprogettato per blind-generator GA, 7 lenti.",
-  "_focus_metrics_design": "Le focus_metrics sono ENFASI per la lente, non filtri. Standardizzate a 4 per stile (cognitive budget). Evitano ridondanze con la sezione 'Regime recente' del USER_TEMPLATE (es. autocorr_baseline non va nel focus, e' gia' visibile). Scelte per essere semantically aligned con la metafora dello stile.",
+  "_schema": "3.2",
+  "_changelog": "v3.2 - Patch consolidamento: ripristinati 3 invarianti regrediti in v3.1 (ASCII-safe, archetipo dominante, hint lookback); voce attiva rinforzata; anti_patterns +2 (chattering, isteresi); output_priorities +1 (#1 coerenza con lente cognitiva); domain_warnings +1 frase (soglia seasonality 0.05); NEW _design_invariants metadata. Lunghezza directive 800-950 char (era 545-614 in v3.1, troppo snellite). v3.1 - Refactor contenuto post-diagnosi. v3.0 - Refactor compositore. v2.2 - Metriche geometrico-frattali. v2.1 - directive estese. v2.0 - Riprogettato per blind-generator GA.",
+  "_focus_metrics_design": "Le focus_metrics sono ENFASI per la lente, non filtri. Standardizzate a 4 per stile (cognitive budget). Evitano ridondanze con la sezione 'Regime recente' del USER_TEMPLATE.",
+  "_design_invariants": "Caratteristiche che future versioni DEVONO preservare: (1) ASCII-safe: nessun carattere > U+007F nelle directive (es. il carattere circa-uguale U+2248 era una regressione di v3.1, ripristinato in v3.2); (2) Archetipo dominante: ogni directive chiude con 'Archetipo dominante: <metafora>.' come ancora identitaria resistente a mutate_prompt_llm; (3) Lookback consigliato: ogni directive include un range numerico diversificato per stile per orientare il parametro evoluto lookback_window del genoma; (4) Metafora ancorante: la lente cognitiva e' descritta in forma 'Il mercato e ...' come prima frase; (5) Lunghezza directive: tra 800 e 950 char (sweet spot per robustezza a mutate_prompt_llm).",

-  "agent_role": "Sei un agente generatore di ipotesi di trading quantitativo per un sistema swarm coevolutivo specializzato in mercati crypto. Sei parte di una popolazione che esplora collettivamente lo spazio delle strategie: la diversita delle ipotesi e' un asset critico per il sistema. Preferisci esplorare territori meno ovvi rispetto a quelli che la tua lente cognitiva renderebbe predicibili.",
+  "agent_role": "Sei un agente generatore di ipotesi di trading quantitativo per un sistema swarm coevolutivo specializzato in mercati crypto. Sei parte di una popolazione che esplora collettivamente lo spazio delle strategie: la diversita delle ipotesi e un asset critico per il sistema. Preferisci esplorare territori meno ovvi rispetto a quelli che la tua lente cognitiva renderebbe predicibili.",

-  "pattern_guidance": "Forme di curva da cercare:\n  - Trend strutturale (direzione persistente con basso ritracciamento)\n  - Compressione di volatilita (pre-breakout, energia accumulata)\n  - Espansione di volatilita (regime di shock: momentum o cattura difensiva)\n  - Mean reversion strutturale (deviazione eccessiva dalla media -> ritorno atteso)\n  - Esaurimento direzionale (estremi di oscillatori, divergenza prezzo-momento)\n\nRipetibilita da sfruttare:\n  - Eventi crossover ricorrenti su serie correlate\n  - Cicli intra-day se la seasonality oraria e' significativa\n  - Cicli settimanali se la seasonality settimanale e' significativa\n  - Pattern doppio (top/bottom) con conferma su livello simile\n  - Range breakout dopo periodo di compressione\n\nCerca pattern che si REPLICANO nei dati storici, non singoli eventi rari.",
+  "pattern_guidance": "Forme di curva da cercare:\n  - Trend strutturale (direzione persistente con basso ritracciamento)\n  - Compressione di volatilita (pre-breakout, energia accumulata)\n  - Espansione di volatilita (regime di shock: momentum o cattura difensiva)\n  - Mean reversion strutturale (deviazione eccessiva dalla media verso ritorno atteso)\n  - Esaurimento direzionale (estremi di oscillatori, divergenza prezzo-momento)\n\nRipetibilita da sfruttare:\n  - Eventi crossover ricorrenti su serie correlate\n  - Cicli intra-day se la seasonality oraria e significativa\n  - Cicli settimanali se la seasonality settimanale e significativa\n  - Pattern doppio (top/bottom) con conferma su livello simile\n  - Range breakout dopo periodo di compressione\n\nCerca pattern che si REPLICANO nei dati storici, non singoli eventi rari.",

  "instruction": "Genera una strategia che cerchi anomalie sfruttabili in questo regime crypto.",

-  "domain_warnings": "Crypto trada 24/7 senza CME gap: non assumere chiusure settimanali. Tail pesanti e vol clustering esteso caratterizzano BTC/ETH: evita ipotesi gaussiane. NON tentare di inferire funding rate, news o eventi macro: non sono accessibili. Le statistiche fornite sono l'unica informazione su cui basarsi.",
+  "domain_warnings": "Crypto trada 24/7 senza CME gap: non assumere chiusure settimanali. Tail pesanti e vol clustering esteso caratterizzano BTC/ETH: evita ipotesi gaussiane. NON tentare di inferire funding rate, news o eventi macro: non sono accessibili. Le statistiche fornite sono l'unica informazione su cui basarsi. Una seasonality_hour o seasonality_dow > 0 NON significa che la stagionalita sia significativa: usa la soglia 0.05 come gate minimo (sotto e' rumore statistico).",

-  "anti_patterns": "Evita: (1) basare strategia su singolo evento estremo (kurt outlier senza ricorrenza nelle 500 barre recenti); (2) usare piu di 4 condizioni in AND (sovra-fitting combinatorio, brittle a piccoli shift di regime); (3) confondere correlazione storica con causalita (autocorr o pattern temporali sono BIAS, non leggi); (4) assumere stazionarieta perfetta del regime (i delta 'recente vs baseline' indicano lo scostamento); (5) usare feature temporali (hour, dow, is_weekend) quando la seasonality corrispondente e' < 0.05 (rumore, no signal).",
+  "anti_patterns": "Evita: (1) basare strategia su singolo evento estremo (kurt outlier senza ricorrenza nelle 500 barre recenti); (2) usare piu di 4 condizioni in AND (sovra-fitting combinatorio, brittle a piccoli shift di regime); (3) confondere correlazione storica con causalita (autocorr o pattern temporali sono BIAS, non leggi); (4) assumere stazionarieta perfetta del regime (i delta 'recente vs baseline' indicano lo scostamento); (5) usare feature temporali (hour, dow, is_weekend) quando la seasonality corrispondente e' < 0.05 (rumore, no signal); (6) crossover tra indicatori dello stesso tipo con lookback vicini (chattering: oscilla tra entry/exit senza segnale netto); (7) soglie hard senza isteresi tra entry ed exit (genera chattering al confine; usa soglie diverse per entry vs exit, almeno 10-20% di gap).",

-  "output_priorities": "Quando emerge trade-off: (1) robustezza cross-regime > ottimalita su singolo regime; (2) semplicita interpretabile (2-3 condizioni con razionale chiaro) > complessita raffinata (5+ condizioni accordate); (3) selettivita (poche entry forti, alto SNR) > attivita (molte entry deboli); (4) condizioni con razionale meccanico > pattern statistici fortuiti.",
+  "output_priorities": "Quando emerge trade-off: (1) coerenza con la lente cognitiva: la strategia deve essere riconoscibile come emanata dal tuo stile (es. engineer = pochi gate robusti, psychologist = contrarian su estremi). E' il fondamento del design swarm: ipotesi omogeneizzate riducono la diversita della popolazione; (2) robustezza cross-regime > ottimalita su singolo regime; (3) semplicita interpretabile (2-3 condizioni con razionale chiaro) > complessita raffinata (5+ condizioni accordate); (4) selettivita (poche entry forti, alto SNR) > attivita (molte entry deboli); (5) condizioni con razionale meccanico > pattern statistici fortuiti.",

  "styles": {
    "physicist": {
-      "directive": "Il mercato e un sistema fisico con energia (std), simmetrie (skew) e memoria (autocorr). Leggi kurt come densita di eventi estremi (fat tails = fuori equilibrio), skew come forzante asimmetrica. AR(1) positivo significativamente sopra baseline = memoria coerente, momentum legittimo; Hurst > 0.55 conferma persistenza di scala; vol percentile alto + kurt bassa = energia immagazzinata non ancora rilasciata; efficiency_ratio alto = movimento efficiente; spectral_entropy bassa + dominant_cycle definito = modi armonici sfruttabili. Pattern coerenti su piu scale temporali sono robusti, pattern singoli sono rumore.",
+      "directive": "Il mercato e un sistema fisico con energia (std), simmetrie (skew) e memoria (autocorr). Leggi kurt come densita di eventi estremi (fat tails = fuori equilibrio), skew come forzante asimmetrica. AR(1) positivo molto sopra baseline = memoria coerente, costruisci ipotesi di momentum legittimo; Hurst > 0.55 conferma persistenza di scala su orizzonti multipli; vol_pct alto con kurt bassa = energia immagazzinata non ancora rilasciata, cattura il rilascio; efficiency_ratio elevato = movimento efficiente, sfrutta direzionalita; spectral_entropy bassa con dominant_cycle definito = modi armonici sfruttabili, combina con conferma sulla fase. Preferisci pattern coerenti su piu lookback rispetto a singoli eventi rumorosi. Diagnostica regimi simmetrici e rotture di simmetria. Lookback consigliato: 150-300 barre. Archetipo dominante: sistema fisico in equilibrio (o pre-rottura di simmetria).",
      "focus_metrics": ["hurst", "dominant_cycle", "efficiency_ratio", "spectral_entropy"]
    },
-
    "biologist": {
-      "directive": "Il mercato e un ecosistema dove strategie competono per alpha finito. Skew negativo = predazione asimmetrica (vol-selling crowded che subisce shock), positivo = predatori che cacciano breakout. Kurt alta = eventi di estinzione/fioritura. AR(1) positivo persistente = una specie sta colonizzando la nicchia (overcrowding imminente, fade); Hurst > 0.55 + vol percentile basso = nicchia stabile (occupa); tail asimmetrico (left << right) = predazione asimmetrica strutturale; structural_uptrend persistente = specie dominante stabile. Cattura la coda opposta al consensus.",
+      "directive": "Il mercato e un ecosistema dove strategie competono per alpha finito. Skew negativo segnala predazione asimmetrica (vol-selling crowded subisce shock), positivo predatori che cacciano breakout. Kurt alta = eventi di estinzione o fioritura. AR(1) positivo persistente = una specie sta colonizzando la nicchia (overcrowding imminente, preferisci fade); Hurst > 0.55 con vol_pct basso = nicchia stabile (occupa con strategie direzionali); tail asimmetrico (left molto piu pesante di right) = predazione asimmetrica strutturale, costruisci contrarian sulla coda; structural_uptrend persistente = specie dominante stabile. Combina seasonality con uno o due gate di regime per evitare di sovrapporti a fasi gia mature. Cattura la coda opposta al consensus. Lookback consigliato: 80-200 barre. Archetipo dominante: ecosistema con dinamiche predator-prey e nicchie evolutive.",
      "focus_metrics": ["tail_left", "tail_right", "structural_uptrend", "seasonality_hour"]
    },
-
    "historian": {
-      "directive": "Il mercato attraversa fasi cicliche che si ripetono in forma simile. Mean = drift strutturale, std = ampiezza ciclo. Kurt alta + vol regime medium/high = fase tardiva (pre-transizione); kurt bassa + skew ≈0 = accumulazione/stabilita. AR(1) recente >> baseline storica = regime accelera rispetto al normale; Hurst > 0.55 + vol percentile alto = fase markup matura, mean reversion attesa; structural_uptrend sostenuto = fase di accumulo/markup; compression < 1 = consolidamento pre-fase nuova. Identifica analogie tra il regime corrente e fasi tipiche.",
+      "directive": "Il mercato attraversa fasi cicliche che si ripetono in forma simile. Mean = drift strutturale, std = ampiezza ciclo, kurt alta + vol regime medium/high = fase tardiva (pre-transizione); kurt bassa + skew vicino a zero = fase di accumulazione o stabilita. AR(1) recente molto sopra baseline storica = regime accelera rispetto al normale, diagnostica se markup o distribuzione; Hurst > 0.55 con vol_pct alto = fase markup matura, costruisci ipotesi di mean reversion strutturale attesa; structural_uptrend sostenuto = fase di accumulo o markup attiva, sfrutta la direzionalita; compression < 1 = consolidamento pre-fase nuova, preferisci breakout direzionali a conferma di rottura. Identifica analogie tra il regime corrente e fasi tipiche (accumulazione, markup, distribuzione, markdown). Lookback consigliato: 200-500 barre. Archetipo dominante: ciclo storico ricorrente in fasi tipiche.",
      "focus_metrics": ["autocorr_recent", "structural_uptrend", "compression", "hurst"]
    },
-
    "meteorologist": {
-      "directive": "La volatilita ha climi persistenti e fronti di transizione. Vol_regime + std + kurt definiscono microclima: std bassa + kurt bassa = calma stabile (vendi vol), std alta + kurt alta = tempesta (compra convexity), std bassa + kurt alta = calma ingannevole (riduci esposizione). AR(1) recente > baseline = fronte persistente in arrivo; Hurst > 0.55 = sistema su scala lunga, Hurst < 0.45 = turbolenza locale; vol percentile estremo = posizione nel ciclo seasonal; compression < 1 = compressione vol (pre-rilascio); tail pesanti = clima instabile. Pattern multi-regime sono robusti.",
+      "directive": "La volatilita ha climi persistenti e fronti di transizione. Vol_regime + std + kurt definiscono il microclima: std bassa + kurt bassa = calma stabile (preferisci vendere vol con gate sicuri); std alta + kurt alta = tempesta (compra convexity o resta flat); std bassa + kurt alta = calma ingannevole pre-fronte (riduci esposizione). AR(1) recente sopra baseline = fronte persistente in arrivo, cattura la direzione del fronte; Hurst > 0.55 = sistema su scala lunga (ciclone), Hurst < 0.45 = turbolenza locale (no trend persistente, preferisci range-trading); vol_pct estremo = posizione nel ciclo seasonal, modula la size; compression < 1 = compressione vol pre-rilascio, posizionati per il breakout. Costruisci strategie con gate espliciti su vol che attivano logiche diverse. Lookback consigliato: 50-150 barre. Archetipo dominante: clima atmosferico con fronti e regimi persistenti.",
      "focus_metrics": ["vol_pct", "compression", "tail_right", "dominant_cycle"]
    },
-
    "engineer": {
-      "directive": "Tratta ogni segnale come un sistema di controllo: SNR favorevole, causalita, robustezza. Std e il rumore di fondo. Kurt alta riduce affidabilita dei segnali medi. AR(1) > 0.05 con std contenuta = SNR favorevole; AR(1) ≈0 = random walk, non costruirci; Hurst < 0.45 = filtro mean-reversion causale efficace; efficiency_ratio < 0.2 = no signal, non costruire; spectral_entropy > 0.8 = white noise, regime non modellabile; tail_index < 2.5 = saturazione, riduci leverage; seasonality < 0.05 = feature temporali sono rumore. Robustezza > ottimalita.",
+      "directive": "Tratta ogni segnale come un sistema di controllo: serve SNR favorevole, causalita, robustezza. Std e il rumore di fondo. Kurt alta riduce affidabilita dei segnali medi (gli estremi dominano le statistiche). AR(1) > 0.05 con std contenuta = SNR favorevole, costruisci ipotesi di momentum filtrato; AR(1) vicino a zero = random walk, evita di costruire signal su questo; Hurst < 0.45 = filtro mean-reversion causale efficace, sfrutta con isteresi; efficiency_ratio < 0.2 = no signal, non costruire; spectral_entropy > 0.8 = white noise, regime non modellabile; tail_index < 2.5 = saturazione dei sensori, riduci leverage; seasonality < 0.05 = feature temporali sono rumore, NON usarle. Preferisci pattern semplici e tarabili: poche condizioni in AND, soglie con margine, isteresi entry/exit. Lookback consigliato: 60-120 barre. Archetipo dominante: sistema di controllo ingegneristico con SNR e robustezza.",
      "focus_metrics": ["efficiency_ratio", "spectral_entropy", "tail_left", "autocorr_recent"]
    },
-
    "military_strategist": {
-      "directive": "Distingui campagna offensiva da campagna difensiva e adatta dottrina. Vol regime medium/low + skew positivo + kurt moderata = terreno offensivo (entry direzionali). Vol regime high + kurt elevata = terreno ostile (difesa: posizioni limitate). AR(1) > 0 = vento alle spalle, carica con momentum; structural_uptrend > 0.7 = terreno occupato, hold; compression < 0.5 = preparazione attacco, posizionati; vol percentile alta = artiglieria nemica, ritirata; seasonality forte = via predicibile. Concentrazione: poche condizioni forti. Sorpresa: contrarian su consensus estremo.",
+      "directive": "Distingui campagna offensiva da campagna difensiva e adatta dottrina. Vol regime medium/low + skew positivo + kurt moderata = terreno favorevole all'attacco (costruisci entry direzionali su breakout o momentum). Vol regime high + kurt elevata = terreno ostile (difesa: posizioni limitate, exit rapide, gate restrittivi). AR(1) > 0 = vento alle spalle, carica con momentum; AR(1) negativo = imboscata possibile, preferisci contrarian; Hurst > 0.55 = posizione difendibile, hold trade; structural_uptrend > 0.7 = terreno occupato dall'avversario (decidi se attaccare o ritirarti); compression < 0.5 = preparazione attacco silenziosa, posizionati per breakout; vol_pct alta = artiglieria nemica attiva, ritirata. Concentrazione: poche condizioni forti. Sorpresa: contrarian su consensus estremo. Lookback consigliato: 100-200 barre. Archetipo dominante: stratega militare che bilancia offesa e difesa.",
      "focus_metrics": ["structural_uptrend", "compression", "vol_pct", "tail_left"]
    },
-
    "psychologist": {
-      "directive": "Il mercato e folla con emozioni misurabili. Skew e kurt sono il termometro emotivo: skew neg + kurt alta = paura ricorrente (capitulation, fade gli estremi ribasso); skew pos + kurt alta = euforia (FOMO, fade rialzo); skew ≈0 + kurt bassa = apatia. AR(1) recente >> baseline = euforia coordinata in corso, posizionati contro l'ultimo arrivato; Hurst > 0.55 = trance collettiva (dura piu del razionale); tail_left pesante (Hill < 2.5) = paura sistemica; spectral_entropy alta = caos comportamentale; vol percentile estremo = momentum emozionale. Contrarian sugli estremi.",
+      "directive": "Il mercato e folla con emozioni misurabili. Skew e kurt sono il termometro emotivo: skew neg + kurt alta = paura ricorrente (capitulation spikes, cattura il rimbalzo); skew pos + kurt alta = euforia (FOMO spikes, preferisci fade gli estremi al rialzo); skew vicino a zero + kurt bassa = apatia o range (gioca i bordi del range). AR(1) recente molto sopra baseline = euforia coordinata in corso, posizionati contro l'ultimo arrivato; Hurst > 0.55 = trance collettiva (trend trance, dura piu del razionale); tail_left pesante (Hill < 2.5) = paura sistemica strutturale, contrarian sulla capitulation; spectral_entropy alta = caos comportamentale, riduci dimensionalita del signal; vol_pct estremo = momentum emozionale puro, fade gli estremi. Sfrutta crossover di oscillatori in regimi razionali (kurt vicina a 3). Lookback consigliato: 50-120 barre. Archetipo dominante: psicologo del comportamento collettivo.",
      "focus_metrics": ["tail_left", "tail_right", "autocorr_recent", "spectral_entropy"]
    }
  }
@@ -0,0 +1,140 @@
+{
+  "rules": [
+    {
+      "condition": {
+        "op": "and",
+        "args": [
+          {
+            "op": "gt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "realized_vol",
+                "params": [150]
+              },
+              {
+                "kind": "literal",
+                "value": 0.01
+              }
+            ]
+          },
+          {
+            "op": "lt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "atr_pct",
+                "params": [150]
+              },
+              {
+                "kind": "literal",
+                "value": 0.02
+              }
+            ]
+          },
+          {
+            "op": "gt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "sma_pct",
+                "params": [150]
+              },
+              {
+                "kind": "literal",
+                "value": 0.05
+              }
+            ]
+          }
+        ]
+      },
+      "action": "entry-long"
+    },
+    {
+      "condition": {
+        "op": "and",
+        "args": [
+          {
+            "op": "lt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "realized_vol",
+                "params": [150]
+              },
+              {
+                "kind": "literal",
+                "value": 0.005
+              }
+            ]
+          },
+          {
+            "op": "gt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "atr_pct",
+                "params": [150]
+              },
+              {
+                "kind": "literal",
+                "value": 0.03
+              }
+            ]
+          },
+          {
+            "op": "lt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "sma_pct",
+                "params": [150]
+              },
+              {
+                "kind": "literal",
+                "value": -0.05
+              }
+            ]
+          }
+        ]
+      },
+      "action": "entry-short"
+    },
+    {
+      "condition": {
+        "op": "or",
+        "args": [
+          {
+            "op": "eq",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "sma_pct",
+                "params": [150]
+              },
+              {
+                "kind": "literal",
+                "value": 0.0
+              }
+            ]
+          },
+          {
+            "op": "lt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "realized_vol",
+                "params": [150]
+              },
+              {
+                "kind": "literal",
+                "value": 0.001
+              }
+            ]
+          }
+        ]
+      },
+      "action": "exit"
+    }
+  ]
+}
@@ -0,0 +1,84 @@
+{
+  "rules": [
+    {
+      "condition": {
+        "op": "and",
+        "args": [
+          {
+            "op": "gt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "realized_vol",
+                "params": [
+                  150
+                ]
+              },
+              {
+                "kind": "literal",
+                "value": 0.007
+              }
+            ]
+          },
+          {
+            "op": "lt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "atr_pct",
+                "params": [
+                  150
+                ]
+              },
+              {
+                "kind": "literal",
+                "value": 0.0042
+              }
+            ]
+          }
+        ]
+      },
+      "action": "entry-long"
+    },
+    {
+      "condition": {
+        "op": "or",
+        "args": [
+          {
+            "op": "crossunder",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "rsi",
+                "params": [
+                  14
+                ]
+              },
+              {
+                "kind": "literal",
+                "value": 70.0
+              }
+            ]
+          },
+          {
+            "op": "gt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "atr_pct",
+                "params": [
+                  150
+                ]
+              },
+              {
+                "kind": "literal",
+                "value": 0.007
+              }
+            ]
+          }
+        ]
+      },
+      "action": "exit"
+    }
+  ]
+}
@@ -0,0 +1,120 @@
+{
+  "rules": [
+    {
+      "condition": {
+        "op": "and",
+        "args": [
+          {
+            "op": "lt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "rsi",
+                "params": [
+                  14
+                ]
+              },
+              {
+                "kind": "literal",
+                "value": 30.0
+              }
+            ]
+          },
+          {
+            "op": "gt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "atr_pct",
+                "params": [
+                  14
+                ]
+              },
+              {
+                "kind": "literal",
+                "value": 0.01
+              }
+            ]
+          },
+          {
+            "op": "lt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "macd_pct",
+                "params": [
+                  12,
+                  26,
+                  9
+                ]
+              },
+              {
+                "kind": "literal",
+                "value": -0.005
+              }
+            ]
+          }
+        ]
+      },
+      "action": "entry-long"
+    },
+    {
+      "condition": {
+        "op": "or",
+        "args": [
+          {
+            "op": "gt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "rsi",
+                "params": [
+                  14
+                ]
+              },
+              {
+                "kind": "literal",
+                "value": 70.0
+              }
+            ]
+          },
+          {
+            "op": "lt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "atr_pct",
+                "params": [
+                  14
+                ]
+              },
+              {
+                "kind": "literal",
+                "value": 0.005
+              }
+            ]
+          },
+          {
+            "op": "gt",
+            "args": [
+              {
+                "kind": "indicator",
+                "name": "macd_pct",
+                "params": [
+                  12,
+                  26,
+                  9
+                ]
+              },
+              {
+                "kind": "literal",
+                "value": 0.005
+              }
+            ]
+          }
+        ]
+      },
+      "action": "exit"
+    }
+  ]
+}
@@ -34,8 +34,10 @@ def test_frontend_imports() -> None:
 def test_strategies_json_loadable() -> None:
    files = importlib.resources.files("strategy_crypto") / "strategies"
    found = sorted(p.name for p in files.iterdir() if p.name.endswith(".json"))
-    assert "btc_fb63e851.json" in found
-    assert "eth_facd6af85d5d.json" in found
+    # Convenzione: almeno un winner BTC e uno ETH shippati col package.
+    # Non hard-codare hash specifici: cambiano ad ogni swap di paper winner.
+    assert any(n.startswith("btc_") for n in found), f"no btc_*.json in {found}"
+    assert any(n.startswith("eth_") for n in found), f"no eth_*.json in {found}"
    for fname in found:
        data = json.loads((files / fname).read_text())
        assert "rules" in data, f"{fname} missing 'rules' key"
@@ -0,0 +1,70 @@
+"""Regression guard: end-of-window snap deve seguire il timeframe dell'asset.
+
+Bug originale (scripts/run_paper_trading.py): ``end = now.replace(minute=0,...)``
+snappava sempre all'ora; ETH 5m veniva quindi valutato 1 volta ogni 60 min
+invece di ogni 5 min, riducendo la fedelta' al backtest tunato 5m.
+"""
+
+from __future__ import annotations
+
+import importlib.util
+import sys
+from datetime import UTC, datetime
+from pathlib import Path
+
+import pytest
+
+_REPO_ROOT = Path(__file__).resolve().parents[3]
+_RUNNER_PATH = _REPO_ROOT / "scripts" / "run_paper_trading.py"
+
+
+def _load_runner_module():
+    spec = importlib.util.spec_from_file_location("run_paper_trading", _RUNNER_PATH)
+    assert spec is not None and spec.loader is not None
+    module = importlib.util.module_from_spec(spec)
+    sys.modules["run_paper_trading"] = module
+    spec.loader.exec_module(module)
+    return module
+
+
+@pytest.fixture(scope="module")
+def runner():
+    return _load_runner_module()
+
+
+@pytest.mark.parametrize(
+    "now, tf, expected",
+    [
+        # 5m: snap al boundary di 5 min, NON all'ora
+        (datetime(2026, 5, 18, 14, 37, 42, tzinfo=UTC), "5m", datetime(2026, 5, 18, 14, 35, tzinfo=UTC)),
+        (datetime(2026, 5, 18, 14, 35, 0, tzinfo=UTC), "5m", datetime(2026, 5, 18, 14, 35, tzinfo=UTC)),
+        (datetime(2026, 5, 18, 14, 34, 59, tzinfo=UTC), "5m", datetime(2026, 5, 18, 14, 30, tzinfo=UTC)),
+        # 1h: comportamento storico preservato
+        (datetime(2026, 5, 18, 14, 37, 42, tzinfo=UTC), "1h", datetime(2026, 5, 18, 14, 0, tzinfo=UTC)),
+        (datetime(2026, 5, 18, 14, 0, 0, tzinfo=UTC), "1h", datetime(2026, 5, 18, 14, 0, tzinfo=UTC)),
+        # 15m / 4h
+        (datetime(2026, 5, 18, 14, 22, 0, tzinfo=UTC), "15m", datetime(2026, 5, 18, 14, 15, tzinfo=UTC)),
+        (datetime(2026, 5, 18, 14, 22, 0, tzinfo=UTC), "4h", datetime(2026, 5, 18, 12, 0, tzinfo=UTC)),
+    ],
+)
+def test_align_end_to_timeframe(runner, now, tf, expected) -> None:
+    assert runner._align_end_to_timeframe(now, tf) == expected
+
+
+def test_align_end_5m_advances_every_5_minutes(runner) -> None:
+    """Bug-regression: chiamate consecutive a 5 min di distanza devono
+    produrre end DIVERSI per tf=5m (prima del fix erano identici)."""
+    a = datetime(2026, 5, 18, 14, 30, 0, tzinfo=UTC)
+    b = datetime(2026, 5, 18, 14, 35, 0, tzinfo=UTC)
+    c = datetime(2026, 5, 18, 14, 40, 0, tzinfo=UTC)
+    ends = {runner._align_end_to_timeframe(t, "5m") for t in (a, b, c)}
+    assert len(ends) == 3
+
+
+def test_align_end_1h_stable_within_hour(runner) -> None:
+    """Per tf=1h, chiamate dentro la stessa ora devono dare lo stesso end."""
+    ends = {
+        runner._align_end_to_timeframe(datetime(2026, 5, 18, 14, m, 0, tzinfo=UTC), "1h")
+        for m in (0, 15, 30, 45, 59)
+    }
+    assert ends == {datetime(2026, 5, 18, 14, 0, tzinfo=UTC)}
@@ -0,0 +1,110 @@
+# Libro dei Frattali — Riassunto
+
+**Autori:** Luca Serleto, Corrado Malanga
+**Editore/Stampa:** Adobe InDesign, nov 2024
+**Pagine:** 58 — **testo embedded ≈ nullo (~1.6 KB)**
+**Estratto raw:** [_extracted/Libro_frattali.txt](_extracted/Libro_frattali.txt)
+
+## Natura del documento
+
+Il "Libro dei Frattali" è un **catalogo grafico** di pattern candlestick, non un testo discorsivo. Le pagine sono composte quasi interamente da figure (candle-chart estratti via trasformata di Fourier secondo la metodologia esposta nel companion paper [Pythagoras Trading Prediction](Pythagoras_Trading_Prediction.summary.md)).
+
+L'estrazione testuale rivela **solo le etichette**:
+
+| Pagina | Contenuto testuale |
+|---|---|
+| 1 | "Luca Serleto / Corrado Malanga / Libro dei Frattali" (frontespizio) |
+| 2 | "Pattern da 3 candele a massimo 6 candele (LONG)" |
+| 3–58 | "Pattern 1", "Pattern 2", …, "Pattern 57" — una riga per pagina |
+
+**Tutto il resto è immagine.** L'utente ha esplicitamente richiesto che gli agenti **non passino dalle immagini** ma costruiscano schemi dai numeri.
+
+## Struttura numerica deducibile (senza guardare le figure)
+
+### Conteggio dei pattern
+- **Totale pattern catalogati: 57**
+- Numerati progressivamente da 1 a 57
+- Tutti dichiarati "**LONG**" (segnali rialzisti)
+- Tutti nel range **3–6 candele** (range 1 della tassonomia esposta in *Pythagoras Trading Prediction*, pp. 49)
+
+### Coerenza col paper companion
+Il paper *Pythagoras Trading Prediction* (p. 49) afferma testualmente:
+> "Questa tipologia di pattern può estendersi fino a 6 candele, per questo motivo nel libro i pattern sono identificati con range da 3 a 6, da 6 a 12, da 12 a 24, da 24 a 39 e da 39 a 56."
+
+⇒ Il Libro dei Frattali pubblicato copre **solo il primo range** (3–6 candele). Il numero massimo (56) della tassonomia coincide quasi esattamente con il numero di pattern catalogati (57) — possibile errore tipografico o discrepanza intenzionale di 1 unità.
+
+### Pattern composti citati nel paper
+Sempre nel paper companion (p. 53), è dichiarato che pattern di range superiore sono composti dai pattern semplici di range 1:
+- **Pattern 13 = Pattern 3 + Pattern 11**
+
+Implicazione: la numerazione del Libro NON è una sequenza temporale ma una **enumerazione di forme distinte** che fungono da "lettere" dell'alfabeto frattale; i pattern complessi sono "parole" composte.
+
+## Spazio numerico (3–6 candele LONG)
+
+Per i 57 pattern, l'**unico contenuto numerico estraibile senza vision** è il loro indice 1–57. Tuttavia il **dominio combinatorio** che il catalogo intende coprire può essere ricostruito teoricamente:
+
+### Encoding compatto di una candela
+Una candela in approssimazione discreta può essere ridotta a un simbolo dell'alfabeto:
+- `U` = up (close > open) — bullish body
+- `D` = down (close < open) — bearish body
+- `0` = doji (close ≈ open)
+
+Quindi una sequenza di 3–6 candele appartiene allo spazio `{U,D,0}^k` con k ∈ {3,4,5,6}.
+
+| k | |Σᵏ| = 3ᵏ |
+|---|---|
+| 3 | 27 |
+| 4 | 81 |
+| 5 | 243 |
+| 6 | 729 |
+| **Tot** | **1080** sequenze teoriche |
+
+57 pattern catalogati su 1080 sequenze ⇒ il catalogo seleziona ≈ **5.3%** dello spazio totale.
+
+### Filtro implicito "LONG"
+"LONG" implica che il pattern preceda un breakout/continuazione al rialzo. Restrizione plausibile:
+- pattern terminanti con candela bullish forte, o
+- pattern con dominanza U / 0 nelle ultime k-1 candele, o
+- pattern strutturali noti dell'analisi tecnica (martello, inverted hammer, morning star, ecc.) — limitati a 3–6 candele.
+
+## Numeri ricavabili dal Libro (l'unico set "puro")
+
+Dal solo testo del Libro, gli agenti hanno accesso a:
+1. **Cardinalità del catalogo**: 57
+2. **Numerazione**: serie 1, 2, 3, …, 57
+3. **Range candele**: [3, 6] (chiuso, intero)
+4. **Direzione**: solo LONG
+5. **Costante autori**: 2 (Serleto, Malanga)
+6. **Edizione**: nov 2024
+7. **Numero di pagine**: 58
+
+## Indizi numerici per l'analisi di similitudine
+
+| Numero | Match con Pythagoras Trading Prediction |
+|---|---|
+| 57 pattern | tassonomia bordo superiore = 56 (errore di 1) |
+| 56 (bordo tassonomia) | numerologia: 56 = 8·7 = ottava·sacro |
+| 3 (min candele) | numero sacro pitagorico (triade); 3 = lati triangolo Evideon |
+| 6 (max candele) | numero perfetto (1+2+3); φ²-... |
+| 6/3 = 2 | ratio 2 = ottava musicale |
+| (max-min)/min = 1 | unità |
+| log(57)/log(3) ≈ 3.69 | dimensione frattale di Hausdorff di insieme dimensionale 4 ≈ 3.69 |
+
+## Punti utili per estrazione numerica (per il task di similitudini)
+
+1. **N = 57** pattern → cardinality, da confrontare con 56 (bordo del paper), 528 Hz (Solfeggio), 588 candele indicatore
+2. **k ∈ {3, 4, 5, 6}** → estremi del range
+3. **Σ³⁺⁴⁺⁵⁺⁶ = 1080** sequenze teoriche (3ᵏ encoded) → 1080 = 8·135 = 2³·3³·5
+4. **Tassonomia bordi**: 3, 6, 12, 24, 39, 56 — sequenza utile per fingerprint
+5. **Direzione "LONG"** → variabile binaria
+6. **Composizionalità**: Pattern 13 = Pattern 3 + Pattern 11 → grammar di pattern frattali
+
+## Conclusione
+
+Il Libro dei Frattali, **letto senza immagini**, è un oggetto numerico molto povero (57 etichette su 58 pagine). Per produrre similitudini significative tra i due testi, occorre:
+
+1. Trattare il Libro come **schema combinatorio implicito** (1080 sequenze possibili, 57 selezionate)
+2. Confrontare i **numeri-chiave** (3, 6, 57, range) con i numeri estratti dal paper teorico (φ, π, e, 528, 588, 137.0359, etc.)
+3. Lasciare agli agenti il compito di **trovare ricorrenze, ratios, e dimensioni frattali** che colleghino i due insiemi numerici.
+
+Vedere [Pythagoras_Trading_Prediction.summary.md](Pythagoras_Trading_Prediction.summary.md) per il dominio numerico del paper teorico.
@@ -0,0 +1,182 @@
+# Pythagoras Trading Prediction — Riassunto
+
+**Autori:** Luca Serleto, Corrado Malanga
+**Editore/Stampa:** Adobe InDesign, dic 2024
+**Pagine:** 66 — testo embedded ~91 KB (no immagini estraibili: figure presenti ma non analizzate per scelta esplicita dell'utente)
+**Estratto raw:** [_extracted/Pythagoras_Trading_Prediction_.txt](_extracted/Pythagoras_Trading_Prediction_.txt)
+
+## Tesi centrale
+
+Il prezzo dei mercati finanziari è descrivibile come **funzione frattale**, scomponibile via **trasformata di Fourier** in componenti sinusoidali periodiche, e quindi proiettabile nel futuro. Il "parametro nascosto" che la fisica quantistica cerca da decenni è la **Coscienza** (entropia di Shannon ΔS), che modula sia la capacità di osservare/predire che la traiettoria stessa dei prezzi.
+
+## Struttura del libro (4 parti)
+
+1. **Parte teorica / coscienziale** (pp. 2–24): framework filosofico-fisico
+2. **Frattali** (pp. 37–44): indicatore frattale H-C
+3. **Trasformata di Fourier** (pp. 45–58): pattern catalogati e applicazione
+4. **Piattaforma Pythagoras** (pp. 59–66): manuale operativo (Auto + Manual mode)
+
+## Concetti chiave (in ordine di apparizione)
+
+### 1. Approccio "femminile" vs "maschile" alla predizione
+- **Femminile/animico**: visione olografica del tempo (Pribram, Bohm), tutto co-presente nel presente; il cervello come "lettore di ologrammi"
+- **Maschile/spirituale**: approccio galileiano-statistico via algoritmo
+- Tesi: i due metodi vanno integrati
+
+### 2. Probabilità → Certezza
+> "Se il sistema predittivo è corretto al 10%, possiamo trovare il parametro che trasformi quella probabilità nella certezza al 100%."
+
+Il parametro mancante è la **Coscienza**.
+
+### 3. Teorema di ricorrenza di Poincaré
+In meccanica hamiltoniana, in un sistema dinamico a spazio delle fasi limitato, lo stato può tornare arbitrariamente vicino a quello iniziale dopo tempo sufficiente. → Il tempo è **circolare**, gli eventi ricorrono → mercati ripetibili.
+
+### 4. Entropia di Shannon S
+- S = -log(W), W = microstati
+- Coscienza ≡ S; **Consapevolezza ≡ ΔS** (variazione di entropia tra istante e successivo)
+- Inizio universo: S = -∞; fine: S = 0; entropia totale = ∞
+
+### 5. Hartman-Curry → "Indicatore frattale H-C"
+- Introdotto da Corrado Malanga al XCongress di Pescara (2018)
+- Prima applicazione su grafico EUR/USD
+- Struttura tridimensionale costruita su "numeri universali" che genera ricorrenze
+
+### 6. Numeri Universali (Evideon)
+> "I numeri dell'universo sono: la sezione aurea φ, π, il numero di Eulero e, √2, ecc. Numeri irrazionali e trascendenti."
+
+Triangolo rettangolo con costanti universali fisiche; lati moltiplicati per opportuni coefficienti basati su π, e, φ e temperatura assoluta di Kelvin/100.
+
+**Costanti numeriche citate esplicitamente:**
+- 137.0359 × √π / 2 = **121.449**
+- 266.87 × 1 / √e = **161.86**
+- Frequenze Solfeggio: **741, 528, 852, 639, 396, 417** (p. 27)
+- Numero di Avogadro N = 6.022·10²³ (citato come ipotesi sul numero di sub-componenti dell'Universo)
+
+### 7. Trasformata di Fourier
+Scomposizione di qualsiasi funzione continua in seni e coseni:
+- Frequenze positive (orario) e negative (antiorario) → propagazione **avanti e indietro nel tempo**
+- Ricostruzione approssimata con N componenti → **N = misura della consapevolezza**
+- Pochi N: previsione approssimativa; molti N: previsione accurata
+
+### 8. Equazione di Schrödinger ↔ trading
+Identificazione formale:
+
+| Schrödinger | Trading |
+|---|---|
+| iℏ ∂Ψ/∂t = -ℏ²/(2m) ∂²Ψ/∂x² + V(x)Ψ | dinamica del prezzo |
+| derivata seconda di Ψ in x | "piegatura" spazio-temporale del grafico |
+| m | massa fittizia ≈ volume d'affari |
+| V(Ψ) | perturbazioni esterne (manipolazione di mercato) |
+| Ψ ampiezza | altezza candele (asse energia/prezzo) |
+| Ψ² probabilità | probabilità che l'evento si materializzi |
+
+Identificazione con energia libera di Gibbs: termine -TΔS della formula di Gibbs ≡ termine entropico in Schrödinger.
+
+### 9. Centro di inversione (operazione geometrica chiave)
+Sul modello "Evideon" (3 assi: spazio, tempo, energia), un punto A del passato si trasforma nel punto E del futuro tramite:
+- **Riflessione speculare sull'asse verticale** (energia/prezzo)
+- **Riflessione speculare sull'asse orizzontale** (tempo)
+- = traslazione temporale + inversione di pattern
+
+Inoltre il pattern subisce:
+- **Compressione** o **dilatazione** sull'asse x (tempo)
+- **Compressione** o **dilatazione** sull'asse y (prezzo)
+A seconda della posizione del punto sull'Evideon (tempo sotto-multiplo → contrazione).
+
+### 10. Finestra di Overton
+- Apertura temporale che definisce "il presente"
+- Analogia con SAR (Synthetic Aperture Radar): più larga la finestra, più precisa la previsione
+- Pythagoras applica la stessa finestra al passato per proiettare nel futuro
+
+### 11. ZPE (Zero Point Energy)
+Universo non vuoto: pieno di particelle e anti-particelle che si annichiliscono. Permette di "conoscere perfettamente coordinate spazio-temporali e valore nullo dell'energia": non c'è incertezza vera (contro Heisenberg).
+
+### 12. Indicatore frattale su TradingView
+Disponibile su licenza dedicata. Parametri operativi:
+- Tracciato da minimo → massimo
+- **25 linee verticali** all'interno di **588 candele**
+- 90% segnala accelerazione (aumento volatilità), 10% lateralizzazione
+- Tolleranza ±3 ore su time frame 1H/4H
+- **Linee blu inclinate** = supporti/resistenze tempo-condizionati
+- **Cerchi bianchi** = obiettivi di prezzo circolari
+
+### 13. Pattern range (taxonomy del Book dei Frattali)
+| Range | Numero candele |
+|---|---|
+| Range 1 | 3 – 6 candele |
+| Range 2 | 6 – 12 candele |
+| Range 3 | 12 – 24 candele |
+| Range 4 | 24 – 39 candele |
+| Range 5 | 39 – 56 candele |
+
+I pattern oltre 6 candele diventano "frattali complessi": composizione di pattern semplici (es. Pattern 13 = Pattern 3 + Pattern 11).
+
+### 14. Variabili che NON alterano il target
+- Inclinazione del pattern
+- Ampiezza
+- Curvatura
+- Numero di candele entro lo stesso range
+
+→ il target di proiezione è invariante per queste deformazioni.
+
+### 15. Variabili che alterano l'applicabilità
+- **Time frame**: 15m, 1H, 4H, 1D
+- **Volatilità dell'asset** ("consapevolezza dell'asset")
+- **Leva finanziaria**
+
+## Esempi concreti citati
+
+| Pattern | Asset | Data/Ora | TF |
+|---|---|---|---|
+| 1 | BTC | 28 ago 2024 00:00 | 15m |
+| 1 | ETH | 28 ago 2024 00:00 | 15m |
+| 1 | Oro | 29 ago 2024 14:30 | 15m |
+| 1 | Argento | 17 lug 2024 18:15 | 15m |
+| 13 | BTC | 24 ago 2024 23:15 | 15m |
+| 13 | ETH | 24 ago 2024 23:45 | 15m |
+| 13 | Oro | 02 ago 2024 16:45 | 15m |
+| 13 | Argento | 02 ago 2024 16:45 | 15m |
+| 15 | DJT (Trump Media) | 12 lug 2024 (chiusura) | 1D |
+
+Il caso DJT è presentato come "predizione" dell'attentato a Trump del 14 lug 2024 (frattale complesso al rialzo formatosi il 12 lug → escludeva evento negativo).
+
+## Piattaforma Pythagoras
+
+### Auto Mode
+1. Seleziona asset
+2. Seleziona TF (15m, 1H, 4H)
+3. Click "Predict" → Fourier identifica frattale e proietta futuro (colore giallo)
+
+### Manual Mode
+1. Seleziona asset
+2. Seleziona frattale dal Book per data
+3. Click "Predict" → proiezione
+
+## Disclaimer interno al testo
+
+> "Anche se si conosce un potenziale futuro, questo non è garanzia di guadagno. Solo la consapevolezza del trader, basata sull'esperienza, potrà trasformare la predizione di un potenziale futuro in un profitto concreto."
+
+Il testo enfatizza che il **target del trader non è guadagnare ma capire sé stesso** (approccio "coscienziale"). Il denaro è strumento di apprendimento dell'entropia/dualità.
+
+## Punti utili per estrazione numerica (per il task di similitudini)
+
+1. **Numeri universali esatti**: φ ≈ 1.6180339887, π ≈ 3.1415926535, e ≈ 2.71828, √2 ≈ 1.41421
+2. **Costanti derivate dal testo**: 137.0359, 121.449, 266.87, 161.86
+3. **Frequenze Solfeggio (Hz)**: 396, 417, 528, 639, 741, 852
+4. **Conteggi operativi**: 25 linee/588 candele = ratio 0.0425
+5. **Numeri ricorrenti**: 6.022·10²³ (Avogadro)
+6. **Range candele**: bordi 3, 6, 12, 24, 39, 56 — sequenza con incrementi 3, 6, 12, 15, 17 (NON Fibonacci puro, ma sequenza con ratios ~2x)
+7. **Ratios tra bordi**: 6/3=2, 12/6=2, 24/12=2, 39/24=1.625, 56/39=1.436 — gli ultimi due si avvicinano a φ ≈ 1.618 e √2 ≈ 1.414
+8. **Sequenza pattern numerati**: 1, 2, 3, ... 57 (matching col Libro dei Frattali)
+9. **Operazioni geometriche**: mirror H, mirror V, scaling x/y (4 trasformazioni)
+10. **TF candidati**: 15m, 1H, 4H, 1D (multipli 4x e 6x)
+
+## Bibliografia interna / link YouTube citati
+- youtube.com/shorts/OnM09uMCS50 (teorema di Poincaré)
+- youtu.be/hHig-PP4Mcs (gravità come Coscienza)
+- youtu.be/HTIYZBrzZbM (struttura dell'Evideon)
+
+## Opere citate degli autori
+- Corrado Malanga, *Io e Dio*, Spazio Interiore, Roma
+- Corrado Malanga, *Trilogia della Coscienza*, Spazio Interiore, Roma
+- Corrado Malanga al XCongress, Pescara, 2018 (intro pattern H-C)
@@ -0,0 +1,227 @@
+
+
+===PAGE 1===
+Luca Serleto
+Corrado Malanga
+Libro dei Libro dei 
+FrattaliFrattali
+
+===PAGE 2===
+Pattern da 3 candele a massimo 6 candele (LONG)
+Pattern 1
+
+===PAGE 3===
+Pattern 2
+
+===PAGE 4===
+Pattern 3
+
+===PAGE 5===
+Pattern 4
+
+===PAGE 6===
+Pattern 5
+
+
+===PAGE 7===
+Pattern 6
+
+
+===PAGE 8===
+Pattern 7
+
+
+===PAGE 9===
+Pattern 8
+
+
+===PAGE 10===
+Pattern 9
+
+
+===PAGE 11===
+Pattern 10
+
+
+===PAGE 12===
+Pattern 11
+
+
+===PAGE 13===
+Pattern 12
+
+
+===PAGE 14===
+Pattern 13
+
+
+===PAGE 15===
+Pattern 14
+
+
+===PAGE 16===
+Pattern 15
+
+
+===PAGE 17===
+Pattern 16
+
+
+===PAGE 18===
+Pattern 17
+
+
+===PAGE 19===
+Pattern 18
+
+
+===PAGE 20===
+Pattern 19
+
+
+===PAGE 21===
+Pattern 20
+
+
+===PAGE 22===
+Pattern 21
+
+
+===PAGE 23===
+Pattern 22
+
+
+===PAGE 24===
+Pattern 23
+
+
+===PAGE 25===
+Pattern 24
+
+
+===PAGE 26===
+Pattern 25
+
+
+===PAGE 27===
+Pattern 26
+
+
+===PAGE 28===
+Pattern 27
+
+
+===PAGE 29===
+Pattern 28
+
+
+===PAGE 30===
+Pattern 29
+
+
+===PAGE 31===
+Pattern 30
+
+
+===PAGE 32===
+Pattern 31
+
+
+===PAGE 33===
+Pattern 32
+
+
+===PAGE 34===
+Pattern 33
+
+
+===PAGE 35===
+Pattern 34
+
+
+===PAGE 36===
+Pattern 35
+
+
+===PAGE 37===
+Pattern 36
+
+
+===PAGE 38===
+Pattern 37
+
+
+===PAGE 39===
+Pattern 38
+
+
+===PAGE 40===
+Pattern 39
+
+
+===PAGE 41===
+Pattern 40
+
+
+===PAGE 42===
+Pattern 41
+
+
+===PAGE 43===
+Pattern 42
+
+
+===PAGE 44===
+Pattern 43
+
+
+===PAGE 45===
+Pattern 44
+
+
+===PAGE 46===
+Pattern 45
+
+
+===PAGE 47===
+Pattern 46
+
+
+===PAGE 48===
+Pattern 47
+
+
+===PAGE 49===
+Pattern 48
+
+
+===PAGE 50===
+Pattern 49
+
+
+===PAGE 51===
+Pattern 50
+
+
+===PAGE 52===
+Pattern 51
+
+
+===PAGE 53===
+Pattern 52
+
+
+===PAGE 54===
+Pattern 53
+
+===PAGE 55===
+Pattern 54
+
+===PAGE 56===
+Pattern 55
+
+===PAGE 57===
+Pattern 56
+
+===PAGE 58===
+Pattern 57
@@ -0,0 +1,78 @@
+# strategy_pythagoras
+
+Strategia di trading basata sul framework **Pythagoras-Malanga**
+(candle-pattern + geometria frattale), evoluta via GA sul core
+`multi_swarm_core`. Workspace member del monorepo `multi_swarm_coevolutive`.
+
+## Scope
+
+Pipeline coevolutiva su candele OHLC: il GA del core esplora combinazioni
+di indicatori candle-pattern, geometria pitagorica e ratio frattali, e
+produce strategie JSON freezate. Questo member le esegue in
+**paper-trading forward-test** ed espone una dashboard NiceGUI per
+analisi invarianza scala/tick e candle.
+
+## Layout
+
+```
+strategy_pythagoras/
+├── backend/        paper trading runner (PaperExecutor, Portfolio, PaperRepository)
+│   └── schema.py   tabelle paper_trading_* (DB locale dedicato)
+├── frontend/       NiceGUI dashboard (tab invariance, candle, equity, ticks)
+├── strategies/     JSON freezate input al runner
+│                   (pythagoras_*.json)
+└── prompts.json    7 stili di prompt LLM (candle-pattern, frattale, ratio,
+                    pivot, kagi, renko, hybrid)
+```
+
+## Run paper-trading smoke
+
+```bash
+uv run python scripts/run_pythagoras_smoke.py \
+    --name pythagoras-smoke-001 \
+    --initial-capital 1000 \
+    --poll-seconds 300
+```
+
+Il default `--strategies-dir` punta ai JSON shippati col package via
+`importlib.resources.files("strategy_pythagoras") / "strategies"`.
+
+## Dashboard
+
+```bash
+uv run python -m strategy_pythagoras.frontend.nicegui_app
+```
+
+In produzione: `https://swarm.tielogic.xyz/strategy_pythagoras_gui/`
+(root_path configurato via `DASHBOARD_ROOT_PATH=/strategy_pythagoras_gui`).
+
+## DB schema
+
+Schema isolato dal core e dalle altre strategie. Due DB distinti:
+
+- `state/strategy_pythagoras.db` — GA + analisi invarianza
+  (env `STRATEGY_PYTHAGORAS_DB_PATH`)
+- `state/strategy_pythagoras_paper.db` — paper-trading runs
+  (env `STRATEGY_PYTHAGORAS_PAPER_DB_PATH`)
+
+Tabelle paper-trading:
+
+- `paper_trading_runs` — metadata run (id, name, capital, status)
+- `paper_trading_positions` — posizioni aperte (long/short)
+- `paper_trading_trades` — trade realized (entry/exit, pnl, fees)
+- `paper_trading_equity` — equity curve snapshot
+- `paper_trading_ticks` — log signal/action per ogni bar
+
+DDL gestito da `strategy_pythagoras.backend.schema.init_schema()`.
+
+La dashboard legge **anche** il `runs.db` del core GA (env `GA_DB_PATH`)
+per correlare paper performance con i genomi di provenienza e con i
+risultati di fitness invariance.
+
+## References
+
+- Spec: `docs/superpowers/specs/2026-05-19-strategy-pythagoras-design.md`
+- Plan: `docs/superpowers/plans/2026-05-19-strategy-pythagoras.md`
+- Summary paper: `Pythagoras/Pythagoras_Trading_Prediction.summary.md`
+- Summary frattali: `Pythagoras/Libro_frattali.summary.md`
+- Pattern member: vedi `src/strategy_crypto/README.md`
@@ -0,0 +1,21 @@
+[project]
+name = "strategy-pythagoras"
+version = "0.1.0"
+description = "Strategy Pythagoras: candle-pattern GA su framework Pythagoras-Malanga, paper-trading runner + NiceGUI dashboard"
+authors = [{ name = "Adriano Dal Pastro", email = "adrianodalpastro@tielogic.com" }]
+requires-python = ">=3.13"
+dependencies = [
+    "multi-swarm-core",
+    "nicegui>=3.11.1",
+    "plotly>=5.24",
+    "pandas>=2.2",
+    "pyarrow>=18.0",
+]
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build.targets.wheel.force-include]
+"strategy_pythagoras/strategies" = "strategy_pythagoras/strategies"
+"strategy_pythagoras/prompts.json" = "strategy_pythagoras/prompts.json"
@@ -0,0 +1,23 @@
+"""Backend paper-trading per la strategia strategy_pythagoras.
+
+Espone le classi principali per import ergonomici in scripts/runner:
+
+    from strategy_pythagoras.backend import PaperExecutor, Portfolio, PaperRepository
+
+Per i tipi interni (TickResult, OpenPosition, Trade) importare dal sotto-modulo.
+"""
+
+from .executor import PaperExecutor, TickResult
+from .persistence import PaperRepository
+from .portfolio import OpenPosition, Portfolio
+from .schema import PAPER_SCHEMA_SQL, init_schema
+
+__all__ = [
+    "PAPER_SCHEMA_SQL",
+    "OpenPosition",
+    "PaperExecutor",
+    "PaperRepository",
+    "Portfolio",
+    "TickResult",
+    "init_schema",
+]
@@ -0,0 +1,97 @@
+"""PaperExecutor: applica un segnale di strategia a un Portfolio.
+
+Il flusso per ogni tick:
+
+    bar OHLCV chiuso -> compile_strategy(strategy) -> Series[Side]
+    -> last_signal = series.iloc[-1]
+    -> match con posizione attuale -> open / close / hold
+
+Niente delay 1-bar: in paper-trading il segnale viene calcolato sulla
+barra appena chiusa e applicato al prezzo close della stessa. La latenza
+reale tra tick e ordine va misurata separatamente (Phase 3 spec).
+"""
+
+from __future__ import annotations
+
+import json
+from dataclasses import dataclass
+from datetime import datetime
+from pathlib import Path
+
+import pandas as pd  # type: ignore[import-untyped]
+
+from multi_swarm_core.backtest.orders import Side, Trade
+from multi_swarm_core.protocol.compiler import compile_strategy
+from multi_swarm_core.protocol.parser import parse_strategy
+
+from .portfolio import OpenPosition, Portfolio
+
+
+@dataclass
+class TickResult:
+    ts: datetime
+    symbol: str
+    bar_ts: datetime
+    close_price: float
+    signal: Side
+    action_taken: str  # "open_long" | "open_short" | "close" | "reverse" | "hold"
+    trade: Trade | None = None
+    new_position: OpenPosition | None = None
+
+
+class PaperExecutor:
+    def __init__(self, strategy_json_path: Path, symbol: str) -> None:
+        text = strategy_json_path.read_text()
+        # parse_strategy si aspetta JSON pulito, non fence; il file e' gia' JSON.
+        self._strategy = parse_strategy(text)
+        self._compiled = compile_strategy(self._strategy)
+        self.symbol = symbol
+        self.strategy_path = strategy_json_path
+
+    def execute_tick(
+        self,
+        portfolio: Portfolio,
+        ohlcv: pd.DataFrame,
+        now: datetime,
+    ) -> TickResult:
+        """Esegui un tick: calcola segnale su tutto ``ohlcv`` (per indicatori
+        con lookback), prendi l'ultimo, e applica al portfolio."""
+        if len(ohlcv) == 0:
+            raise ValueError("Empty OHLCV passed to execute_tick")
+        signals = self._compiled(ohlcv)
+        # ultimo bar chiuso
+        bar_ts = ohlcv.index[-1]
+        close_price = float(ohlcv["close"].iloc[-1])
+        signal = Side(signals.iloc[-1]) if signals.iloc[-1] is not None else Side.FLAT
+
+        current = portfolio.positions.get(self.symbol)
+        action = "hold"
+        trade: Trade | None = None
+        new_position: OpenPosition | None = None
+
+        if current is None and signal != Side.FLAT:
+            new_position = portfolio.open(self.symbol, signal, close_price, now)
+            action = f"open_{signal.value}"
+        elif current is not None and signal == Side.FLAT:
+            trade = portfolio.close(self.symbol, close_price, now)
+            action = "close"
+        elif current is not None and signal != current.side:
+            # reverse: chiudi e riapri opposto
+            trade = portfolio.close(self.symbol, close_price, now)
+            new_position = portfolio.open(self.symbol, signal, close_price, now)
+            action = "reverse"
+
+        return TickResult(
+            ts=now,
+            symbol=self.symbol,
+            bar_ts=bar_ts.to_pydatetime() if hasattr(bar_ts, "to_pydatetime") else bar_ts,
+            close_price=close_price,
+            signal=signal,
+            action_taken=action,
+            trade=trade,
+            new_position=new_position,
+        )
+
+    @property
+    def strategy_dict(self) -> dict:
+        return json.loads(self.strategy_path.read_text())
@@ -0,0 +1,123 @@
+"""Persistenza paper-trading: scrive su un DB dedicato (state/strategy_pythagoras_paper.db)
+con le tabelle ``paper_trading_*`` definite localmente in :mod:`.schema`.
+
+Il DB e' isolato dal ``runs.db`` del core GA: nessun naming conflict con
+future strategie (state/strategy_<asset>.db), nessuna contention di lock
+fra writer GA e writer paper.
+"""
+
+from __future__ import annotations
+
+import json
+import sqlite3
+import uuid
+from datetime import UTC, datetime
+from pathlib import Path
+from typing import Any
+
+from .executor import TickResult
+from .portfolio import Portfolio
+from .schema import init_schema as _init_paper_schema
+
+
+class PaperRepository:
+    def __init__(self, db_path: Path | str):
+        self.db_path = Path(db_path)
+
+    def init_schema(self) -> None:
+        """Crea (se mancanti) le tabelle paper_trading_* su ``self.db_path``."""
+        _init_paper_schema(self.db_path)
+
+    def _conn(self) -> sqlite3.Connection:
+        conn = sqlite3.connect(self.db_path, isolation_level=None)
+        conn.row_factory = sqlite3.Row
+        conn.execute("PRAGMA foreign_keys = ON")
+        conn.execute("PRAGMA journal_mode = WAL")
+        return conn
+
+    @staticmethod
+    def _now() -> str:
+        return datetime.now(UTC).isoformat()
+
+    def create_run(self, name: str, initial_capital: float, config: dict[str, Any]) -> str:
+        rid = uuid.uuid4().hex
+        with self._conn() as conn:
+            conn.execute(
+                "INSERT INTO paper_trading_runs "
+                "(id, name, started_at, status, initial_capital, config_json) "
+                "VALUES (?,?,?,?,?,?)",
+                (rid, name, self._now(), "running", initial_capital, json.dumps(config)),
+            )
+        return rid
+
+    def stop_run(self, run_id: str, status: str = "stopped") -> None:
+        with self._conn() as conn:
+            conn.execute(
+                "UPDATE paper_trading_runs SET stopped_at=?, status=? WHERE id=?",
+                (self._now(), status, run_id),
+            )
+
+    def save_tick(self, run_id: str, tick: TickResult) -> None:
+        with self._conn() as conn:
+            conn.execute(
+                "INSERT INTO paper_trading_ticks "
+                "(paper_run_id, symbol, ts, bar_ts, close_price, signal, action_taken) "
+                "VALUES (?,?,?,?,?,?,?)",
+                (
+                    run_id,
+                    tick.symbol,
+                    tick.ts.isoformat(),
+                    tick.bar_ts.isoformat() if hasattr(tick.bar_ts, "isoformat") else str(tick.bar_ts),
+                    tick.close_price,
+                    tick.signal.value,
+                    tick.action_taken,
+                ),
+            )
+            if tick.trade is not None:
+                t = tick.trade
+                conn.execute(
+                    "INSERT INTO paper_trading_trades "
+                    "(paper_run_id, symbol, side, qty, entry_price, exit_price, "
+                    "entry_ts, exit_ts, pnl, fees) VALUES (?,?,?,?,?,?,?,?,?,?)",
+                    (
+                        run_id,
+                        tick.symbol,
+                        t.side.value,
+                        t.size,
+                        t.entry_price,
+                        t.exit_price,
+                        t.entry_ts.isoformat(),
+                        t.exit_ts.isoformat(),
+                        t.net_pnl,
+                        t.fees,
+                    ),
+                )
+
+    def save_equity_snapshot(
+        self,
+        run_id: str,
+        ts: datetime,
+        equity: float,
+        cash: float,
+        positions_value: float,
+    ) -> None:
+        with self._conn() as conn:
+            conn.execute(
+                "INSERT INTO paper_trading_equity "
+                "(paper_run_id, ts, equity, cash, positions_value) VALUES (?,?,?,?,?)",
+                (run_id, ts.isoformat(), equity, cash, positions_value),
+            )
+
+    def sync_open_positions(self, run_id: str, portfolio: Portfolio) -> None:
+        """Sostituisce snapshot posizioni aperte. Idempotente: cancella e reinserisce."""
+        with self._conn() as conn:
+            conn.execute(
+                "DELETE FROM paper_trading_positions WHERE paper_run_id=?", (run_id,)
+            )
+            for sym, pos in portfolio.positions.items():
+                conn.execute(
+                    "INSERT INTO paper_trading_positions "
+                    "(paper_run_id, symbol, side, qty, entry_price, entry_ts) "
+                    "VALUES (?,?,?,?,?,?)",
+                    (run_id, sym, pos.side.value, pos.qty, pos.entry_price, pos.entry_ts.isoformat()),
+                )
@@ -0,0 +1,104 @@
+"""Portfolio multi-asset per paper-trading.
+
+Modello semplificato: capitale unico ``cash``, allocazione equal-weight
+fra N posizioni (sleeve = 1/N del capitale iniziale per ogni simbolo).
+Niente leva, niente liquidation, fees su entry+exit (bp del notional).
+
+Una :class:`Position` rappresenta una posizione aperta su un singolo
+simbolo (long/short, qty in unita' dell'asset, prezzo di entry). La
+posizione viene chiusa con :meth:`Portfolio.close` che produce un
+:class:`Trade` realized e accredita ``cash``.
+
+Mark-to-market via :meth:`Portfolio.equity`.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from datetime import datetime
+
+from multi_swarm_core.backtest.orders import Side, Trade
+
+
+@dataclass(frozen=True)
+class OpenPosition:
+    symbol: str
+    side: Side
+    qty: float
+    entry_price: float
+    entry_ts: datetime
+
+
+@dataclass
+class Portfolio:
+    initial_capital: float
+    fees_bp: float = 5.0
+    n_sleeves: int = 2  # numero strategie / asset previsti
+    cash: float = field(init=False)
+    positions: dict[str, OpenPosition] = field(default_factory=dict)
+    closed_trades: list[Trade] = field(default_factory=list)
+
+    def __post_init__(self) -> None:
+        self.cash = self.initial_capital
+
+    @property
+    def sleeve_capital(self) -> float:
+        return self.initial_capital / self.n_sleeves
+
+    def open(
+        self,
+        symbol: str,
+        side: Side,
+        price: float,
+        ts: datetime,
+    ) -> OpenPosition:
+        if symbol in self.positions:
+            raise ValueError(f"Position already open on {symbol}")
+        if side == Side.FLAT:
+            raise ValueError("Cannot open a FLAT position")
+        # sleeve fisso: alloca 1/n_sleeves del capitale iniziale, qty = notional/price.
+        notional = self.sleeve_capital
+        qty = notional / price
+        fees = notional * (self.fees_bp / 10000.0)
+        self.cash -= fees
+        pos = OpenPosition(symbol=symbol, side=side, qty=qty, entry_price=price, entry_ts=ts)
+        self.positions[symbol] = pos
+        return pos
+
+    def close(
+        self,
+        symbol: str,
+        price: float,
+        ts: datetime,
+    ) -> Trade:
+        if symbol not in self.positions:
+            raise ValueError(f"No open position on {symbol}")
+        pos = self.positions.pop(symbol)
+        trade = Trade(
+            entry_ts=pos.entry_ts,
+            exit_ts=ts,
+            side=pos.side,
+            size=pos.qty,
+            entry_price=pos.entry_price,
+            exit_price=price,
+            fees_bp=self.fees_bp,
+        )
+        # net_pnl include gia' i fees sull'intero round-trip; abbiamo gia'
+        # addebitato meta' fees all'open, ora addebitiamo il resto.
+        self.cash += trade.gross_pnl - (trade.fees / 2.0)
+        self.closed_trades.append(trade)
+        return trade
+
+    def equity(self, last_prices: dict[str, float]) -> tuple[float, float]:
+        """Ritorna (equity_totale, positions_value) marcando posizioni aperte
+        al ``last_prices[symbol]``. Posizioni senza prezzo disponibile valgono
+        notional di entry (fallback conservativo)."""
+        positions_value = 0.0
+        for sym, pos in self.positions.items():
+            price = last_prices.get(sym, pos.entry_price)
+            unreal = pos.qty * (
+                price - pos.entry_price if pos.side == Side.LONG
+                else pos.entry_price - price
+            )
+            positions_value += pos.qty * pos.entry_price + unreal
+        return self.cash + positions_value, positions_value
@@ -0,0 +1,87 @@
+"""Schema SQLite per le tabelle paper-trading della strategia pythagoras.
+
+Owned dal member strategy_pythagoras: il DDL e' standalone rispetto al core,
+e scrive su un database dedicato (state/strategy_pythagoras_paper.db, env
+STRATEGY_PYTHAGORAS_PAPER_DB_PATH) separato dal runs.db del core GA. Pattern
+replicabile per future strategie.
+"""
+
+from __future__ import annotations
+
+import sqlite3
+from pathlib import Path
+
+PAPER_SCHEMA_SQL = """
+CREATE TABLE IF NOT EXISTS paper_trading_runs (
+    id TEXT PRIMARY KEY,
+    name TEXT NOT NULL,
+    started_at TEXT NOT NULL,
+    stopped_at TEXT,
+    status TEXT NOT NULL DEFAULT 'running',
+    initial_capital REAL NOT NULL,
+    config_json TEXT NOT NULL
+);
+
+CREATE TABLE IF NOT EXISTS paper_trading_positions (
+    paper_run_id TEXT NOT NULL,
+    symbol TEXT NOT NULL,
+    side TEXT NOT NULL,
+    qty REAL NOT NULL,
+    entry_price REAL NOT NULL,
+    entry_ts TEXT NOT NULL,
+    PRIMARY KEY (paper_run_id, symbol),
+    FOREIGN KEY (paper_run_id) REFERENCES paper_trading_runs(id)
+);
+
+CREATE TABLE IF NOT EXISTS paper_trading_trades (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    paper_run_id TEXT NOT NULL,
+    symbol TEXT NOT NULL,
+    side TEXT NOT NULL,
+    qty REAL NOT NULL,
+    entry_price REAL NOT NULL,
+    exit_price REAL NOT NULL,
+    entry_ts TEXT NOT NULL,
+    exit_ts TEXT NOT NULL,
+    pnl REAL NOT NULL,
+    fees REAL NOT NULL,
+    FOREIGN KEY (paper_run_id) REFERENCES paper_trading_runs(id)
+);
+
+CREATE TABLE IF NOT EXISTS paper_trading_equity (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    paper_run_id TEXT NOT NULL,
+    ts TEXT NOT NULL,
+    equity REAL NOT NULL,
+    cash REAL NOT NULL,
+    positions_value REAL NOT NULL,
+    FOREIGN KEY (paper_run_id) REFERENCES paper_trading_runs(id)
+);
+
+CREATE TABLE IF NOT EXISTS paper_trading_ticks (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    paper_run_id TEXT NOT NULL,
+    symbol TEXT NOT NULL,
+    ts TEXT NOT NULL,
+    bar_ts TEXT NOT NULL,
+    close_price REAL NOT NULL,
+    signal TEXT NOT NULL,
+    action_taken TEXT NOT NULL,
+    FOREIGN KEY (paper_run_id) REFERENCES paper_trading_runs(id)
+);
+
+CREATE INDEX IF NOT EXISTS idx_paper_trades_run ON paper_trading_trades(paper_run_id, exit_ts);
+CREATE INDEX IF NOT EXISTS idx_paper_equity_run ON paper_trading_equity(paper_run_id, ts);
+CREATE INDEX IF NOT EXISTS idx_paper_ticks_run ON paper_trading_ticks(paper_run_id, ts);
+"""
+
+
+def init_schema(db_path: Path | str) -> None:
+    """Crea (se mancanti) le tabelle paper_trading_* sul db indicato."""
+    Path(db_path).parent.mkdir(parents=True, exist_ok=True)
+    conn = sqlite3.connect(str(db_path))
+    try:
+        conn.executescript(PAPER_SCHEMA_SQL)
+        conn.commit()
+    finally:
+        conn.close()
@@ -0,0 +1,50 @@
+"""Bonus di asset-invariance per la fitness del GA.
+
+corr_signal = frazione di entries su asset A che hanno corrispondenza su asset B
+entro +/-tolerance_bars (default 36 = 3h su 5m TF, vedi paper Pythagoras p. 43).
+"""
+from __future__ import annotations
+
+import os
+
+import pandas as pd
+
+GA_INVARIANCE_ALPHA = float(os.getenv("GA_INVARIANCE_ALPHA", "0.3"))
+GA_INVARIANCE_TOLERANCE_BARS = int(os.getenv("GA_INVARIANCE_TOLERANCE_BARS", "36"))
+
+
+def corr_signal(
+    entries_a: pd.Series,
+    entries_b: pd.Series,
+    tolerance_bars: int = GA_INVARIANCE_TOLERANCE_BARS,
+) -> float:
+    """Frazione di entries A con match in B entro +/-tolerance_bars.
+
+    Args:
+        entries_a, entries_b: Series binarie {0,1} sullo stesso index temporale (interi).
+        tolerance_bars: finestra di tolleranza in barre.
+
+    Returns:
+        In [0, 1]. 0 se entries_a non ha alcuna entry o nessun match.
+    """
+    a_idx = entries_a[entries_a > 0].index.tolist()
+    b_idx = entries_b[entries_b > 0].index.tolist()
+    if not a_idx or not b_idx:
+        return 0.0
+    b_set = set(b_idx)
+    matched = 0
+    for ti in a_idx:
+        for delta in range(-tolerance_bars, tolerance_bars + 1):
+            if (ti + delta) in b_set:
+                matched += 1
+                break
+    return matched / len(a_idx)
+
+
+def apply_invariance_bonus(
+    base_fitness: float,
+    invariance_score: float,
+    alpha: float = GA_INVARIANCE_ALPHA,
+) -> float:
+    """``fitness * (1 + alpha * invariance_score)``."""
+    return base_fitness * (1.0 + alpha * invariance_score)
@@ -0,0 +1,215 @@
+"""Paper-trading data access functions for the strategy_pythagoras dashboard.
+
+Reads exclusively from strategy_pythagoras_paper.db (paper_trading_* tables)
+per il paper-trading; le funzioni dedicate ai winner Pythagoras leggono
+invece dal runs.db del core GA (env ``GA_DB_PATH``), default
+``state/strategy_pythagoras.db`` via env ``STRATEGY_PYTHAGORAS_DB_PATH`` quando
+si vuole usare una sotto-tabella locale.
+"""
+
+from __future__ import annotations
+
+import json
+import sqlite3
+import time
+from pathlib import Path
+from typing import Any
+
+import pandas as pd  # type: ignore[import-untyped]
+
+
+def _paper_conn(db_path: str | Path) -> sqlite3.Connection:
+    # Cold-start race: GUI può avviarsi prima che il paper writer crei il file.
+    db_path_str = str(db_path)
+    deadline = time.monotonic() + 5.0
+    while True:
+        try:
+            conn = sqlite3.connect(db_path_str, timeout=5.0)
+            conn.row_factory = sqlite3.Row
+            return conn
+        except sqlite3.OperationalError:
+            if time.monotonic() >= deadline:
+                raise
+            time.sleep(1.0)
+
+
+def paper_runs_df(db_path: str | Path) -> pd.DataFrame:
+    with _paper_conn(db_path) as conn:
+        rows = conn.execute(
+            "SELECT id, name, started_at, stopped_at, status, initial_capital, config_json "
+            "FROM paper_trading_runs ORDER BY started_at DESC"
+        ).fetchall()
+    return pd.DataFrame([dict(r) for r in rows])
+
+
+def paper_equity_df(db_path: str | Path, run_id: str) -> pd.DataFrame:
+    with _paper_conn(db_path) as conn:
+        rows = conn.execute(
+            "SELECT ts, equity, cash, positions_value FROM paper_trading_equity "
+            "WHERE paper_run_id=? ORDER BY ts ASC",
+            (run_id,),
+        ).fetchall()
+    return pd.DataFrame([dict(r) for r in rows])
+
+
+def paper_positions_df(db_path: str | Path, run_id: str) -> pd.DataFrame:
+    with _paper_conn(db_path) as conn:
+        rows = conn.execute(
+            "SELECT symbol, side, qty, entry_price, entry_ts "
+            "FROM paper_trading_positions WHERE paper_run_id=? ORDER BY symbol",
+            (run_id,),
+        ).fetchall()
+    return pd.DataFrame([dict(r) for r in rows])
+
+
+def paper_trades_df(db_path: str | Path, run_id: str, limit: int = 100) -> pd.DataFrame:
+    with _paper_conn(db_path) as conn:
+        rows = conn.execute(
+            "SELECT symbol, side, qty, entry_price, exit_price, entry_ts, exit_ts, pnl, fees "
+            "FROM paper_trading_trades WHERE paper_run_id=? ORDER BY exit_ts DESC LIMIT ?",
+            (run_id, limit),
+        ).fetchall()
+    return pd.DataFrame([dict(r) for r in rows])
+
+
+def paper_ticks_df(db_path: str | Path, run_id: str, limit: int = 50) -> pd.DataFrame:
+    with _paper_conn(db_path) as conn:
+        rows = conn.execute(
+            "SELECT ts, bar_ts, symbol, close_price, signal, action_taken "
+            "FROM paper_trading_ticks WHERE paper_run_id=? ORDER BY ts DESC LIMIT ?",
+            (run_id, limit),
+        ).fetchall()
+    return pd.DataFrame([dict(r) for r in rows])
+
+
+def paper_run_summary(db_path: str | Path, run_id: str) -> dict[str, Any]:
+    """Aggrega metriche sintetiche per la pagina paper trading."""
+    with _paper_conn(db_path) as conn:
+        run = conn.execute(
+            "SELECT id, name, started_at, stopped_at, status, initial_capital, config_json "
+            "FROM paper_trading_runs WHERE id=?",
+            (run_id,),
+        ).fetchone()
+        if run is None:
+            return {}
+        run = dict(run)
+
+        eq_row = conn.execute(
+            "SELECT equity, cash, positions_value, ts FROM paper_trading_equity "
+            "WHERE paper_run_id=? ORDER BY ts DESC LIMIT 1",
+            (run_id,),
+        ).fetchone()
+
+        trades_agg = conn.execute(
+            "SELECT COUNT(*) AS n, COALESCE(SUM(pnl),0) AS sum_pnl, "
+            "COALESCE(SUM(fees),0) AS sum_fees FROM paper_trading_trades "
+            "WHERE paper_run_id=?",
+            (run_id,),
+        ).fetchone()
+
+        tick_agg = conn.execute(
+            "SELECT COUNT(*) AS n, MAX(ts) AS last_ts FROM paper_trading_ticks "
+            "WHERE paper_run_id=?",
+            (run_id,),
+        ).fetchone()
+
+        positions_n = conn.execute(
+            "SELECT COUNT(*) AS n FROM paper_trading_positions WHERE paper_run_id=?",
+            (run_id,),
+        ).fetchone()["n"]
+
+    initial = float(run["initial_capital"])
+    current_equity = float(eq_row["equity"]) if eq_row is not None else initial
+    pnl_pct = (current_equity - initial) / initial * 100.0 if initial else 0.0
+
+    return {
+        "id": run["id"],
+        "name": run["name"],
+        "status": run["status"],
+        "started_at": run["started_at"],
+        "stopped_at": run["stopped_at"],
+        "initial_capital": initial,
+        "config": json.loads(run["config_json"]),
+        "current_equity": current_equity,
+        "current_cash": float(eq_row["cash"]) if eq_row is not None else initial,
+        "current_positions_value": float(eq_row["positions_value"]) if eq_row is not None else 0.0,
+        "last_equity_ts": eq_row["ts"] if eq_row is not None else None,
+        "pnl_abs": current_equity - initial,
+        "pnl_pct": pnl_pct,
+        "n_trades": int(trades_agg["n"]),
+        "trades_pnl": float(trades_agg["sum_pnl"]),
+        "trades_fees": float(trades_agg["sum_fees"]),
+        "n_ticks": int(tick_agg["n"]),
+        "last_tick_ts": tick_agg["last_ts"],
+        "n_open_positions": int(positions_n),
+    }
+
+
+# ---------------------------------------------------------------------------
+# Pythagoras-specific helpers (winners invariance + candle pattern usage)
+# ---------------------------------------------------------------------------
+
+
+def load_invariance_metrics(ga_db_path: str) -> "pd.DataFrame":
+    """Per ogni winner ritorna (genome_id, cognitive_style, fitness, sharpe_btc, sharpe_eth, invariance_score).
+
+    Lo schema atteso e' la tabella ``pythagoras_winners`` creata dal runner
+    ``scripts/run_pythagoras_smoke.py`` (Task 6.1).
+    """
+    import sqlite3
+
+    import pandas as pd
+
+    con = sqlite3.connect(ga_db_path)
+    try:
+        return pd.read_sql_query(
+            "SELECT genome_id, cognitive_style, fitness, sharpe_btc, sharpe_eth, "
+            "invariance_score FROM pythagoras_winners ORDER BY fitness DESC",
+            con,
+        )
+    finally:
+        con.close()
+
+
+def load_candle_pattern_usage(ga_db_path: str) -> "pd.DataFrame":
+    """Per ogni winner estrae le sequenze candle_pattern usate (per heatmap)."""
+    import json
+    import sqlite3
+
+    import pandas as pd
+
+    con = sqlite3.connect(ga_db_path)
+    try:
+        df = pd.read_sql_query(
+            "SELECT genome_id, cognitive_style, rules_json FROM pythagoras_winners",
+            con,
+        )
+    finally:
+        con.close()
+    records: list[dict] = []
+    for _, row in df.iterrows():
+        rules = json.loads(row["rules_json"]).get("rules", [])
+        for r in rules:
+            for ind_name, params in _walk_indicators(r["condition"]):
+                if ind_name == "candle_pattern":
+                    length = int(params[0])
+                    syms = [int(s) for s in params[1: 1 + length]]
+                    seq_str = "".join({0: "U", 1: "D", 2: "0"}[s] for s in syms)
+                    records.append(
+                        {
+                            "genome_id": row["genome_id"],
+                            "cognitive_style": row["cognitive_style"],
+                            "sequence": seq_str,
+                            "length": length,
+                        }
+                    )
+    return pd.DataFrame.from_records(records)
+
+
+def _walk_indicators(node: dict):
+    """Yields (indicator_name, params) for every IndicatorNode in the AST."""
+    if "op" in node:
+        for a in node.get("args", []):
+            yield from _walk_indicators(a)
+    elif node.get("kind") == "indicator":
+        yield node["name"], node["params"]
@@ -0,0 +1,421 @@
+"""Strategy Pythagoras Dashboard — paper-trading + GA winners page: /.
+
+Avvio: ``uv run python -m strategy_pythagoras.frontend.nicegui_app``
+Default port 8080. Legge il paper DB (``strategy_pythagoras_paper.db``) per il
+tab ``Paper`` e il GA DB (``strategy_pythagoras.db``) per i tab pythagoras-specifici
+(Genomes / Patterns / Ratios / Invariance).
+
+Palette "Neon Trading Dashboard" (ispirata screenshot 2026-05-11):
+- BG: #0A0A0F (near-black con tinge blu)
+- Surface: #13131A (card base)
+- Surface elevata: #1C1C26 (hover/active)
+- Primary pink: #FF2D87 (highlight key metrics, max fitness)
+- Secondary cyan: #00D9FF (median, secondary curves)
+- Accent amber: #FFB800 (warnings, p90)
+- Success neon green: #00E676, Danger neon red: #FF3D60
+- Text: #FFFFFF (primary), #7A7A8C (muted)
+"""
+
+from __future__ import annotations
+
+import os
+from pathlib import Path
+from typing import Any
+
+import pandas as pd  # type: ignore[import-untyped]
+import plotly.graph_objects as go  # type: ignore[import-untyped]
+from nicegui import app, ui
+
+from strategy_pythagoras.frontend.data import (
+    load_candle_pattern_usage,
+    load_invariance_metrics,
+    paper_equity_df,
+    paper_positions_df,
+    paper_run_summary,
+    paper_runs_df,
+    paper_ticks_df,
+    paper_trades_df,
+)
+from multi_swarm_core.dashboard.theme import (
+    COLOR_PRIMARY,
+    COLOR_SURFACE,
+    COLOR_SURFACE_2,
+    COLOR_TEXT,
+    COLOR_TEXT_MUTED,
+    _STATUS_BADGE,
+    _apply_theme,
+    _build_header,
+)
+
+PAPER_DB_PATH = os.environ.get(
+    "STRATEGY_PYTHAGORAS_PAPER_DB_PATH", "./state/strategy_pythagoras_paper.db"
+)
+GA_DB_PATH = os.environ.get(
+    "STRATEGY_PYTHAGORAS_DB_PATH", "./state/strategy_pythagoras.db"
+)
+DASHBOARD_ROOT_PATH = os.environ.get("DASHBOARD_ROOT_PATH", "/strategy_pythagoras_gui")
+REFRESH_INTERVAL_S = 3.0
+
+
+def _paper_runs_options(only_running: bool = False) -> dict[str, str]:
+    try:
+        runs = paper_runs_df(PAPER_DB_PATH)
+    except Exception:
+        return {}
+    if runs.empty:
+        return {}
+    if only_running:
+        runs = runs[runs["status"] == "running"]
+        if runs.empty:
+            return {}
+    return {
+        row["id"]: f"{row['name']} — {row['status']} ({row['started_at'][:16]})"
+        for _, row in runs.iterrows()
+    }
+
+
+def _paper_equity_figure(eq_df: Any, initial_capital: float) -> go.Figure:
+    fig = go.Figure()
+    if eq_df is not None and not eq_df.empty:
+        ts = pd.to_datetime(eq_df["ts"])
+        fig.add_trace(
+            go.Scatter(
+                x=ts,
+                y=eq_df["equity"],
+                mode="lines",
+                line={"color": COLOR_PRIMARY, "width": 2},
+                name="equity",
+            )
+        )
+        fig.add_hline(
+            y=initial_capital,
+            line={"color": COLOR_TEXT_MUTED, "width": 1, "dash": "dash"},
+            annotation_text=f"initial ${initial_capital:.0f}",
+            annotation_position="bottom right",
+            annotation_font_color=COLOR_TEXT_MUTED,
+        )
+    fig.update_layout(
+        title=None,
+        paper_bgcolor=COLOR_SURFACE,
+        plot_bgcolor=COLOR_SURFACE,
+        font={"color": COLOR_TEXT, "family": "Inter"},
+        xaxis={"gridcolor": COLOR_SURFACE_2, "title": None},
+        yaxis={"gridcolor": COLOR_SURFACE_2, "title": "Equity ($)"},
+        margin={"l": 60, "r": 20, "t": 10, "b": 40},
+        height=320,
+        showlegend=False,
+    )
+    return fig
+
+
+def _render_paper_panel() -> None:
+    """Rende il pannello paper-trading (equivalente alla pagina root di strategy_crypto)."""
+    options = _paper_runs_options()
+    if not options:
+        ui.label("Nessuna paper-trading run nel database.").classes("text-h6 text-warning")
+        ui.label(
+            "Avvia un paper run per popolare strategy_pythagoras_paper.db."
+        ).classes("text-caption")
+        return
+
+    state: dict[str, Any] = {"run_id": next(iter(options))}
+
+    with ui.row().classes("w-full items-center gap-4 q-mb-md"):
+        select = ui.select(options=options, value=state["run_id"], label="Paper run").classes(
+            "flex-grow"
+        )
+        status_badge = ui.badge("…", color="primary").classes("text-body1 q-pa-sm")
+        ui.button("🔄 Refresh", on_click=lambda: refresh()).props("outline color=primary")
+
+    with ui.row().classes("w-full gap-4"):
+        with ui.card().classes("flex-grow metric-card accent-cyan"):
+            ui.label("Equity").classes("text-caption")
+            equity_lbl = ui.label("—").classes("text-h4")
+        with ui.card().classes("flex-grow metric-card accent-purple"):
+            ui.label("P/L cumulato").classes("text-caption")
+            pnl_lbl = ui.label("—").classes("text-h4")
+        with ui.card().classes("flex-grow metric-card accent-amber"):
+            ui.label("Trades chiusi").classes("text-caption")
+            trades_lbl = ui.label("—").classes("text-h4")
+        with ui.card().classes("flex-grow metric-card accent-green"):
+            ui.label("Open / Tick").classes("text-caption")
+            ticks_lbl = ui.label("—").classes("text-h4")
+
+    with ui.row().classes("w-full gap-4 q-mt-md"):
+        started_lbl = ui.label("Started: —")
+        last_tick_lbl = ui.label("Last tick: —")
+        cash_lbl = ui.label("Cash: —")
+
+    ui.separator()
+    ui.label("Equity curve").classes("text-subtitle1 q-mt-md")
+    equity_plot = ui.plotly(_paper_equity_figure(None, 0.0)).classes("w-full")
+
+    ui.separator()
+    ui.label("Open positions").classes("text-subtitle1 q-mt-md")
+    positions_table = ui.table(
+        columns=[
+            {"name": "symbol", "label": "symbol", "field": "symbol"},
+            {"name": "side", "label": "side", "field": "side"},
+            {"name": "qty", "label": "qty", "field": "qty"},
+            {"name": "entry_price", "label": "entry", "field": "entry_price"},
+            {"name": "entry_ts", "label": "entry ts", "field": "entry_ts"},
+        ],
+        rows=[],
+        row_key="symbol",
+    ).classes("w-full")
+
+    ui.separator()
+    ui.label("Ultimi 30 tick").classes("text-subtitle1 q-mt-md")
+    ticks_table = ui.table(
+        columns=[
+            {"name": "ts", "label": "ts", "field": "ts"},
+            {"name": "symbol", "label": "symbol", "field": "symbol"},
+            {"name": "bar_ts", "label": "bar", "field": "bar_ts"},
+            {"name": "close_price", "label": "close", "field": "close_price"},
+            {"name": "signal", "label": "signal", "field": "signal"},
+            {"name": "action_taken", "label": "action", "field": "action_taken"},
+        ],
+        rows=[],
+        row_key="ts",
+    ).classes("w-full")
+
+    ui.separator()
+    ui.label("Trades chiusi (ultimi 50)").classes("text-subtitle1 q-mt-md")
+    trades_table = ui.table(
+        columns=[
+            {"name": "exit_ts", "label": "exit ts", "field": "exit_ts"},
+            {"name": "symbol", "label": "symbol", "field": "symbol"},
+            {"name": "side", "label": "side", "field": "side"},
+            {"name": "qty", "label": "qty", "field": "qty"},
+            {"name": "entry_price", "label": "entry", "field": "entry_price"},
+            {"name": "exit_price", "label": "exit", "field": "exit_price"},
+            {"name": "pnl", "label": "pnl", "field": "pnl"},
+            {"name": "fees", "label": "fees", "field": "fees"},
+        ],
+        rows=[],
+        row_key="exit_ts",
+    ).classes("w-full")
+
+    def refresh() -> None:
+        run_id = select.value
+        if not run_id:
+            return
+        try:
+            summary = paper_run_summary(PAPER_DB_PATH, run_id)
+            eq = paper_equity_df(PAPER_DB_PATH, run_id)
+            positions = paper_positions_df(PAPER_DB_PATH, run_id)
+            ticks = paper_ticks_df(PAPER_DB_PATH, run_id, limit=30)
+            trades = paper_trades_df(PAPER_DB_PATH, run_id, limit=50)
+        except Exception as e:  # noqa: BLE001
+            ui.notify(f"Errore: {e}", type="negative")
+            return
+
+        text, color = _STATUS_BADGE.get(summary["status"], (summary["status"], "primary"))
+        status_badge.text = text
+        status_badge.props(f"color={color}")
+
+        equity_lbl.text = f"${summary['current_equity']:.2f}"
+        pnl_lbl.text = f"{summary['pnl_pct']:+.2f}%"
+        trades_lbl.text = str(summary["n_trades"])
+        ticks_lbl.text = f"{summary['n_open_positions']} / {summary['n_ticks']}"
+
+        started_lbl.text = f"Started: {summary['started_at']}"
+        last_tick_lbl.text = f"Last tick: {summary['last_tick_ts'] or '—'}"
+        cash_lbl.text = (
+            f"Cash: ${summary['current_cash']:.2f}  |  "
+            f"Pos value: ${summary['current_positions_value']:.2f}"
+        )
+
+        equity_plot.update_figure(_paper_equity_figure(eq, summary["initial_capital"]))
+
+        positions_table.rows = (
+            [
+                {col: (round(v, 6) if isinstance(v, float) else v) for col, v in row.items()}
+                for _, row in positions.iterrows()
+            ]
+            if not positions.empty
+            else []
+        )
+        positions_table.update()
+
+        ticks_table.rows = (
+            [
+                {col: (round(v, 6) if isinstance(v, float) else v) for col, v in row.items()}
+                for _, row in ticks.iterrows()
+            ]
+            if not ticks.empty
+            else []
+        )
+        ticks_table.update()
+
+        trades_table.rows = (
+            [
+                {col: (round(v, 6) if isinstance(v, float) else v) for col, v in row.items()}
+                for _, row in trades.iterrows()
+            ]
+            if not trades.empty
+            else []
+        )
+        trades_table.update()
+
+    def on_select_change() -> None:
+        state["run_id"] = select.value
+        refresh()
+
+    select.on_value_change(on_select_change)
+    ui.timer(REFRESH_INTERVAL_S, refresh)
+    refresh()
+
+
+@ui.page("/")
+def index() -> None:
+    _apply_theme()
+    _build_header(
+        active="/",
+        brand_subtitle="Strategy Pythagoras",
+        nav_items=[("/", "Dashboard")],
+        db_label=f"⛁ {Path(GA_DB_PATH).resolve().name}",
+    )
+
+    with ui.tabs() as tabs:
+        t_genomes = ui.tab("Genomes")
+        t_patterns = ui.tab("Patterns")
+        t_ratios = ui.tab("Ratios")
+        t_invariance = ui.tab("Invariance")
+        t_paper = ui.tab("Paper")
+
+    with ui.tab_panels(tabs, value=t_genomes).classes("w-full"):
+        with ui.tab_panel(t_genomes):
+            ui.label("GA Winners (post pythagoras-smoke-001)").classes("text-h6")
+            try:
+                df = load_invariance_metrics(GA_DB_PATH)
+                if df.empty:
+                    ui.label(
+                        "No winners yet. Run scripts/run_pythagoras_smoke.py first."
+                    ).classes("text-warning")
+                else:
+                    ui.table(
+                        rows=df.to_dict("records"),
+                        columns=[
+                            {"name": "genome_id", "label": "Genome ID", "field": "genome_id"},
+                            {
+                                "name": "cognitive_style",
+                                "label": "Style",
+                                "field": "cognitive_style",
+                            },
+                            {"name": "fitness", "label": "Fitness", "field": "fitness"},
+                            {"name": "sharpe_btc", "label": "Sharpe BTC", "field": "sharpe_btc"},
+                            {"name": "sharpe_eth", "label": "Sharpe ETH", "field": "sharpe_eth"},
+                            {
+                                "name": "invariance_score",
+                                "label": "Invariance",
+                                "field": "invariance_score",
+                            },
+                        ],
+                        pagination=10,
+                    ).classes("w-full")
+            except Exception as e:  # noqa: BLE001
+                ui.label(f"DB not ready: {e}").classes("text-warning")
+
+        with ui.tab_panel(t_patterns):
+            ui.label(
+                "Candle pattern sequences emerged (per cognitive style)"
+            ).classes("text-h6")
+            try:
+                df_pat = load_candle_pattern_usage(GA_DB_PATH)
+                if df_pat.empty:
+                    ui.label(
+                        "No patterns yet. Run scripts/run_pythagoras_smoke.py first."
+                    ).classes("text-warning")
+                else:
+                    grouped = (
+                        df_pat.groupby(["cognitive_style", "sequence"])
+                        .size()
+                        .reset_index(name="count")
+                    )
+                    grouped = grouped.sort_values("count", ascending=False).head(50)
+                    ui.table(
+                        rows=grouped.to_dict("records"),
+                        columns=[
+                            {
+                                "name": "cognitive_style",
+                                "label": "Style",
+                                "field": "cognitive_style",
+                            },
+                            {
+                                "name": "sequence",
+                                "label": "Sequence (U/D/0)",
+                                "field": "sequence",
+                            },
+                            {"name": "count", "label": "Count", "field": "count"},
+                        ],
+                        pagination=15,
+                    ).classes("w-full")
+            except Exception as e:  # noqa: BLE001
+                ui.label(f"DB not ready: {e}").classes("text-warning")
+
+        with ui.tab_panel(t_ratios):
+            ui.label(
+                "Pythagorean ratio literals — distance from universal constants"
+            ).classes("text-h6")
+            try:
+                df = load_invariance_metrics(GA_DB_PATH)
+                if df.empty:
+                    ui.label("No winners yet.").classes("text-warning")
+                else:
+                    ui.label(f"Total winners: {len(df)}").classes("text-body2")
+                    ui.label(
+                        "(Ratio literal histogram available after GA run produces "
+                        "pythagorean_ratio entries.)"
+                    ).classes("text-caption")
+            except Exception as e:  # noqa: BLE001
+                ui.label(f"DB not ready: {e}").classes("text-warning")
+
+        with ui.tab_panel(t_invariance):
+            ui.label(
+                "Cross-asset invariance: Sharpe BTC vs Sharpe ETH"
+            ).classes("text-h6")
+            try:
+                df = load_invariance_metrics(GA_DB_PATH)
+                if df.empty:
+                    ui.label("No winners yet.").classes("text-warning")
+                else:
+                    import plotly.express as px  # type: ignore[import-untyped]
+
+                    fig = px.scatter(
+                        df,
+                        x="sharpe_btc",
+                        y="sharpe_eth",
+                        color="invariance_score",
+                        hover_data=["genome_id", "cognitive_style", "fitness"],
+                        title="Sharpe BTC vs Sharpe ETH (color = invariance_score)",
+                    )
+                    ui.plotly(fig).classes("w-full")
+            except Exception as e:  # noqa: BLE001
+                ui.label(f"DB not ready: {e}").classes("text-warning")
+
+        with ui.tab_panel(t_paper):
+            _render_paper_panel()
+
+
+def main() -> None:
+    app.on_startup(
+        lambda: print(
+            f"GA DB: {Path(GA_DB_PATH).resolve()} | "
+            f"Paper DB: {Path(PAPER_DB_PATH).resolve()} | "
+            f"root_path: {DASHBOARD_ROOT_PATH or '/'}"
+        )
+    )
+    ui.run(
+        host="0.0.0.0",
+        port=int(os.environ.get("SWARM_DASHBOARD_PORT", "8080")),
+        title="Strategy Pythagoras Dashboard",
+        reload=False,
+        show=False,
+        dark=True,
+        root_path=DASHBOARD_ROOT_PATH,
+    )
+
+
+if __name__ in {"__main__", "__mp_main__"}:
+    main()
@@ -0,0 +1,70 @@
+"""Indicatori candle Pythagoras.
+
+Vincoli grammar: ``IndicatorNode.params`` e' sempre ``list[float]``. Quindi:
+- candle_pattern: params = [length, sym0, sym1, ..., sym_{length-1}]
+  length in [3,12]; sym in {0=U, 1=D, 2=doji}
+- pythagorean_ratio: params = [lookback]  lookback in [12,200]
+- fractal_mirror: params = [k, axis_int]  k in [3,12]; axis_int=0(h) 1(v)
+"""
+from __future__ import annotations
+
+import numpy as np
+import pandas as pd
+
+_DOJI_THRESHOLD = 0.001
+
+
+def _symbol_series(df: pd.DataFrame) -> pd.Series:
+    """Mappa ogni candela in {0=U, 1=D, 2=doji}."""
+    close = df["close"]
+    open_ = df["open"]
+    rel = (close - open_).abs() / open_.replace(0, np.nan)
+    sym = np.where(close > open_, 0, np.where(close < open_, 1, 2))
+    sym = np.where(rel.values < _DOJI_THRESHOLD, 2, sym)
+    return pd.Series(sym, index=df.index, dtype="int64")
+
+
+def candle_pattern(df: pd.DataFrame, params: list[float]) -> pd.Series:
+    """1.0 se le ultime ``length`` candele matchano la sequenza, 0.0 altrimenti."""
+    length = int(params[0])
+    target = [int(s) for s in params[1:1 + length]]
+    syms = _symbol_series(df)
+    out = pd.Series(0.0, index=df.index, dtype="float64")
+    if len(syms) < length:
+        return out
+    arr = syms.values
+    target_arr = np.array(target, dtype=arr.dtype)
+    for i in range(length - 1, len(arr)):
+        if np.array_equal(arr[i - length + 1: i + 1], target_arr):
+            out.iat[i] = 1.0
+    return out
+
+
+def pythagorean_ratio(df: pd.DataFrame, params: list[float]) -> pd.Series:
+    """``max(close[-lookback:]) / min(close[-lookback:])`` rolling."""
+    lookback = int(params[0])
+    close = df["close"]
+    hi = close.rolling(lookback, min_periods=1).max()
+    lo = close.rolling(lookback, min_periods=1).min().replace(0, np.nan)
+    return (hi / lo).fillna(1.0)
+
+
+def fractal_mirror(df: pd.DataFrame, params: list[float]) -> pd.Series:
+    """Correlation tra close[-k:] e suo mirror su asse axis."""
+    k = int(params[0])
+    axis_int = int(params[1])
+    close = df["close"].values
+    out = np.full(len(close), 0.0)
+    for i in range(k - 1, len(close)):
+        window = close[i - k + 1: i + 1]
+        if axis_int == 0:  # h: mirror temporale
+            mirror = window[::-1]
+        else:              # v: mirror prezzo
+            mirror = window.max() - (window - window.min())
+        std_w = window.std()
+        std_m = mirror.std()
+        if std_w < 1e-12 or std_m < 1e-12:
+            out[i] = 0.0
+        else:
+            out[i] = float(np.corrcoef(window, mirror)[0, 1])
+    return pd.Series(out, index=df.index, dtype="float64")
@@ -0,0 +1,53 @@
+{
+  "_comment": "Stili cognitivi del GA per strategy_pythagoras. Dominio: pattern frattali ricorrenti su mercati crypto secondo il framework Pythagoras-Malanga (Fourier + frattali H-C + Evideon).",
+  "_schema": "3.2",
+  "_changelog": "v1.0 - Initial release. Schema clonato da strategy_crypto v3.2 con contenuto Pythagoras-aligned.",
+  "_focus_metrics_design": "Le focus_metrics sono ENFASI per la lente, non filtri. 4 per stile. Devono includere almeno 1 dei 3 indicatori Pythagoras (candle_pattern, pythagorean_ratio, fractal_mirror).",
+  "_design_invariants": "(1) ASCII-safe; (2) Archetipo dominante: <metafora> come ancora; (3) Lookback range esplicito; (4) Prima frase 'Il mercato e ...'; (5) Lunghezza directive 800-950 char.",
+  "_param_encoding_note": "Promosso a campo top-level 'custom_indicators_spec' (v3.2). Vedi sotto.",
+
+  "custom_indicators_spec": "Indicatori Pythagoras (oltre a sma/sma_pct/rsi/atr/atr_pct/realized_vol/macd/macd_pct gia documentati sopra). REGOLA CRITICA: params accetta SOLO numeri float, MAI stringhe e MAI altri nodi.\n\n  {\"kind\": \"indicator\", \"name\": \"candle_pattern\", \"params\": [length, sym0, sym1, ..., sym_{length-1}]}\n    length: int in [3,12]  (numero di candele consecutive che devono matchare)\n    sym_i:  int in {0, 1, 2}  (0=U up close>open, 1=D down close<open, 2=doji)\n    Esempio: pattern U-D-U di 3 candele = params=[3, 0, 1, 0]\n    Esempio: pattern D-D-D-U (reversal classico) = params=[4, 1, 1, 1, 0]\n    Output: 1.0 se le ultime length candele matchano, 0.0 altrimenti. Confronta solo con literal 0.0 o 1.0.\n\n  {\"kind\": \"indicator\", \"name\": \"pythagorean_ratio\", \"params\": [lookback]}\n    lookback: int in [12,200]  (finestra rolling per max/min close)\n    Esempio: ratio su finestra 89 (Fibonacci) = params=[89]\n    Output: max(close[-lookback:]) / min(close[-lookback:]), adimensionale >= 1.0.\n    Confronta con literal vicini a phi=1.618, 1/phi=0.618 (in realta usa 1.618 perche ratio >= 1), sqrt2=1.414, pi/2=1.571, e=2.718.\n\n  {\"kind\": \"indicator\", \"name\": \"fractal_mirror\", \"params\": [k, axis_int]}\n    k: int in [3,12]  (lunghezza finestra di confronto)\n    axis_int: int in {0, 1}  (0=h mirror temporale, 1=v mirror prezzo)\n    Esempio: mirror temporale su 8 candele = params=[8, 0]\n    Esempio: mirror prezzo su 6 candele = params=[6, 1]\n    Output: correlation di Pearson in [-1.0, 1.0] tra finestra e suo mirror. Confronta con literal in (-1, 1), tipicamente |0.5|-|0.8|.\n\nESEMPI di confronti CORRETTI:\n  candle_pattern di 3 candele U-D-U attiva: {\"op\": \"eq\", \"args\": [{\"kind\":\"indicator\",\"name\":\"candle_pattern\",\"params\":[3,0,1,0]}, {\"kind\":\"literal\",\"value\":1.0}]}\n  ratio phi entro 0.5%: {\"op\": \"and\", \"args\": [{\"op\":\"gt\",\"args\":[{\"kind\":\"indicator\",\"name\":\"pythagorean_ratio\",\"params\":[89]},{\"kind\":\"literal\",\"value\":1.610}]}, {\"op\":\"lt\",\"args\":[{\"kind\":\"indicator\",\"name\":\"pythagorean_ratio\",\"params\":[89]},{\"kind\":\"literal\",\"value\":1.626}]}]}\n  mirror temporale forte: {\"op\": \"gt\", \"args\": [{\"kind\":\"indicator\",\"name\":\"fractal_mirror\",\"params\":[8,0]}, {\"kind\":\"literal\",\"value\":0.7}]}\n\nESEMPI ERRATI (dead branch o validation error):\n  candle_pattern con stringa: params=[\"UDU\"]                  (errato: params devono essere numeri)\n  candle_pattern: params=[3, \"U\", \"D\", \"U\"]                    (errato: usa codici 0/1/2)\n  pythagorean_ratio: params=[\"close\"]                          (errato: params devono essere numeri, non nomi feature)\n  pythagorean_ratio: params=[5]                                (errato: lookback < 12)\n  pythagorean_ratio: params=[89, 1.618]                        (errato: arity 1, non 2)\n  fractal_mirror: params=[0, 1]                                (errato: k < 3)\n  fractal_mirror: params=[100, 0]                              (errato: k > 12)\n  fractal_mirror: params=[8]                                   (errato: arity 2, manca axis_int)",
+
+  "agent_role": "Sei un agente generatore di ipotesi di trading quantitativo per un sistema swarm coevolutivo che cerca pattern frattali ricorrenti sui mercati crypto secondo il framework Pythagoras-Malanga (frattali H-C, trasformata di Fourier, geometria Evideon). Sei parte di una popolazione che esplora collettivamente lo spazio dei pattern: la diversita delle ipotesi e un asset critico. Preferisci esplorare territori meno ovvi rispetto a quelli che la tua lente cognitiva renderebbe predicibili. La strategia che produci deve essere riconoscibile come emanata dal tuo stile.",
+
+  "pattern_guidance": "Forme da cercare:\n  - Sequenze candle 3-12 lunghezza in alfabeto {U,D,doji} via candle_pattern\n  - Ratios di prezzo vicini a 1.618 (phi), 0.618 (1/phi), 1.414 (sqrt2), 1.571 (pi/2), 2.718 (e) via pythagorean_ratio\n  - Mirror H/V come proiezioni via fractal_mirror\n  - Pattern composti: lunghi come concatenazione di corti (Pattern 13 = Pattern 3 + Pattern 11 nel paper)\n  - Cicli ricorrenti via Poincare: stesso pattern firato a distanze regolari\nCerca pattern che si REPLICANO su BTC e ETH al ~stesso timestamp (entro 36 barre = 3h su 5m TF).",
+
+  "instruction": "Genera una strategia che cerchi pattern frattali ricorrenti coerenti col framework Pythagoras-Malanga, riconoscibile come emanata dal tuo stile.",
+
+  "domain_warnings": "Crypto trada 24/7 senza CME gap. I numeri sacri (Solfeggio 396-417-528-639-741-852 Hz, 137.0359, 121.449, 161.86) sono prior teorici, NON leggi: usali come scale candidate per literal, non come dogma. Il paper Pythagoras e' esplicitamente non-falsificabile (cita 'consapevolezza del trader' come jolly): il backtest e' l'unico arbitro. La tolleranza +/-3h del paper su 5m TF = +/-36 barre per il bonus di asset-invariance.",
+
+  "anti_patterns": "Evita: (1) sequenza candle_pattern con piu di 7 simboli vincolati (overfitting); (2) pythagorean_ratio con tolleranza > 2% sui literal (numerologia spuria); (3) fractal_mirror con k = lookback_window (tautologico, sempre prossimo a 1.0); (4) letterali con piu di 4 decimali (es. 1.6180339 -> usa 1.618); (5) piu di 4 condizioni in AND; (6) crossover tra indicatori dello stesso tipo con lookback vicini (chattering); (7) time_in_market > 80% (leveraged buy&hold camuffato).",
+
+  "output_priorities": "Trade-off: (1) coerenza con la lente cognitiva: la strategia deve essere riconoscibile come emanata dal tuo stile (es. pythagorean usa ratios, candle_grammarian sequenze esplicite); (2) asset-invariance: segnali che attivano sia su BTC che su ETH entro 36 barre (e' il bonus della fitness); (3) selettivita: poche entry forti, alto SNR; (4) composizionalita: pattern lunghi come somma di corti; (5) robustezza vs random baseline (3-sigma richiesto dal skeptic_quant).",
+
+  "styles": {
+    "pythagorean": {
+      "directive": "Il mercato e un'armonia di ratios sacri: la sezione aurea (phi=1.618 e 1/phi=0.618), la radice di due (1.414), pi mezzi (1.571) ed e mezzi (1.359) strutturano i livelli di supporto, resistenza e target. Usa pythagorean_ratio con lookback Fibonacci (89, 144) per misurare l'estensione del range recente e cerca punti dove esso e' prossimo a phi entro 0.5%: indica un swing maturo che tende all'inversione armonica. Combina con candle_pattern di 3-5 candele per timing. Evita di chiudere troppo presto: la potenza di un ratio sacro si manifesta quando il prezzo passa il livello e poi torna a testarlo. Mai usare letterali con piu di 4 decimali: l'armonia non e' nei decimali ma nella struttura. Lookback consigliato: 89-144. Archetipo dominante: musicista che ascolta gli accordi nascosti del mercato.",
+      "focus_metrics": ["pythagorean_ratio", "candle_pattern", "sma_pct", "realized_vol"]
+    },
+    "fractal_geometer": {
+      "directive": "Il mercato e autosimilare: una sequenza di 3 candele si ripete dilatata su 6, 12 candele e oltre. La tua firma e' detectare la stessa struttura a scale diverse. Usa candle_pattern di lunghezza 3 per identificare l'unita atomica, poi cerca conferma con fractal_mirror su finestra 2x o 3x: se la correlation e' alta significa che la scala maggiore replica la struttura. Pythagorean_ratio con lookback grande (>100) misura l'ampiezza globale del frattale e i suoi target di proiezione. Atr_pct conferma che la volatilita corrente e coerente con la scala. Evita pattern di lunghezza > 6: i frattali complessi vanno composti, non vincolati esplicitamente. Mai usare fractal_mirror con k uguale al lookback_window. Lookback consigliato: 48-144. Archetipo dominante: geometra che misura la dimensione di Hausdorff del prezzo.",
+      "focus_metrics": ["candle_pattern", "fractal_mirror", "atr_pct", "pythagorean_ratio"]
+    },
+    "fourier_analyst": {
+      "directive": "Il mercato e somma di seni: una funzione complessa scomposta in frequenze armoniche distinte. Il tuo compito e' isolare le componenti ricorrenti dominanti. Usa sma_pct su 3 lookback diversi (es. 30, 60, 120) come proxy delle componenti a bassa frequenza; quando 2 su 3 si allineano nella stessa direzione, la frequenza dominante e' chiara e attiva l'entry. Realized_vol misura l'energia totale del segnale; un calo improvviso indica trasformazione di fase. Candle_pattern di 4-6 candele identifica l'inviluppo locale del segnale composito. Cerca pattern dove il presente e' la proiezione del passato dopo una traslazione temporale costante (mappa di Poincare). Evita oscillatori dello stesso tipo a lookback vicini: producono chattering inutile. Lookback consigliato: 60-200. Archetipo dominante: ingegnere del segnale che ricostruisce la portante armonica nascosta.",
+      "focus_metrics": ["sma_pct", "realized_vol", "candle_pattern", "atr"]
+    },
+    "evideonic_projector": {
+      "directive": "Il mercato e il presente proiettato dal passato via due operazioni geometriche: mirror H (inversione temporale) e mirror V (inversione prezzo), poi scalato per phi, 1/phi, sqrt2. Il tuo segnale principale e' fractal_mirror su entrambi gli assi: quando l'h mirror su 6-12 candele e' altamente positivo (>0.7), il pattern recente e' un riflesso temporale di un pattern precedente, segnale di prosecuzione armonica; quando il v mirror e' positivo, e' un riflesso direzionale, segnale di inversione imminente. Usa pythagorean_ratio per fissare i target di proiezione (entry moltiplicata per phi). Candle_pattern serve da gate per filtrare il contesto. Evita target con ratio > 3 (zona di rumore, fuori scala phi). Lookback consigliato: 24-96. Archetipo dominante: proiettore evideonico che riflette ogni evento nei suoi quattro alter-ego speculari.",
+      "focus_metrics": ["fractal_mirror", "pythagorean_ratio", "candle_pattern", "sma_pct"]
+    },
+    "candle_grammarian": {
+      "directive": "Il mercato e una lingua di tre simboli atomici: U (close>open), D (close<open), doji (close circa open). Ogni parola di 3-12 lettere e' una frase compiuta con significato direzionale. Il tuo dominio e' la sintassi diretta: usa candle_pattern come segnale principale, con sequenze di lunghezza 3-5 per la decisione di entry. Cerca pattern dove l'ultima candela e' U dopo una serie 'DD' o '0D' (reversal classico) o sequenze 'UU0U' che indicano accumulazione con pausa. Volume conferma la pronuncia del pattern: alto volume = pattern ben articolato. Atr modula la dimensione di stop e target. Combina al massimo due candle_pattern indipendenti in AND, mai di piu. Mai pattern di lunghezza > 7 (frasi troppo lunghe = overfitting). Lookback consigliato: 12-48. Archetipo dominante: grammatico che decifra la sintassi delle candele.",
+      "focus_metrics": ["candle_pattern", "volume", "atr", "realized_vol"]
+    },
+    "recurrence_theorist": {
+      "directive": "Il mercato e regolato dal teorema di Poincare: ogni configurazione torna prima o poi vicino a se stessa nel tempo. Il tuo lavoro e' trovare il pattern di oggi che firo' anche ieri o la settimana scorsa a distanza k regolare. Usa candle_pattern su 5-8 candele come firma riconoscibile, poi fractal_mirror con axis=0 (h, mirror temporale) su lookback grande (100-200) per misurare quanto la struttura corrente assomigli a quella passata invertita: se >0.5, il ciclo e' nella fase di ripetizione attiva. Pythagorean_ratio fornisce i target di estensione del ciclo. Sma_pct su lookback medio conferma il regime macro coerente. Evita pattern troppo corti (3 candele): troppo poco distinti per ricorrenza affidabile. Lookback consigliato: 100-200. Archetipo dominante: archeologo del prezzo che cerca repliche di antichi pattern dimenticati.",
+      "focus_metrics": ["candle_pattern", "fractal_mirror", "pythagorean_ratio", "sma_pct"]
+    },
+    "skeptic_quant": {
+      "directive": "Il mercato e rumore con piccoli edge statistici nascosti tra il caos. Tu sei l'anticorpo del sistema: non fidarti mai dei pattern senza significativita 3-sigma vs random baseline. Usa realized_vol e atr_pct come gate di regime (entra solo in regimi di volatilita media, percentile 30-70). Sma_pct conferma la direzione del trend macro corrente. Candle_pattern di 3-4 candele puo' essere usato, ma solo come conferma di un setup gia validato da volatilita e trend: mai come segnale primario isolato. Vincolo duro: max 3 condizioni in AND (poche e robuste); literal con almeno il 20% di margine di sicurezza tra entry ed exit. Nessun pythagorean_ratio (numerologia spuria) ne fractal_mirror (tautologici nella tua lente scettica). Lookback: 60-150. Archetipo dominante: scettico quantitativo che falsifica prima di confermare.",
+      "focus_metrics": ["realized_vol", "atr_pct", "sma_pct", "candle_pattern"]
+    }
+  }
+}
@@ -0,0 +1,43 @@
+"""Bonus invariance: pattern che firano simultaneamente su 2 asset entro tolleranza."""
+from __future__ import annotations
+
+import pandas as pd
+import pytest
+
+from strategy_pythagoras.fitness_invariance import (
+    apply_invariance_bonus,
+    corr_signal,
+)
+
+
+def test_corr_signal_perfect_alignment() -> None:
+    entries_btc = pd.Series([0, 1, 0, 1, 0], index=[0, 1, 2, 3, 4])
+    entries_eth = pd.Series([0, 1, 0, 1, 0], index=[0, 1, 2, 3, 4])
+    assert corr_signal(entries_btc, entries_eth, tolerance_bars=0) == pytest.approx(1.0)
+
+
+def test_corr_signal_no_overlap() -> None:
+    entries_btc = pd.Series([0, 1, 0, 0, 0], index=[0, 1, 2, 3, 4])
+    entries_eth = pd.Series([0, 0, 0, 0, 1], index=[0, 1, 2, 3, 4])
+    assert corr_signal(entries_btc, entries_eth, tolerance_bars=0) == pytest.approx(0.0)
+
+
+def test_corr_signal_within_tolerance() -> None:
+    # entry su BTC a t=1, su ETH a t=3, tolerance=2 -> match
+    entries_btc = pd.Series([0, 1, 0, 0, 0], index=[0, 1, 2, 3, 4])
+    entries_eth = pd.Series([0, 0, 0, 1, 0], index=[0, 1, 2, 3, 4])
+    assert corr_signal(entries_btc, entries_eth, tolerance_bars=2) == pytest.approx(1.0)
+
+
+def test_apply_invariance_bonus_increases_fitness() -> None:
+    assert apply_invariance_bonus(1.0, 0.5, 0.3) == pytest.approx(1.0 * (1.0 + 0.3 * 0.5))
+
+
+def test_apply_invariance_bonus_alpha_zero() -> None:
+    assert apply_invariance_bonus(1.0, 0.7, 0.0) == pytest.approx(1.0)
+
+
+def test_corr_signal_zero_entries() -> None:
+    entries_btc = pd.Series([0, 0, 0], index=[0, 1, 2])
+    entries_eth = pd.Series([0, 0, 0], index=[0, 1, 2])
+    assert corr_signal(entries_btc, entries_eth, tolerance_bars=0) == 0.0
@@ -0,0 +1,82 @@
+"""Unit-test dei 3 indicatori Pythagoras."""
+from __future__ import annotations
+
+import numpy as np
+import pandas as pd
+import pytest
+
+from strategy_pythagoras.indicators import (
+    candle_pattern,
+    fractal_mirror,
+    pythagorean_ratio,
+)
+
+
+@pytest.fixture
+def ohlcv_30() -> pd.DataFrame:
+    """30 candele sintetiche: pattern alternato U,D,U,D,..."""
+    n = 30
+    close = np.array([100.0 + i for i in range(n)])  # monotone
+    open_ = close - np.tile([1.0, -1.0], n // 2)     # U,D,U,D,...
+    return pd.DataFrame({"open": open_, "high": close + 0.5, "low": open_ - 0.5, "close": close, "volume": [1.0] * n})
+
+
+# -- candle_pattern -----------------------------------------------------------
+
+def test_candle_pattern_matches_recent(ohlcv_30: pd.DataFrame) -> None:
+    # Verifica che il symbol mapping U=0,D=1,doji=2 lavori correttamente.
+    # Con la fixture: indices pari sono U (close>open), dispari sono D (close<open).
+    # ultime 3 candele (idx 27,28,29) -> D,U,D
+    out = candle_pattern(ohlcv_30, [3, 1, 0, 1])  # D, U, D
+    assert out.iloc[-1] == 1.0
+    out2 = candle_pattern(ohlcv_30, [3, 0, 0, 0])  # U, U, U: no match
+    assert out2.iloc[-1] == 0.0
+
+
+def test_candle_pattern_zero_for_short_history(ohlcv_30: pd.DataFrame) -> None:
+    out = candle_pattern(ohlcv_30, [3, 0, 0, 0])
+    assert out.iloc[0] == 0.0
+    assert out.iloc[1] == 0.0
+
+
+def test_candle_pattern_doji_symbol(ohlcv_30: pd.DataFrame) -> None:
+    df = ohlcv_30.copy()
+    # forza la candela [-1] doji: |close-open|/open < 0.001
+    df.loc[df.index[-1], "open"] = df["close"].iloc[-1] * (1 - 1e-5)
+    # ultime 3 dopo modifica: D (idx 27), U (idx 28), doji (idx 29)
+    out = candle_pattern(df, [3, 1, 0, 2])
+    assert out.iloc[-1] == 1.0
+
+
+# -- pythagorean_ratio --------------------------------------------------------
+
+def test_pythagorean_ratio_basic(ohlcv_30: pd.DataFrame) -> None:
+    out = pythagorean_ratio(ohlcv_30, [12])
+    # ultimi 12 close: 118..129 -> max/min = 129/118
+    expected = 129.0 / 118.0
+    assert abs(out.iloc[-1] - expected) < 1e-9
+
+
+def test_pythagorean_ratio_no_lookahead(ohlcv_30: pd.DataFrame) -> None:
+    out = pythagorean_ratio(ohlcv_30, [12])
+    out0 = pythagorean_ratio(ohlcv_30.iloc[:13], [12])
+    assert abs(out.iloc[12] - out0.iloc[-1]) < 1e-9
+
+
+# -- fractal_mirror -----------------------------------------------------------
+
+def test_fractal_mirror_h_pattern_inverted(ohlcv_30: pd.DataFrame) -> None:
+    # mirror temporale: correlazione tra close[-k:] e close[-k:][::-1]
+    # Per close monotono, correlation(seq, seq_reversed) = -1
+    out = fractal_mirror(ohlcv_30, [6, 0])  # axis=0 (h)
+    assert out.iloc[-1] < -0.99
+
+
+def test_fractal_mirror_v_axis(ohlcv_30: pd.DataFrame) -> None:
+    out = fractal_mirror(ohlcv_30, [6, 1])
+    assert out.iloc[-1] < -0.99
+
+
+def test_fractal_mirror_clamps_initial(ohlcv_30: pd.DataFrame) -> None:
+    out = fractal_mirror(ohlcv_30, [6, 0])
+    assert len(out) == len(ohlcv_30)
@@ -0,0 +1,67 @@
+"""prompts.json valido, 7 stili Pythagoras-aligned, schema v3.2."""
+from __future__ import annotations
+
+import json
+from importlib.resources import files
+
+import pytest
+
+EXPECTED_STYLES = {
+    "pythagorean", "fractal_geometer", "fourier_analyst",
+    "evideonic_projector", "candle_grammarian", "recurrence_theorist",
+    "skeptic_quant",
+}
+
+
+@pytest.fixture
+def prompts() -> dict:
+    p = files("strategy_pythagoras") / "prompts.json"
+    return json.loads(p.read_text(encoding="utf-8"))
+
+
+def test_schema_version(prompts: dict) -> None:
+    assert prompts["_schema"] == "3.2"
+
+
+def test_top_level_fields(prompts: dict) -> None:
+    for k in (
+        "agent_role", "pattern_guidance", "instruction", "domain_warnings",
+        "anti_patterns", "output_priorities", "styles",
+    ):
+        assert k in prompts, f"missing field: {k}"
+
+
+def test_styles_set(prompts: dict) -> None:
+    assert set(prompts["styles"].keys()) == EXPECTED_STYLES
+
+
+def test_styles_have_directive_and_focus_metrics(prompts: dict) -> None:
+    for name, style in prompts["styles"].items():
+        assert "directive" in style, f"{name} missing directive"
+        assert "focus_metrics" in style, f"{name} missing focus_metrics"
+        assert len(style["focus_metrics"]) == 4, f"{name} focus_metrics len != 4"
+
+
+def test_directives_ascii_safe(prompts: dict) -> None:
+    for name, style in prompts["styles"].items():
+        directive = style["directive"]
+        for ch in directive:
+            assert ord(ch) <= 0x7F, f"{name} directive contains non-ASCII char {ch!r}"
+
+
+def test_directives_length(prompts: dict) -> None:
+    for name, style in prompts["styles"].items():
+        n = len(style["directive"])
+        assert 800 <= n <= 950, f"{name} directive length {n} outside [800,950]"
+
+
+def test_directives_first_phrase(prompts: dict) -> None:
+    for name, style in prompts["styles"].items():
+        assert style["directive"].startswith("Il mercato e "), \
+            f"{name} directive must start with 'Il mercato e '"
+
+
+def test_directives_end_archetype(prompts: dict) -> None:
+    for name, style in prompts["styles"].items():
+        assert "Archetipo dominante:" in style["directive"][-200:], \
+            f"{name} missing Archetipo dominante in last 200 chars"
@@ -0,0 +1,71 @@
+"""Validator accetta i 3 nuovi indicatori candle con arity corrette."""
+from __future__ import annotations
+
+import json
+
+import pytest
+
+from multi_swarm_core.protocol.parser import parse_strategy
+from multi_swarm_core.protocol.validator import ValidationError, validate_strategy
+
+
+def _strategy_with_indicator(name: str, params: list[float]) -> str:
+    return json.dumps(
+        {
+            "rules": [
+                {
+                    "condition": {
+                        "op": "gt",
+                        "args": [
+                            {"kind": "indicator", "name": name, "params": params},
+                            {"kind": "literal", "value": 0.5},
+                        ],
+                    },
+                    "action": "entry-long",
+                }
+            ]
+        }
+    )
+
+
+def test_candle_pattern_valid_min_arity() -> None:
+    s = parse_strategy(_strategy_with_indicator("candle_pattern", [3, 0, 1, 0]))
+    validate_strategy(s)
+
+
+def test_candle_pattern_valid_max_arity() -> None:
+    s = parse_strategy(_strategy_with_indicator("candle_pattern", [12] + [0] * 12))
+    validate_strategy(s)
+
+
+def test_candle_pattern_rejects_too_few_params() -> None:
+    s = parse_strategy(_strategy_with_indicator("candle_pattern", [3, 0, 1]))
+    with pytest.raises(ValidationError):
+        validate_strategy(s)
+
+
+def test_pythagorean_ratio_valid() -> None:
+    s = parse_strategy(_strategy_with_indicator("pythagorean_ratio", [89]))
+    validate_strategy(s)
+
+
+def test_pythagorean_ratio_rejects_zero_params() -> None:
+    s = parse_strategy(_strategy_with_indicator("pythagorean_ratio", []))
+    with pytest.raises(ValidationError):
+        validate_strategy(s)
+
+
+def test_fractal_mirror_valid_h() -> None:
+    s = parse_strategy(_strategy_with_indicator("fractal_mirror", [12, 0]))
+    validate_strategy(s)
+
+
+def test_fractal_mirror_valid_v() -> None:
+    s = parse_strategy(_strategy_with_indicator("fractal_mirror", [12, 1]))
+    validate_strategy(s)
+
+
+def test_fractal_mirror_rejects_one_param() -> None:
+    s = parse_strategy(_strategy_with_indicator("fractal_mirror", [12]))
+    with pytest.raises(ValidationError):
+        validate_strategy(s)
@@ -0,0 +1,634 @@
+{
+  "run_id": "0392aa1c2d644459afa5a23f43c38ac6",
+  "run_name": "phase1-btc-100-001",
+  "n_folds": 4,
+  "top_k_requested": 10,
+  "top_k_evaluated": 10,
+  "symbol": "BTC-PERPETUAL",
+  "timeframe": "1h",
+  "start": "2018-09-01T00:00:00+00:00",
+  "end": "2026-01-01T00:00:00+00:00",
+  "ohlcv_bars": 64297,
+  "results": [
+    {
+      "genome_id": "23a24989e2ed0f84",
+      "fitness_is": 0.25047738452013774,
+      "sharpe_is": 0.5152551943136504,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.4454407113532186,
+          "sharpe": 0.940612398713799,
+          "dsr": 0.09856838950479485,
+          "dsr_pvalue": 0.9014316104952051,
+          "return": 0.12691347502077277,
+          "max_dd": 0.08467873586477132,
+          "n_trades": 50,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.33651846595831003,
+          "sharpe": 0.6297236131089199,
+          "dsr": 0.05704792862404472,
+          "dsr_pvalue": 0.9429520713759553,
+          "return": 0.16916039262594973,
+          "max_dd": 0.2420995418754207,
+          "n_trades": 61,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.08496628060413243,
+          "sharpe": -0.291593157960215,
+          "dsr": 0.006828013272159182,
+          "dsr_pvalue": 0.9931719867278408,
+          "return": -0.06496567446731383,
+          "max_dd": 0.1933746053658072,
+          "n_trades": 72,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.10029133262703777,
+          "sharpe": -0.08634860278039096,
+          "dsr": 0.01165220864726802,
+          "dsr_pvalue": 0.988347791352732,
+          "return": -0.007636913661893563,
+          "max_dd": 0.061872083556258554,
+          "n_trades": 29,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.2418041976356747,
+      "fitness_oos_min": 0.08496628060413243,
+      "fitness_oos_max": 0.4454407113532186,
+      "fitness_oos_std": 0.15416115393045135,
+      "sharpe_oos_mean": 0.2980985627705282,
+      "sharpe_oos_min": -0.291593157960215,
+      "robust_score": 0.08496628060413243
+    },
+    {
+      "genome_id": "ddda3a5d7fcf95d8",
+      "fitness_is": 0.24345612215631274,
+      "sharpe_is": 0.4859910845049414,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.38630034957174436,
+          "sharpe": 0.6292230751631145,
+          "dsr": 0.05660411470808308,
+          "dsr_pvalue": 0.9433958852919169,
+          "return": 0.0808908197444953,
+          "max_dd": 0.08123461559976199,
+          "n_trades": 44,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.3444344428619903,
+          "sharpe": 0.670768621640302,
+          "dsr": 0.06172291934756436,
+          "dsr_pvalue": 0.9382770806524356,
+          "return": 0.1769344040247678,
+          "max_dd": 0.24038922925189188,
+          "n_trades": 46,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.08496628060413243,
+          "sharpe": -0.291593157960215,
+          "dsr": 0.006828013272159182,
+          "dsr_pvalue": 0.9931719867278408,
+          "return": -0.06496567446731383,
+          "max_dd": 0.1933746053658072,
+          "n_trades": 72,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.10134367585820397,
+          "sharpe": -0.25028965416710486,
+          "dsr": 0.0070613574740692985,
+          "dsr_pvalue": 0.9929386425259307,
+          "return": -0.01793962898000656,
+          "max_dd": 0.05380115145734951,
+          "n_trades": 18,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.22926118722401778,
+      "fitness_oos_min": 0.08496628060413243,
+      "fitness_oos_max": 0.38630034957174436,
+      "fitness_oos_std": 0.13703109785336132,
+      "sharpe_oos_mean": 0.18952722116902415,
+      "sharpe_oos_min": -0.291593157960215,
+      "robust_score": 0.08496628060413243
+    },
+    {
+      "genome_id": "75fffb926a15ff30",
+      "fitness_is": 0.2317839302261713,
+      "sharpe_is": 0.5074946608465971,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.06821040478798467,
+          "sharpe": -0.42408026342979865,
+          "dsr": 0.004964388516380099,
+          "dsr_pvalue": 0.9950356114836199,
+          "return": -0.004117833402575766,
+          "max_dd": 0.01551842276077859,
+          "n_trades": 12,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.30367484453258525,
+          "sharpe": 0.7648008345130556,
+          "dsr": 0.0596988345287043,
+          "dsr_pvalue": 0.9403011654712957,
+          "return": 0.040989700605122525,
+          "max_dd": 0.036878373324561994,
+          "n_trades": 31,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.014172300674502565,
+          "sharpe": -1.443859476065376,
+          "dsr": 9.400808942425867e-05,
+          "dsr_pvalue": 0.9999059919105757,
+          "return": -0.02894431062955649,
+          "max_dd": 0.036019686142963456,
+          "n_trades": 7,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.15455497647301103,
+          "sharpe": 0.11384980407793575,
+          "dsr": 0.018924861238402986,
+          "dsr_pvalue": 0.981075138761597,
+          "return": 0.004017688385377749,
+          "max_dd": 0.034520559216801125,
+          "n_trades": 21,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.1351531316170209,
+      "fitness_oos_min": 0.014172300674502565,
+      "fitness_oos_max": 0.30367484453258525,
+      "fitness_oos_std": 0.1094231344810833,
+      "sharpe_oos_mean": -0.24732227522604583,
+      "sharpe_oos_min": -1.443859476065376,
+      "robust_score": 0.014172300674502565
+    },
+    {
+      "genome_id": "1cba64abfb67fd63",
+      "fitness_is": 0.24779915639787098,
+      "sharpe_is": 0.686744434641618,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.23623418911693914,
+          "sharpe": 0.3904002210885913,
+          "dsr": 0.03526702699157904,
+          "dsr_pvalue": 0.964732973008421,
+          "return": 0.04525398577066886,
+          "max_dd": 0.09020089585606307,
+          "n_trades": 71,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.018226313030461766,
+          "sharpe": -1.287441608051841,
+          "dsr": 0.0005995136178100196,
+          "dsr_pvalue": 0.99940048638219,
+          "return": -0.08712458661694344,
+          "max_dd": 0.08774369104277313,
+          "n_trades": 23,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.005524367416662587,
+          "sharpe": -1.78701596924334,
+          "dsr": 2.01452172810692e-05,
+          "dsr_pvalue": 0.9999798547827189,
+          "return": -0.338652173037419,
+          "max_dd": 0.3725940753872713,
+          "n_trades": 60,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0633884962634609,
+          "sharpe": -0.551177699295004,
+          "dsr": 0.003506443748025072,
+          "dsr_pvalue": 0.9964935562519749,
+          "return": -0.05484487559184925,
+          "max_dd": 0.10783012035662387,
+          "n_trades": 53,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0808433414568811,
+      "fitness_oos_min": 0.005524367416662587,
+      "fitness_oos_max": 0.23623418911693914,
+      "fitness_oos_std": 0.09225620203631806,
+      "sharpe_oos_mean": -0.8088087638753985,
+      "sharpe_oos_min": -1.78701596924334,
+      "robust_score": 0.005524367416662587
+    },
+    {
+      "genome_id": "238e481262c1594c",
+      "fitness_is": 0.2604059091373716,
+      "sharpe_is": 0.4370965264203703,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.7584616560730608,
+          "sharpe": 2.414722662010953,
+          "dsr": 0.5606058252346415,
+          "dsr_pvalue": 0.43939417476535847,
+          "return": 0.0933177386269215,
+          "max_dd": 0.023570677223421994,
+          "n_trades": 3,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.16218807661575446,
+          "sharpe": 0.012873968372421442,
+          "dsr": 0.015091968754361085,
+          "dsr_pvalue": 0.9849080312456389,
+          "return": 0.0016924869828314204,
+          "max_dd": 0.14842112455963283,
+          "n_trades": 15,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.6516189601780499,
+          "sharpe": 2.0590304959441346,
+          "dsr": 0.41524419503560983,
+          "dsr_pvalue": 0.5847558049643902,
+          "return": 0.22153641670021273,
+          "max_dd": 0.0736556977759442,
+          "n_trades": 11,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.3930671732167163,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.7584616560730608,
+      "fitness_oos_std": 0.3194405713707209,
+      "sharpe_oos_mean": 1.1216567815818772,
+      "sharpe_oos_min": 0.0,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "9d484d1d7154e3d8",
+      "fitness_is": 0.24281082171633084,
+      "sharpe_is": 0.5383459590297983,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": -0.17966776746614493,
+          "dsr": 0.010596058628339637,
+          "dsr_pvalue": 0.9894039413716603,
+          "return": -0.00014822652679680193,
+          "max_dd": 0.00043372790281119014,
+          "n_trades": 1,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.03510752668391289,
+          "sharpe": -0.9516408190993537,
+          "dsr": 2.2816330745239768e-07,
+          "dsr_pvalue": 0.9999997718366925,
+          "return": -0.022264081410075964,
+          "max_dd": 0.026510937718929387,
+          "n_trades": 3,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": -1.2370529718735044,
+          "dsr": 3.550540879152349e-10,
+          "dsr_pvalue": 0.9999999996449459,
+          "return": -0.005011777500302128,
+          "max_dd": 0.005011777500302128,
+          "n_trades": 1,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.009397695173924475,
+          "sharpe": -1.662121842926161,
+          "dsr": 6.564863915519434e-10,
+          "dsr_pvalue": 0.9999999993435136,
+          "return": -0.023018388061418094,
+          "max_dd": 0.02710730869264358,
+          "n_trades": 3,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.011126305464459342,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.03510752668391289,
+      "fitness_oos_std": 0.014367292814669608,
+      "sharpe_oos_mean": -1.0076208503412911,
+      "sharpe_oos_min": -1.662121842926161,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "e1f8b425fec3b95e",
+      "fitness_is": 0.23502719247621787,
+      "sharpe_is": 0.49764640014032885,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": -1.3160615555061568,
+          "dsr": 2.0480110912755528e-14,
+          "dsr_pvalue": 0.9999999999999796,
+          "return": -0.004197667464114874,
+          "max_dd": 0.004197667464114874,
+          "n_trades": 1,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.0,
+      "fitness_oos_std": 0.0,
+      "sharpe_oos_mean": -0.3290153888765392,
+      "sharpe_oos_min": -1.3160615555061568,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "de4c3cce96bc6233",
+      "fitness_is": 0.22983732104518312,
+      "sharpe_is": 0.3042292766075092,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.7099634427811606,
+          "sharpe": 2.175438924887508,
+          "dsr": 0.4516683289747405,
+          "dsr_pvalue": 0.5483316710252595,
+          "return": 0.05162527264229322,
+          "max_dd": 0.013388106636764294,
+          "n_trades": 5,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.11214839774268798,
+          "sharpe": -0.09717412830030508,
+          "dsr": 0.011407831629063609,
+          "dsr_pvalue": 0.9885921683709364,
+          "return": -0.010810784900084469,
+          "max_dd": 0.146727446687109,
+          "n_trades": 17,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.48948263578942797,
+          "sharpe": 1.2111587578615337,
+          "dsr": 0.15099566068576153,
+          "dsr_pvalue": 0.8490043393142385,
+          "return": 0.09392150449517889,
+          "max_dd": 0.09248924622174504,
+          "n_trades": 17,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.32789861907831913,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.7099634427811606,
+      "fitness_oos_std": 0.28554710047880966,
+      "sharpe_oos_mean": 0.8223558886121842,
+      "sharpe_oos_min": -0.09717412830030508,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "14711c6816c009bb",
+      "fitness_is": 0.2231252597460312,
+      "sharpe_is": 0.5100989455793494,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.041279942400821015,
+          "sharpe": -0.7516739863723977,
+          "dsr": 0.0003109810948998818,
+          "dsr_pvalue": 0.9996890189051001,
+          "return": -0.002913092808893847,
+          "max_dd": 0.003336150788066372,
+          "n_trades": 3,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.010319985600205254,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.041279942400821015,
+      "fitness_oos_std": 0.017874739392934696,
+      "sharpe_oos_mean": -0.18791849659309942,
+      "sharpe_oos_min": -0.7516739863723977,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "c7f6f7c415a8c1df",
+      "fitness_is": 0.2231252597460312,
+      "sharpe_is": 0.5100989455793494,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.041279942400821015,
+          "sharpe": -0.7516739863723977,
+          "dsr": 0.0003109810948998818,
+          "dsr_pvalue": 0.9996890189051001,
+          "return": -0.002913092808893847,
+          "max_dd": 0.003336150788066372,
+          "n_trades": 3,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.010319985600205254,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.041279942400821015,
+      "fitness_oos_std": 0.017874739392934696,
+      "sharpe_oos_mean": -0.18791849659309942,
+      "sharpe_oos_min": -0.7516739863723977,
+      "robust_score": 0.0
+    }
+  ]
+}
@@ -0,0 +1,634 @@
+{
+  "run_id": "0392aa1c2d644459afa5a23f43c38ac6",
+  "run_name": "phase1-btc-100-001",
+  "n_folds": 4,
+  "top_k_requested": 10,
+  "top_k_evaluated": 10,
+  "symbol": "BTC-PERPETUAL",
+  "timeframe": "1h",
+  "start": "2018-09-01T00:00:00+00:00",
+  "end": "2026-01-01T00:00:00+00:00",
+  "ohlcv_bars": 64297,
+  "results": [
+    {
+      "genome_id": "23a24989e2ed0f84",
+      "fitness_is": 0.25047738452013774,
+      "sharpe_is": 0.5152551943136504,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.4454407113532186,
+          "sharpe": 0.940612398713799,
+          "dsr": 0.09856838950479485,
+          "dsr_pvalue": 0.9014316104952051,
+          "return": 0.12691347502077277,
+          "max_dd": 0.08467873586477132,
+          "n_trades": 50,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.33651846595831003,
+          "sharpe": 0.6297236131089199,
+          "dsr": 0.05704792862404472,
+          "dsr_pvalue": 0.9429520713759553,
+          "return": 0.16916039262594973,
+          "max_dd": 0.2420995418754207,
+          "n_trades": 61,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.08496628060413243,
+          "sharpe": -0.291593157960215,
+          "dsr": 0.006828013272159182,
+          "dsr_pvalue": 0.9931719867278408,
+          "return": -0.06496567446731383,
+          "max_dd": 0.1933746053658072,
+          "n_trades": 72,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.10029133262703777,
+          "sharpe": -0.08634860278039096,
+          "dsr": 0.01165220864726802,
+          "dsr_pvalue": 0.988347791352732,
+          "return": -0.007636913661893563,
+          "max_dd": 0.061872083556258554,
+          "n_trades": 29,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.2418041976356747,
+      "fitness_oos_min": 0.08496628060413243,
+      "fitness_oos_max": 0.4454407113532186,
+      "fitness_oos_std": 0.15416115393045135,
+      "sharpe_oos_mean": 0.2980985627705282,
+      "sharpe_oos_min": -0.291593157960215,
+      "robust_score": 0.08496628060413243
+    },
+    {
+      "genome_id": "ddda3a5d7fcf95d8",
+      "fitness_is": 0.24345612215631274,
+      "sharpe_is": 0.4859910845049414,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.38630034957174436,
+          "sharpe": 0.6292230751631145,
+          "dsr": 0.05660411470808308,
+          "dsr_pvalue": 0.9433958852919169,
+          "return": 0.0808908197444953,
+          "max_dd": 0.08123461559976199,
+          "n_trades": 44,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.3444344428619903,
+          "sharpe": 0.670768621640302,
+          "dsr": 0.06172291934756436,
+          "dsr_pvalue": 0.9382770806524356,
+          "return": 0.1769344040247678,
+          "max_dd": 0.24038922925189188,
+          "n_trades": 46,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.08496628060413243,
+          "sharpe": -0.291593157960215,
+          "dsr": 0.006828013272159182,
+          "dsr_pvalue": 0.9931719867278408,
+          "return": -0.06496567446731383,
+          "max_dd": 0.1933746053658072,
+          "n_trades": 72,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.10134367585820397,
+          "sharpe": -0.25028965416710486,
+          "dsr": 0.0070613574740692985,
+          "dsr_pvalue": 0.9929386425259307,
+          "return": -0.01793962898000656,
+          "max_dd": 0.05380115145734951,
+          "n_trades": 18,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.22926118722401778,
+      "fitness_oos_min": 0.08496628060413243,
+      "fitness_oos_max": 0.38630034957174436,
+      "fitness_oos_std": 0.13703109785336132,
+      "sharpe_oos_mean": 0.18952722116902415,
+      "sharpe_oos_min": -0.291593157960215,
+      "robust_score": 0.08496628060413243
+    },
+    {
+      "genome_id": "1cba64abfb67fd63",
+      "fitness_is": 0.24779915639787098,
+      "sharpe_is": 0.686744434641618,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.23623418911693914,
+          "sharpe": 0.3904002210885913,
+          "dsr": 0.03526702699157904,
+          "dsr_pvalue": 0.964732973008421,
+          "return": 0.04525398577066886,
+          "max_dd": 0.09020089585606307,
+          "n_trades": 71,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.018226313030461766,
+          "sharpe": -1.287441608051841,
+          "dsr": 0.0005995136178100196,
+          "dsr_pvalue": 0.99940048638219,
+          "return": -0.08712458661694344,
+          "max_dd": 0.08774369104277313,
+          "n_trades": 23,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.005524367416662587,
+          "sharpe": -1.78701596924334,
+          "dsr": 2.01452172810692e-05,
+          "dsr_pvalue": 0.9999798547827189,
+          "return": -0.338652173037419,
+          "max_dd": 0.3725940753872713,
+          "n_trades": 60,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0633884962634609,
+          "sharpe": -0.551177699295004,
+          "dsr": 0.003506443748025072,
+          "dsr_pvalue": 0.9964935562519749,
+          "return": -0.05484487559184925,
+          "max_dd": 0.10783012035662387,
+          "n_trades": 53,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0808433414568811,
+      "fitness_oos_min": 0.005524367416662587,
+      "fitness_oos_max": 0.23623418911693914,
+      "fitness_oos_std": 0.09225620203631806,
+      "sharpe_oos_mean": -0.8088087638753985,
+      "sharpe_oos_min": -1.78701596924334,
+      "robust_score": 0.005524367416662587
+    },
+    {
+      "genome_id": "238e481262c1594c",
+      "fitness_is": 0.2604059091373716,
+      "sharpe_is": 0.4370965264203703,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": 2.414722662010953,
+          "dsr": 0.5606058252346415,
+          "dsr_pvalue": 0.43939417476535847,
+          "return": 0.0933177386269215,
+          "max_dd": 0.023570677223421994,
+          "n_trades": 3,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.16218807661575446,
+          "sharpe": 0.012873968372421442,
+          "dsr": 0.015091968754361085,
+          "dsr_pvalue": 0.9849080312456389,
+          "return": 0.0016924869828314204,
+          "max_dd": 0.14842112455963283,
+          "n_trades": 15,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.6516189601780499,
+          "sharpe": 2.0590304959441346,
+          "dsr": 0.41524419503560983,
+          "dsr_pvalue": 0.5847558049643902,
+          "return": 0.22153641670021273,
+          "max_dd": 0.0736556977759442,
+          "n_trades": 11,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.2034517591984511,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.6516189601780499,
+      "fitness_oos_std": 0.2670869559600687,
+      "sharpe_oos_mean": 1.1216567815818772,
+      "sharpe_oos_min": 0.0,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "9d484d1d7154e3d8",
+      "fitness_is": 0.24281082171633084,
+      "sharpe_is": 0.5383459590297983,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": -0.17966776746614493,
+          "dsr": 0.010596058628339637,
+          "dsr_pvalue": 0.9894039413716603,
+          "return": -0.00014822652679680193,
+          "max_dd": 0.00043372790281119014,
+          "n_trades": 1,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": -0.9516408190993537,
+          "dsr": 2.2816330745239768e-07,
+          "dsr_pvalue": 0.9999997718366925,
+          "return": -0.022264081410075964,
+          "max_dd": 0.026510937718929387,
+          "n_trades": 3,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": -1.2370529718735044,
+          "dsr": 3.550540879152349e-10,
+          "dsr_pvalue": 0.9999999996449459,
+          "return": -0.005011777500302128,
+          "max_dd": 0.005011777500302128,
+          "n_trades": 1,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": -1.662121842926161,
+          "dsr": 6.564863915519434e-10,
+          "dsr_pvalue": 0.9999999993435136,
+          "return": -0.023018388061418094,
+          "max_dd": 0.02710730869264358,
+          "n_trades": 3,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.0,
+      "fitness_oos_std": 0.0,
+      "sharpe_oos_mean": -1.0076208503412911,
+      "sharpe_oos_min": -1.662121842926161,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "e1f8b425fec3b95e",
+      "fitness_is": 0.23502719247621787,
+      "sharpe_is": 0.49764640014032885,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": -1.3160615555061568,
+          "dsr": 2.0480110912755528e-14,
+          "dsr_pvalue": 0.9999999999999796,
+          "return": -0.004197667464114874,
+          "max_dd": 0.004197667464114874,
+          "n_trades": 1,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.0,
+      "fitness_oos_std": 0.0,
+      "sharpe_oos_mean": -0.3290153888765392,
+      "sharpe_oos_min": -1.3160615555061568,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "75fffb926a15ff30",
+      "fitness_is": 0.2317839302261713,
+      "sharpe_is": 0.5074946608465971,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.06821040478798467,
+          "sharpe": -0.42408026342979865,
+          "dsr": 0.004964388516380099,
+          "dsr_pvalue": 0.9950356114836199,
+          "return": -0.004117833402575766,
+          "max_dd": 0.01551842276077859,
+          "n_trades": 12,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.30367484453258525,
+          "sharpe": 0.7648008345130556,
+          "dsr": 0.0596988345287043,
+          "dsr_pvalue": 0.9403011654712957,
+          "return": 0.040989700605122525,
+          "max_dd": 0.036878373324561994,
+          "n_trades": 31,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": -1.443859476065376,
+          "dsr": 9.400808942425867e-05,
+          "dsr_pvalue": 0.9999059919105757,
+          "return": -0.02894431062955649,
+          "max_dd": 0.036019686142963456,
+          "n_trades": 7,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.15455497647301103,
+          "sharpe": 0.11384980407793575,
+          "dsr": 0.018924861238402986,
+          "dsr_pvalue": 0.981075138761597,
+          "return": 0.004017688385377749,
+          "max_dd": 0.034520559216801125,
+          "n_trades": 21,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.13161005644839524,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.30367484453258525,
+      "fitness_oos_std": 0.11343884193961566,
+      "sharpe_oos_mean": -0.24732227522604583,
+      "sharpe_oos_min": -1.443859476065376,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "de4c3cce96bc6233",
+      "fitness_is": 0.22983732104518312,
+      "sharpe_is": 0.3042292766075092,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": 2.175438924887508,
+          "dsr": 0.4516683289747405,
+          "dsr_pvalue": 0.5483316710252595,
+          "return": 0.05162527264229322,
+          "max_dd": 0.013388106636764294,
+          "n_trades": 5,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.11214839774268798,
+          "sharpe": -0.09717412830030508,
+          "dsr": 0.011407831629063609,
+          "dsr_pvalue": 0.9885921683709364,
+          "return": -0.010810784900084469,
+          "max_dd": 0.146727446687109,
+          "n_trades": 17,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.48948263578942797,
+          "sharpe": 1.2111587578615337,
+          "dsr": 0.15099566068576153,
+          "dsr_pvalue": 0.8490043393142385,
+          "return": 0.09392150449517889,
+          "max_dd": 0.09248924622174504,
+          "n_trades": 17,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.150407758383029,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.48948263578942797,
+      "fitness_oos_std": 0.20104759307710415,
+      "sharpe_oos_mean": 0.8223558886121842,
+      "sharpe_oos_min": -0.09717412830030508,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "14711c6816c009bb",
+      "fitness_is": 0.2231252597460312,
+      "sharpe_is": 0.5100989455793494,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": -0.7516739863723977,
+          "dsr": 0.0003109810948998818,
+          "dsr_pvalue": 0.9996890189051001,
+          "return": -0.002913092808893847,
+          "max_dd": 0.003336150788066372,
+          "n_trades": 3,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.0,
+      "fitness_oos_std": 0.0,
+      "sharpe_oos_mean": -0.18791849659309942,
+      "sharpe_oos_min": -0.7516739863723977,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "c7f6f7c415a8c1df",
+      "fitness_is": 0.2231252597460312,
+      "sharpe_is": 0.5100989455793494,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": -0.7516739863723977,
+          "dsr": 0.0003109810948998818,
+          "dsr_pvalue": 0.9996890189051001,
+          "return": -0.002913092808893847,
+          "max_dd": 0.003336150788066372,
+          "n_trades": 3,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.0,
+      "fitness_oos_std": 0.0,
+      "sharpe_oos_mean": -0.18791849659309942,
+      "sharpe_oos_min": -0.7516739863723977,
+      "robust_score": 0.0
+    }
+  ]
+}
@@ -0,0 +1,634 @@
+{
+  "run_id": "1b3ac5a72ade4eeca96f63af1c692529",
+  "run_name": "phase1-btc-100-5m-001",
+  "n_folds": 4,
+  "top_k_requested": 10,
+  "top_k_evaluated": 10,
+  "symbol": "BTC-PERPETUAL",
+  "timeframe": "5m",
+  "start": "2018-09-01T00:00:00+00:00",
+  "end": "2026-01-01T00:00:00+00:00",
+  "ohlcv_bars": 771553,
+  "results": [
+    {
+      "genome_id": "f8ca6642adf7e0cd",
+      "fitness_is": 0.18938156615467444,
+      "sharpe_is": 0.30869023048836236,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.2352965252827188,
+          "sharpe": 0.3693814228315177,
+          "dsr": 1.683506715156222e-10,
+          "dsr_pvalue": 0.9999999998316493,
+          "return": 0.04254736022830463,
+          "max_dd": 0.027161767758352023,
+          "n_trades": 22,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:55:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.225344374241631,
+          "sharpe": 0.2966596569312926,
+          "dsr": 1.0070583060973712e-10,
+          "dsr_pvalue": 0.9999999998992942,
+          "return": 0.0323635859324094,
+          "max_dd": 0.020860357367648123,
+          "n_trades": 4,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:55:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.22012851949138668,
+          "sharpe": 0.2582559733706436,
+          "dsr": 6.7095779615621785e-12,
+          "dsr_pvalue": 0.9999999999932904,
+          "return": 0.024874816913196707,
+          "max_dd": 0.01617874217143948,
+          "n_trades": 6,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:55:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.10824559047403624,
+          "sharpe": -0.19939596313642038,
+          "dsr": 4.446294307701592e-18,
+          "dsr_pvalue": 1.0,
+          "return": -0.012815761172245588,
+          "max_dd": 0.030585031985383666,
+          "n_trades": 4,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:55:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.1972537523724432,
+      "fitness_oos_min": 0.10824559047403624,
+      "fitness_oos_max": 0.2352965252827188,
+      "fitness_oos_std": 0.05167698584229592,
+      "sharpe_oos_mean": 0.18122527249925838,
+      "sharpe_oos_min": -0.19939596313642038,
+      "robust_score": 0.10824559047403624
+    },
+    {
+      "genome_id": "44e4031ccb1dc02a",
+      "fitness_is": 0.22571210498548258,
+      "sharpe_is": 0.3821051381739021,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.2750172552747618,
+          "sharpe": 0.6074693455147703,
+          "dsr": 1.1545160457503457e-26,
+          "dsr_pvalue": 1.0,
+          "return": 0.043108172266182354,
+          "max_dd": 0.001458755248422002,
+          "n_trades": 10,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:55:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": 0.06258366543576925,
+          "dsr": 2.937538417346168e-13,
+          "dsr_pvalue": 0.9999999999997062,
+          "return": 0.005830222073507807,
+          "max_dd": 0.020860357367648123,
+          "n_trades": 2,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:55:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.18648515967821389,
+          "sharpe": 0.06128925460246632,
+          "dsr": 1.8274259811076416e-13,
+          "dsr_pvalue": 0.9999999999998173,
+          "return": 0.004414468283732154,
+          "max_dd": 0.01617874217143948,
+          "n_trades": 3,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:55:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:55:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.11537560373824392,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.2750172552747618,
+      "fitness_oos_std": 0.11954610588306011,
+      "sharpe_oos_mean": 0.18283556638825146,
+      "sharpe_oos_min": 0.0,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "662816b63c4feadf",
+      "fitness_is": 0.20828744896100454,
+      "sharpe_is": 0.20788153582834326,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": 0.09737359146830797,
+          "dsr": 2.4288846142332667e-14,
+          "dsr_pvalue": 0.9999999999999757,
+          "return": 0.0030306913996627216,
+          "max_dd": 0.007271637996670698,
+          "n_trades": 2,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:55:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:55:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:55:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:55:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.0,
+      "fitness_oos_std": 0.0,
+      "sharpe_oos_mean": 0.024343397867076993,
+      "sharpe_oos_min": 0.0,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "cf698463450b5daf",
+      "fitness_is": 0.20181196371847498,
+      "sharpe_is": 0.3623641433398033,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.2266655258766632,
+          "sharpe": 0.2999909794832819,
+          "dsr": 6.8141661859725965e-12,
+          "dsr_pvalue": 0.9999999999931858,
+          "return": 0.018823783888960888,
+          "max_dd": 0.01731422516010631,
+          "n_trades": 4,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:55:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": 0.44955543499020745,
+          "dsr": 9.481303288744627e-40,
+          "dsr_pvalue": 1.0,
+          "return": 0.049849479597766866,
+          "max_dd": 0.004685656497556025,
+          "n_trades": 1,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:55:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.04410712874184496,
+          "dsr": 4.2725586842847596e-14,
+          "dsr_pvalue": 0.9999999999999573,
+          "return": 0.004410420254631564,
+          "max_dd": 0.01996463311196068,
+          "n_trades": 2,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:55:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.4450404801682897,
+          "dsr": 1.4117362531214693e-27,
+          "dsr_pvalue": 1.0,
+          "return": 0.0526088849839641,
+          "max_dd": 0.005792329453230141,
+          "n_trades": 1,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:55:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0566663814691658,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.2266655258766632,
+      "fitness_oos_std": 0.09814905178567468,
+      "sharpe_oos_mean": 0.30967350584590597,
+      "sharpe_oos_min": 0.04410712874184496,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "ddbfbef72f07d76a",
+      "fitness_is": 0.2007403940095307,
+      "sharpe_is": 0.3527941850815434,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.25816377963939535,
+          "sharpe": 0.5144509577982044,
+          "dsr": 8.645767195175167e-13,
+          "dsr_pvalue": 0.9999999999991355,
+          "return": 0.07304433130107668,
+          "max_dd": 0.01915225360494227,
+          "n_trades": 5,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:55:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": 0.44955543499020745,
+          "dsr": 9.481303288744627e-40,
+          "dsr_pvalue": 1.0,
+          "return": 0.049849479597766866,
+          "max_dd": 0.004685656497556025,
+          "n_trades": 1,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:55:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.23195306903928886,
+          "sharpe": 0.32337815238365364,
+          "dsr": 8.912615856501225e-14,
+          "dsr_pvalue": 0.9999999999999108,
+          "return": 0.033935014244555584,
+          "max_dd": 0.010486324211451901,
+          "n_trades": 3,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:55:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.35283933132790596,
+          "dsr": 6.658494490709348e-14,
+          "dsr_pvalue": 0.9999999999999334,
+          "return": 0.05344386510725729,
+          "max_dd": 0.022762761145810856,
+          "n_trades": 1,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:55:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.12252921216967105,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.25816377963939535,
+      "fitness_oos_std": 0.12287913982320425,
+      "sharpe_oos_mean": 0.41005596912499287,
+      "sharpe_oos_min": 0.32337815238365364,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "2e49f8e8e6cec686",
+      "fitness_is": 0.19435604112910157,
+      "sharpe_is": 0.3713282033185845,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.26072350293525676,
+          "sharpe": 0.5284128848027025,
+          "dsr": 2.486680252274251e-13,
+          "dsr_pvalue": 0.9999999999997513,
+          "return": 0.05881454144506426,
+          "max_dd": 0.016516772761264658,
+          "n_trades": 5,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:55:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": 0.44955543499020745,
+          "dsr": 9.481303288744627e-40,
+          "dsr_pvalue": 1.0,
+          "return": 0.049849479597766866,
+          "max_dd": 0.004685656497556025,
+          "n_trades": 1,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:55:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.21422619151698888,
+          "sharpe": 0.2269482258874523,
+          "dsr": 3.6571308232390715e-13,
+          "dsr_pvalue": 0.9999999999996343,
+          "return": 0.02568538444204793,
+          "max_dd": 0.019558622265454236,
+          "n_trades": 3,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:55:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.4450404801682897,
+          "dsr": 1.4117362531214693e-27,
+          "dsr_pvalue": 1.0,
+          "return": 0.0526088849839641,
+          "max_dd": 0.005792329453230141,
+          "n_trades": 1,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:55:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.11873742361306142,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.26072350293525676,
+      "fitness_oos_std": 0.11987003696674377,
+      "sharpe_oos_mean": 0.412489256462163,
+      "sharpe_oos_min": 0.2269482258874523,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "9852e8e4e480ef21",
+      "fitness_is": 0.1922948882127238,
+      "sharpe_is": 0.15925760158030436,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.1494822926244717,
+          "sharpe": 0.0804180937424291,
+          "dsr": 2.443102889370683e-16,
+          "dsr_pvalue": 0.9999999999999998,
+          "return": 0.0014913672331040573,
+          "max_dd": 0.003691144778103467,
+          "n_trades": 3,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:55:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.021783719225659744,
+          "sharpe": -1.222169028737873,
+          "dsr": 2.0819357051846864e-80,
+          "dsr_pvalue": 1.0,
+          "return": -0.018262798724970497,
+          "max_dd": 0.018262798724970497,
+          "n_trades": 16,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:55:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:55:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:55:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.042816502962532865,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.1494822926244717,
+      "fitness_oos_std": 0.06222233226062855,
+      "sharpe_oos_mean": -0.285437733748861,
+      "sharpe_oos_min": -1.222169028737873,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "16a1761c8da604e7",
+      "fitness_is": 0.19121238216295305,
+      "sharpe_is": 0.15255013590499958,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:55:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:55:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:55:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:55:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.0,
+      "fitness_oos_std": 0.0,
+      "sharpe_oos_mean": 0.0,
+      "sharpe_oos_min": 0.0,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "add18cb7382296fd",
+      "fitness_is": 0.1879996666147147,
+      "sharpe_is": 0.28550736974923013,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:55:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": 0.44955543499020745,
+          "dsr": 9.481303288744627e-40,
+          "dsr_pvalue": 1.0,
+          "return": 0.049849479597766866,
+          "max_dd": 0.004685656497556025,
+          "n_trades": 1,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:55:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.150691158813143,
+          "dsr": 4.263193477728488e-17,
+          "dsr_pvalue": 1.0,
+          "return": 0.009379645576139906,
+          "max_dd": 0.008611115582095263,
+          "n_trades": 1,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:55:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.46456569507517853,
+          "dsr": 9.733950727895419e-26,
+          "dsr_pvalue": 1.0,
+          "return": 0.056366295538783584,
+          "max_dd": 0.005792329453230141,
+          "n_trades": 1,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:55:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.0,
+      "fitness_oos_std": 0.0,
+      "sharpe_oos_mean": 0.26620307221963224,
+      "sharpe_oos_min": 0.0,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "4ee566eb5e132d8c",
+      "fitness_is": 0.18745393246665415,
+      "sharpe_is": 0.31006907324235156,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.2166522335961314,
+          "sharpe": 0.23875257586420637,
+          "dsr": 4.3199206915491695e-12,
+          "dsr_pvalue": 0.9999999999956801,
+          "return": 0.01599067972499668,
+          "max_dd": 0.01736213317014228,
+          "n_trades": 3,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:55:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": 0.44955543499020745,
+          "dsr": 9.481303288744627e-40,
+          "dsr_pvalue": 1.0,
+          "return": 0.049849479597766866,
+          "max_dd": 0.004685656497556025,
+          "n_trades": 1,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:55:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.04410712874184496,
+          "dsr": 4.2725586842847596e-14,
+          "dsr_pvalue": 0.9999999999999573,
+          "return": 0.004410420254631564,
+          "max_dd": 0.01996463311196068,
+          "n_trades": 2,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:55:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.46456569507517853,
+          "dsr": 9.733950727895419e-26,
+          "dsr_pvalue": 1.0,
+          "return": 0.056366295538783584,
+          "max_dd": 0.005792329453230141,
+          "n_trades": 1,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:55:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.05416305839903285,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.2166522335961314,
+      "fitness_oos_std": 0.09381316904044511,
+      "sharpe_oos_mean": 0.2992452086678593,
+      "sharpe_oos_min": 0.04410712874184496,
+      "robust_score": 0.0
+    }
+  ]
+}
@@ -0,0 +1,634 @@
+{
+  "run_id": "e263651598894da688d95fda90a34a96",
+  "run_name": "phase1-extended-001",
+  "n_folds": 4,
+  "top_k_requested": 10,
+  "top_k_evaluated": 10,
+  "symbol": "BTC-PERPETUAL",
+  "timeframe": "1h",
+  "start": "2018-09-01T00:00:00+00:00",
+  "end": "2026-01-01T00:00:00+00:00",
+  "ohlcv_bars": 64297,
+  "results": [
+    {
+      "genome_id": "fe6e01eb690d3960",
+      "fitness_is": 0.3513762485888574,
+      "sharpe_is": 0.9011072752402621,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.18429629882863333,
+          "sharpe": 0.18578971215949266,
+          "dsr": 0.022696217108110216,
+          "dsr_pvalue": 0.9773037828918898,
+          "return": 0.03138679502492736,
+          "max_dd": 0.19089436189057732,
+          "n_trades": 90,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.5261711065682141,
+          "sharpe": 1.875154815895858,
+          "dsr": 0.34945842578669783,
+          "dsr_pvalue": 0.6505415742133022,
+          "return": 0.684094224950746,
+          "max_dd": 0.26051011671170043,
+          "n_trades": 108,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.29904777517998976,
+          "sharpe": 0.45027968136531676,
+          "dsr": 0.040236883094469954,
+          "dsr_pvalue": 0.9597631169055301,
+          "return": 0.16192920610625539,
+          "max_dd": 0.25615601205401484,
+          "n_trades": 87,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.06652893772008044,
+          "sharpe": -0.7859068293026578,
+          "dsr": 0.0016949251764253048,
+          "dsr_pvalue": 0.9983050748235747,
+          "return": -0.1801701961968295,
+          "max_dd": 0.3050931306970407,
+          "n_trades": 89,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.2690110295742294,
+      "fitness_oos_min": 0.06652893772008044,
+      "fitness_oos_max": 0.5261711065682141,
+      "fitness_oos_std": 0.16971232602043682,
+      "sharpe_oos_mean": 0.43132934502950243,
+      "sharpe_oos_min": -0.7859068293026578,
+      "robust_score": 0.06652893772008044
+    },
+    {
+      "genome_id": "d98739b2ba8d65e8",
+      "fitness_is": 0.3581811122056351,
+      "sharpe_is": 1.5316294902683918,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.03897301517910094,
+          "sharpe": -1.1213338499931884,
+          "dsr": 0.0007106609032094727,
+          "dsr_pvalue": 0.9992893390967905,
+          "return": -0.18358503193809717,
+          "max_dd": 0.24053109269341416,
+          "n_trades": 65,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.13868820192362147,
+          "sharpe": 0.1139203144076397,
+          "dsr": 0.018885390702584475,
+          "dsr_pvalue": 0.9811146092974156,
+          "return": 0.019694298831973045,
+          "max_dd": 0.1528666578131679,
+          "n_trades": 21,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.43784574830368517,
+          "sharpe": 1.922993791724599,
+          "dsr": 0.36463909799020594,
+          "dsr_pvalue": 0.6353609020097941,
+          "return": 0.31046814355338936,
+          "max_dd": 0.09604869735695161,
+          "n_trades": 48,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.377910610011883,
+          "sharpe": 1.3313322991701542,
+          "dsr": 0.17322722607861918,
+          "dsr_pvalue": 0.8267727739213808,
+          "return": 0.12021252899342505,
+          "max_dd": 0.04712452925993322,
+          "n_trades": 32,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.24835439385457264,
+      "fitness_oos_min": 0.03897301517910094,
+      "fitness_oos_max": 0.43784574830368517,
+      "fitness_oos_std": 0.1647414807695958,
+      "sharpe_oos_mean": 0.5617281388273011,
+      "sharpe_oos_min": -1.1213338499931884,
+      "robust_score": 0.03897301517910094
+    },
+    {
+      "genome_id": "0dd6619fdcbe37f4",
+      "fitness_is": 0.3765498201912705,
+      "sharpe_is": 0.9388498977535525,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.03550929012857722,
+          "sharpe": -1.1093436310362703,
+          "dsr": 0.0005959199839818188,
+          "dsr_pvalue": 0.9994040800160182,
+          "return": -0.24600377155172415,
+          "max_dd": 0.38950670401585935,
+          "n_trades": 122,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.049414033104104804,
+          "sharpe": -0.9995738302518242,
+          "dsr": 0.0007695523975576842,
+          "dsr_pvalue": 0.9992304476024423,
+          "return": -0.15162579158457645,
+          "max_dd": 0.21485725923368693,
+          "n_trades": 79,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.4634730710112187,
+          "sharpe": 1.2076100738049764,
+          "dsr": 0.15231462029944348,
+          "dsr_pvalue": 0.8476853797005566,
+          "return": 0.3072592298707053,
+          "max_dd": 0.15464658494822137,
+          "n_trades": 153,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.1215451856323226,
+          "sharpe": -0.22288474503734051,
+          "dsr": 0.00830898358713792,
+          "dsr_pvalue": 0.9916910164128621,
+          "return": -0.035113534418310444,
+          "max_dd": 0.17145164314561556,
+          "n_trades": 74,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.16748539496905585,
+      "fitness_oos_min": 0.03550929012857722,
+      "fitness_oos_max": 0.4634730710112187,
+      "fitness_oos_std": 0.17398113830545858,
+      "sharpe_oos_mean": -0.28104803313011467,
+      "sharpe_oos_min": -1.1093436310362703,
+      "robust_score": 0.03550929012857722
+    },
+    {
+      "genome_id": "00545b157923dc6b",
+      "fitness_is": 0.34960092770407836,
+      "sharpe_is": 0.9785144736247469,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.027180640826810924,
+          "sharpe": -1.26345444662059,
+          "dsr": 0.00030602669379661417,
+          "dsr_pvalue": 0.9996939733062034,
+          "return": -0.2411006699210636,
+          "max_dd": 0.36676673786660746,
+          "n_trades": 79,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.5208756815430423,
+          "sharpe": 1.770946043023605,
+          "dsr": 0.3128216779179025,
+          "dsr_pvalue": 0.6871783220820975,
+          "return": 0.60240576625387,
+          "max_dd": 0.2331907909757441,
+          "n_trades": 45,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.2959791903863346,
+          "sharpe": 0.3683437692731339,
+          "dsr": 0.033871432018897314,
+          "dsr_pvalue": 0.9661285679811027,
+          "return": 0.134452978611995,
+          "max_dd": 0.19964940601028408,
+          "n_trades": 58,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.15883382666451054,
+          "sharpe": 0.031160944245222175,
+          "dsr": 0.015769276637227145,
+          "dsr_pvalue": 0.9842307233627728,
+          "return": 0.006899603676653765,
+          "max_dd": 0.1947452948903118,
+          "n_trades": 36,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.2507173348551746,
+      "fitness_oos_min": 0.027180640826810924,
+      "fitness_oos_max": 0.5208756815430423,
+      "fitness_oos_std": 0.1826508968673572,
+      "sharpe_oos_mean": 0.2267490774803428,
+      "sharpe_oos_min": -1.26345444662059,
+      "robust_score": 0.027180640826810924
+    },
+    {
+      "genome_id": "eea882db55f8dd5e",
+      "fitness_is": 0.3181543783162782,
+      "sharpe_is": 1.3214317725559044,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.02689139923003529,
+          "sharpe": -1.3020990219857191,
+          "dsr": 0.00036878360260577075,
+          "dsr_pvalue": 0.9996312163973943,
+          "return": -0.22280992288117984,
+          "max_dd": 0.28735423155244094,
+          "n_trades": 69,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.1064185213723357,
+          "sharpe": -0.3110958507357913,
+          "dsr": 0.006879713400940769,
+          "dsr_pvalue": 0.9931202865990593,
+          "return": -0.057958925555868235,
+          "max_dd": 0.19529025657454543,
+          "n_trades": 33,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.37938211463073945,
+          "sharpe": 1.490827385000592,
+          "dsr": 0.2224878866287407,
+          "dsr_pvalue": 0.7775121133712593,
+          "return": 0.2563445845249124,
+          "max_dd": 0.1053935530975972,
+          "n_trades": 59,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.4175480977898098,
+          "sharpe": 1.6336803280087653,
+          "dsr": 0.260079721684657,
+          "dsr_pvalue": 0.739920278315343,
+          "return": 0.157201259896103,
+          "max_dd": 0.04639072483391873,
+          "n_trades": 43,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.23256003325573005,
+      "fitness_oos_min": 0.02689139923003529,
+      "fitness_oos_max": 0.4175480977898098,
+      "fitness_oos_std": 0.16881097094204098,
+      "sharpe_oos_mean": 0.3778282100719617,
+      "sharpe_oos_min": -1.3020990219857191,
+      "robust_score": 0.02689139923003529
+    },
+    {
+      "genome_id": "24a3e592397d4bda",
+      "fitness_is": 0.3181543783162782,
+      "sharpe_is": 1.3214317725559044,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.02689139923003529,
+          "sharpe": -1.3020990219857191,
+          "dsr": 0.00036878360260577075,
+          "dsr_pvalue": 0.9996312163973943,
+          "return": -0.22280992288117984,
+          "max_dd": 0.28735423155244094,
+          "n_trades": 69,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.1064185213723357,
+          "sharpe": -0.3110958507357913,
+          "dsr": 0.006879713400940769,
+          "dsr_pvalue": 0.9931202865990593,
+          "return": -0.057958925555868235,
+          "max_dd": 0.19529025657454543,
+          "n_trades": 33,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.37938211463073945,
+          "sharpe": 1.490827385000592,
+          "dsr": 0.2224878866287407,
+          "dsr_pvalue": 0.7775121133712593,
+          "return": 0.2563445845249124,
+          "max_dd": 0.1053935530975972,
+          "n_trades": 59,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.4175480977898098,
+          "sharpe": 1.6336803280087653,
+          "dsr": 0.260079721684657,
+          "dsr_pvalue": 0.739920278315343,
+          "return": 0.157201259896103,
+          "max_dd": 0.04639072483391873,
+          "n_trades": 43,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.23256003325573005,
+      "fitness_oos_min": 0.02689139923003529,
+      "fitness_oos_max": 0.4175480977898098,
+      "fitness_oos_std": 0.16881097094204098,
+      "sharpe_oos_mean": 0.3778282100719617,
+      "sharpe_oos_min": -1.3020990219857191,
+      "robust_score": 0.02689139923003529
+    },
+    {
+      "genome_id": "7d6636945c3fdee5",
+      "fitness_is": 0.4143633094822953,
+      "sharpe_is": 1.0518304581395579,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.024065408548587387,
+          "sharpe": -1.362275778196952,
+          "dsr": 0.0002459819911643212,
+          "dsr_pvalue": 0.9997540180088357,
+          "return": -0.27317643202118824,
+          "max_dd": 0.283710961847723,
+          "n_trades": 76,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.10523269843210308,
+          "sharpe": -0.32050989428963983,
+          "dsr": 0.006397259401103736,
+          "dsr_pvalue": 0.9936027405988963,
+          "return": -0.058925951660568465,
+          "max_dd": 0.19264026914858506,
+          "n_trades": 115,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.37331175980311637,
+          "sharpe": 0.6268880704230287,
+          "dsr": 0.05821057212982157,
+          "dsr_pvalue": 0.9417894278701784,
+          "return": 0.13285431586579133,
+          "max_dd": 0.11992643530986899,
+          "n_trades": 66,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.058003508089270274,
+          "sharpe": -0.9065054278150867,
+          "dsr": 0.0011919907667399207,
+          "dsr_pvalue": 0.9988080092332601,
+          "return": -0.11553575296643481,
+          "max_dd": 0.21946607715577302,
+          "n_trades": 108,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.14015334371826926,
+      "fitness_oos_min": 0.024065408548587387,
+      "fitness_oos_max": 0.37331175980311637,
+      "fitness_oos_std": 0.13766562991922562,
+      "sharpe_oos_mean": -0.4906007574696624,
+      "sharpe_oos_min": -1.362275778196952,
+      "robust_score": 0.024065408548587387
+    },
+    {
+      "genome_id": "94413564f2f6c03e",
+      "fitness_is": 0.4504680245893446,
+      "sharpe_is": 1.931509193896414,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.1454684156938894,
+          "sharpe": -0.1847503153026734,
+          "dsr": 0.010432618759993285,
+          "dsr_pvalue": 0.9895673812400068,
+          "return": -0.006041376713751601,
+          "max_dd": 0.028928093840880797,
+          "n_trades": 10,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": 0.7414755036663777,
+          "dsr": 0.032908166066855475,
+          "dsr_pvalue": 0.9670918339331446,
+          "return": 0.024390796509991652,
+          "max_dd": 0.01848789755136504,
+          "n_trades": 3,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.6298210284045589,
+          "sharpe": 2.8227093941852077,
+          "dsr": 0.783195485997482,
+          "dsr_pvalue": 0.216804514002518,
+          "return": 0.1153800539734966,
+          "max_dd": 0.009172322853563938,
+          "n_trades": 13,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": -0.3705731110916588,
+          "dsr": 0.005298299731142135,
+          "dsr_pvalue": 0.9947017002688578,
+          "return": -0.01618834464316521,
+          "max_dd": 0.050704690156295765,
+          "n_trades": 8,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.1938223610246121,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.6298210284045589,
+      "fitness_oos_std": 0.2586344704657744,
+      "sharpe_oos_mean": 0.7522153678643133,
+      "sharpe_oos_min": -0.3705731110916588,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "9deb48ad9a1d6a01",
+      "fitness_is": 0.38499883511982724,
+      "sharpe_is": 1.7309660717184492,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.11865902601061347,
+          "sharpe": -0.33840465028379796,
+          "dsr": 0.006577366557928182,
+          "dsr_pvalue": 0.9934226334420718,
+          "return": -0.012941050841296264,
+          "max_dd": 0.03402877313520886,
+          "n_trades": 17,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": 0.5508017623316874,
+          "dsr": 0.028490827818300125,
+          "dsr_pvalue": 0.9715091721816999,
+          "return": 0.01831303651843519,
+          "max_dd": 0.017308885097101046,
+          "n_trades": 7,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 2.1433702644761987,
+          "dsr": 0.4071948898247041,
+          "dsr_pvalue": 0.5928051101752959,
+          "return": 0.10827348451282881,
+          "max_dd": 0.008755897100329497,
+          "n_trades": 8,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 1.0955591683486918,
+          "dsr": 0.07909942404164044,
+          "dsr_pvalue": 0.9209005759583595,
+          "return": 0.01372945300669004,
+          "max_dd": 0.006329669977511941,
+          "n_trades": 5,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.029664756502653367,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.11865902601061347,
+      "fitness_oos_std": 0.05138086545675487,
+      "sharpe_oos_mean": 0.862831636218195,
+      "sharpe_oos_min": -0.33840465028379796,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "1bb0f36ae3f178ae",
+      "fitness_is": 0.3289012666358953,
+      "sharpe_is": 1.2144322416340778,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.03325222099260249,
+          "sharpe": -1.0706470661093082,
+          "dsr": 0.0007248785745670455,
+          "dsr_pvalue": 0.999275121425433,
+          "return": -0.09901039935604483,
+          "max_dd": 0.13711458096989376,
+          "n_trades": 27,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": 0.12138335573748249,
+          "dsr": 0.018713173174332784,
+          "dsr_pvalue": 0.9812868268256673,
+          "return": 0.00555346714044469,
+          "max_dd": 0.02724874190208475,
+          "n_trades": 7,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.32995249837249846,
+          "sharpe": 0.9733164430116448,
+          "dsr": 0.07297295015940324,
+          "dsr_pvalue": 0.9270270498405968,
+          "return": 0.05047188141942249,
+          "max_dd": 0.02617799182077058,
+          "n_trades": 13,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 1.3421850442772436,
+          "dsr": 0.05521595540810741,
+          "dsr_pvalue": 0.9447840445918926,
+          "return": 0.06121506651714692,
+          "max_dd": 0.00995276327591354,
+          "n_trades": 3,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.09080117984127524,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.32995249837249846,
+      "fitness_oos_std": 0.13873981434768828,
+      "sharpe_oos_mean": 0.34155944422926565,
+      "sharpe_oos_min": -1.0706470661093082,
+      "robust_score": 0.0
+    }
+  ]
+}
@@ -0,0 +1,324 @@
+{
+  "run_id": "7f65bd1832d94c638b588aab02fb223e",
+  "run_name": "phase1-hardened-001",
+  "n_folds": 4,
+  "top_k_requested": 5,
+  "top_k_evaluated": 5,
+  "symbol": "BTC-PERPETUAL",
+  "timeframe": "1h",
+  "start": "2018-09-01T00:00:00+00:00",
+  "end": "2026-01-01T00:00:00+00:00",
+  "ohlcv_bars": 64297,
+  "results": [
+    {
+      "genome_id": "9cf506b83bec55f6",
+      "fitness_is": 0.2847465321576384,
+      "sharpe_is": 0.6809931251900289,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.4454407113532186,
+          "sharpe": 0.940612398713799,
+          "dsr": 0.09856838950479485,
+          "dsr_pvalue": 0.9014316104952051,
+          "return": 0.12691347502077277,
+          "max_dd": 0.08467873586477132,
+          "n_trades": 50,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.3348901947844796,
+          "sharpe": 0.6215567075501345,
+          "dsr": 0.05615871498414856,
+          "dsr_pvalue": 0.9438412850158514,
+          "return": 0.1669550204052912,
+          "max_dd": 0.2425926649805225,
+          "n_trades": 60,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.08496628060413243,
+          "sharpe": -0.291593157960215,
+          "dsr": 0.006828013272159182,
+          "dsr_pvalue": 0.9931719867278408,
+          "return": -0.06496567446731383,
+          "max_dd": 0.1933746053658072,
+          "n_trades": 72,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.10547685784422405,
+          "sharpe": -0.04385303190774091,
+          "dsr": 0.013045759744378084,
+          "dsr_pvalue": 0.986954240255622,
+          "return": -0.003908970230222186,
+          "max_dd": 0.05825300658307936,
+          "n_trades": 31,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.24269351114651366,
+      "fitness_oos_min": 0.08496628060413243,
+      "fitness_oos_max": 0.4454407113532186,
+      "fitness_oos_std": 0.15273583802070415,
+      "sharpe_oos_mean": 0.3066807290989944,
+      "sharpe_oos_min": -0.291593157960215,
+      "robust_score": 0.08496628060413243
+    },
+    {
+      "genome_id": "9f1f7ffa40b4900a",
+      "fitness_is": 0.2653851546757454,
+      "sharpe_is": 0.7183647852569522,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.0,
+      "fitness_oos_std": 0.0,
+      "sharpe_oos_mean": 0.0,
+      "sharpe_oos_min": 0.0,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "aff702f86c30e30f",
+      "fitness_is": 0.22070023288031862,
+      "sharpe_is": 0.49220575419744833,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": -1.3160615555061568,
+          "dsr": 2.0480110912755528e-14,
+          "dsr_pvalue": 0.9999999999999796,
+          "return": -0.004197667464114874,
+          "max_dd": 0.004197667464114874,
+          "n_trades": 1,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.0,
+      "fitness_oos_std": 0.0,
+      "sharpe_oos_mean": -0.3290153888765392,
+      "sharpe_oos_min": -1.3160615555061568,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "43236904205071d5",
+      "fitness_is": 0.22070023288031862,
+      "sharpe_is": 0.49220575419744833,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.0,
+          "sharpe": -1.3160615555061568,
+          "dsr": 2.0480110912755528e-14,
+          "dsr_pvalue": 0.9999999999999796,
+          "return": -0.004197667464114874,
+          "max_dd": 0.004197667464114874,
+          "n_trades": 1,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": 0.0,
+          "dsr": 0.0,
+          "dsr_pvalue": 1.0,
+          "return": 0.0,
+          "max_dd": 0.0,
+          "n_trades": 0,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.0,
+      "fitness_oos_std": 0.0,
+      "sharpe_oos_mean": -0.3290153888765392,
+      "sharpe_oos_min": -1.3160615555061568,
+      "robust_score": 0.0
+    },
+    {
+      "genome_id": "78a445d3043e0771",
+      "fitness_is": 0.20950868586433583,
+      "sharpe_is": 0.2940379509455447,
+      "folds": [
+        {
+          "fold": 0,
+          "fitness": 0.07182313189843528,
+          "sharpe": -0.5971649832941454,
+          "dsr": 0.0029614586254508167,
+          "dsr_pvalue": 0.9970385413745492,
+          "return": -0.10657245144370586,
+          "max_dd": 0.17076725093180228,
+          "n_trades": 14,
+          "test_start": "2022-05-02 12:00:00+00:00",
+          "test_end": "2023-04-02 08:00:00+00:00"
+        },
+        {
+          "fold": 1,
+          "fitness": 0.04914620786914511,
+          "sharpe": -0.7350195766937807,
+          "dsr": 0.0007433210570883239,
+          "dsr_pvalue": 0.9992566789429117,
+          "return": -0.03472119511680272,
+          "max_dd": 0.060780295013844174,
+          "n_trades": 20,
+          "test_start": "2023-04-02 09:00:00+00:00",
+          "test_end": "2024-03-02 05:00:00+00:00"
+        },
+        {
+          "fold": 2,
+          "fitness": 0.0,
+          "sharpe": 1.2071345821102195,
+          "dsr": 0.14855640183813817,
+          "dsr_pvalue": 0.8514435981618618,
+          "return": 0.12296871148346566,
+          "max_dd": 0.05065940776643054,
+          "n_trades": 9,
+          "test_start": "2024-03-02 06:00:00+00:00",
+          "test_end": "2025-01-31 02:00:00+00:00"
+        },
+        {
+          "fold": 3,
+          "fitness": 0.0,
+          "sharpe": -1.6874561123211371,
+          "dsr": 9.510326955509528e-06,
+          "dsr_pvalue": 0.9999904896730445,
+          "return": -0.03197955364502458,
+          "max_dd": 0.032239270519921044,
+          "n_trades": 7,
+          "test_start": "2025-01-31 03:00:00+00:00",
+          "test_end": "2025-12-31 23:00:00+00:00"
+        }
+      ],
+      "fitness_oos_mean": 0.0302423349418951,
+      "fitness_oos_min": 0.0,
+      "fitness_oos_max": 0.07182313189843528,
+      "fitness_oos_std": 0.03128704497338385,
+      "sharpe_oos_mean": -0.4531265225497109,
+      "sharpe_oos_min": -1.6874561123211371,
+      "robust_score": 0.0
+    }
+  ]
+}
@@ -18,6 +18,7 @@ members = [
    "multi-swarm-coevolutive",
    "multi-swarm-core",
    "strategy-crypto",
+    "strategy-pythagoras",
 ]

 [[package]]
@@ -900,6 +901,7 @@ source = { virtual = "." }
 dependencies = [
    { name = "multi-swarm-core" },
    { name = "strategy-crypto" },
+    { name = "strategy-pythagoras" },
 ]

 [package.dev-dependencies]
@@ -917,6 +919,7 @@ dev = [
 requires-dist = [
    { name = "multi-swarm-core", editable = "src/multi_swarm_core" },
    { name = "strategy-crypto", editable = "src/strategy_crypto" },
+    { name = "strategy-pythagoras", editable = "src/strategy_pythagoras" },
 ]

 [package.metadata.requires-dev]
@@ -1980,6 +1983,27 @@ requires-dist = [
    { name = "pyarrow", specifier = ">=18.0" },
 ]

+[[package]]
+name = "strategy-pythagoras"
+version = "0.1.0"
+source = { editable = "src/strategy_pythagoras" }
+dependencies = [
+    { name = "multi-swarm-core" },
+    { name = "nicegui" },
+    { name = "pandas" },
+    { name = "plotly" },
+    { name = "pyarrow" },
+]
+
+[package.metadata]
+requires-dist = [
+    { name = "multi-swarm-core", editable = "src/multi_swarm_core" },
+    { name = "nicegui", specifier = ">=3.11.1" },
+    { name = "pandas", specifier = ">=2.2" },
+    { name = "plotly", specifier = ">=5.24" },
+    { name = "pyarrow", specifier = ">=18.0" },
+]
+
 [[package]]
 name = "tenacity"
 version = "9.1.4"
Author	SHA1	Message	Date
Adriano Dal Pastro	b8aa00b883	fix(llm): inject system prompt into user message for OpusAgent OpusAgent topic system_prompt is used for summarization context but not enforced as a binding instruction to Claude. Prepend [SYSTEM] block to the first message of each session so Claude sees the full JSON schema and grammar specification. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-26 09:42:26 +00:00
Adriano Dal Pastro	ba2c23e602	feat(llm): add session_id for retries and summarize for run tracking HypothesisAgent now opens an OpusAgent session on first attempt and reuses it for parse-error retries — Claude sees full conversation context instead of re-stuffing the prompt. Sessions are closed after propose() completes. All completions set summarize=true so topics accumulate incremental summaries of GA activity. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-26 09:18:11 +00:00
Adriano Dal Pastro	646c64dacd	feat(llm): migrate from OpenRouter to OpusAgent Replace OpenRouter (openai SDK) with OpusAgent REST API (httpx + polling). LLMClient now creates topics per system prompt, submits async requests, and polls for completion. Model tiers map to Claude model IDs (opus-4-7, sonnet-4-6, haiku-4-5) configurable via env vars. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-26 09:09:54 +00:00
Adriano Dal Pastro	a702b2090d	chore(smoke): expose RUN_NAME via PYTHAGORAS_SMOKE_RUN_NAME env var	2026-05-20 15:19:13 +00:00
Adriano Dal Pastro	c6bf0f31cc	feat(prompt_library): v3.2 custom_indicators_spec — fix LLM parse rate on Pythagoras indicators	2026-05-20 10:00:31 +00:00
Adriano Dal Pastro	68e0b009e9	feat(strategy_pythagoras): GA smoke-test runner (BTC+ETH dual run + invariance bonus)	2026-05-19 14:16:07 +00:00
Adriano Dal Pastro	4baa1eca62	feat(strategy_pythagoras): NiceGUI dashboard with 4 tabs (Genomes/Patterns/Ratios/Invariance)	2026-05-19 14:09:26 +00:00
Adriano Dal Pastro	035cd1dff3	feat(strategy_pythagoras): port frontend data layer + invariance/candle helpers	2026-05-19 14:00:33 +00:00
Adriano Dal Pastro	b8bf0c186c	feat(strategy_pythagoras): invariance bonus callback (BTC<->ETH corr_signal) corr_signal: frazione di entries A con match in B entro +/-tolerance_bars (default 36 barre = 3h su 5m TF, env GA_INVARIANCE_TOLERANCE_BARS). apply_invariance_bonus: fitness * (1 + alpha * invariance_score), alpha=0.3 (env GA_INVARIANCE_ALPHA). Spec plan Pythagoras §4. GA integration approach (Step 5 findings): - multi_swarm_core.ga.fitness.compute_fitness e compute_combined_fitness NON espongono callback/hook per post-processing della fitness. - orchestrator.run.run_phase1(...) -> str ritorna solo il run_id; le evaluations (incl. fitness scalare) vengono persistite via repo.save_evaluation dentro al loop GA (run.py:264-277). - I winner sono recuperati dopo il run con repo.list_evaluations(run_id) e ri-ordinati per fitness (vedi pattern run.py:302-310 per WFA re-eval). - Pattern (b) confermato: il runner Task 6.1 chiamera' run_phase1 due volte (BTC, ETH), poi per ogni coppia di evaluations matchera' i genome_id, calcolera' corr_signal sulle entries dei rispettivi backtests e applichera' apply_invariance_bonus per ri-rankare esternamente i winner. Nessuna modifica a multi_swarm_core necessaria in questo task. Tests: 6/6 PASS (perfect alignment, no overlap, within tolerance, bonus formula, alpha=0, zero entries).	2026-05-19 13:58:39 +00:00
Adriano Dal Pastro	af68bc44b4	feat(strategy_pythagoras): port paper-trading backend (Portfolio, Executor, Repository)	2026-05-19 13:55:23 +00:00
Adriano Dal Pastro	074ebe0379	feat(strategy_pythagoras): port DB schema from strategy_crypto	2026-05-19 13:53:08 +00:00
Adriano Dal Pastro	7f2db19a7c	feat(strategy_pythagoras): prompts.json with 7 Pythagoras-aligned cognitive styles	2026-05-19 13:51:32 +00:00
Adriano Dal Pastro	369a77b5cf	feat(strategy_pythagoras): implement candle_pattern, pythagorean_ratio, fractal_mirror + register in compiler	2026-05-19 13:38:40 +00:00
Adriano Dal Pastro	2aa5646aeb	feat(protocol): validate arity + semantics of 3 Pythagoras indicators	2026-05-19 13:29:16 +00:00
Adriano Dal Pastro	6a9e2c28b1	feat(protocol): register 3 Pythagoras indicators in KNOWN_INDICATORS	2026-05-19 13:26:40 +00:00
Adriano Dal Pastro	37558a34f5	feat(strategy_pythagoras): scaffold workspace member + register in uv	2026-05-19 13:26:01 +00:00
Adriano Dal Pastro	14f476dd09	docs(strategy_pythagoras): spec + plan + summaries dei PDF Base per l'esecuzione del sub-project strategy_pythagoras: - docs/superpowers/specs/2026-05-19-strategy-pythagoras-design.md (spec) - docs/superpowers/plans/2026-05-19-strategy-pythagoras.md (14 task, 72 step TDD) - src/strategy_pythagoras/Pythagoras/.summary.md (riassunti numerici dei 2 PDF) - src/strategy_pythagoras/Pythagoras/_extracted/.txt (estrazione testo grezzo via pypdf) I PDF stessi non vengono committati (vedi .gitignore).	2026-05-19 13:03:09 +00:00
Adriano Dal Pastro	05c7b5e89e	test(strategy_crypto): non hard-codare hash dei winner shippati L'assert su btc_fb63e851.json/eth_facd6af85d5d.json era diventato stale dopo i swap dei paper winner (commit `8b767da`, `23b7273`). Verifica strutturale (almeno un btc_.json e un eth_.json) evita il fail ad ogni futuro swap di winner. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-18 17:25:27 +00:00
Adriano Dal Pastro	6655e425fa	fix(paper): ETH 5m allineato al tick + hardening GUI/compose Bug principale: in scripts/run_paper_trading.py il fetch usava end = now.replace(minute=0,...), troncando sempre all'ora. ETH è dichiarato timeframe=5m (commit `23b7273`) ma di fatto veniva valutato 1 volta ogni 60 min — 502 poll del run 39e027df hanno prodotto solo 43 evaluazioni/asset, tutte a HH:00. Il commento in load_assets segnala esplicitamente che a 1h la strategia perde -33% su 7y: regressione vs backtest. Fix: helper _align_end_to_timeframe(now, timeframe) snappa end al boundary nativo dell'asset. Mappa 1m/5m/15m/30m/1h/4h/1d. Test regression in src/strategy_crypto/tests con 9 casi. Hardening accessorio incluso nello stesso commit: - docker-compose.yml: state/ in RW per strategy-crypto-gui (SQLite WAL richiede SHM writable anche da reader). - multi_swarm_core/dashboard/nicegui_app.py: ui.timer ora deactivate on_disconnect su 3 pagine (index/convergence/genomes) per evitare leak di timer dopo client disconnect. - strategy_crypto/frontend/data.py: retry 5s su sqlite.connect per cold-start race quando GUI parte prima del paper writer. - state/validation-hardened-001.json: output WFA tooling multi-fold del run phase1-hardened-001. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-18 17:04:15 +00:00
Adriano Dal Pastro	23b7273e71	feat(paper): ETH a tick 5m + tooling per-year/per-tick analysis scripts/run_paper_trading.py: AssetConfig ETH ora usa timeframe="5m" invece del default 1h. Il winner c04dff7086 e' stato trovato dal GA su dati 5m e a 1h la strategia perde: - ETH @ 5m (native): +359.50% cum 7y, 77% winrate, max DD/yr 19% - ETH @ 1h (precedente): -33.03% cum 7y, 67% winrate, max DD 74% BTC resta a 1h (winner 238e4812 native a 1h, +104% 7y, Sharpe 2+ in 3 anni). Nuovi script di analisi: - scripts/yearly_strategies.py: breakdown per anno (2019-2025) di 4 strategie su tick di discovery (trade/winrate/return/maxDD/Sharpe). - scripts/multi_tick_strategies.py: confronto cross-tick (5m/15m/1h) per i 2 winner correnti. Documenta la divergenza tick-paper di ETH. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-16 22:10:38 +00:00
Adriano Dal Pastro	9c871d1d86	feat(validation): WFA tooling + multi-fold results phase1-100 runs Aggiunge 2 script di analisi per validare i top-K genomi cross-fold: - scripts/analyze_btc_winners.py: per-trade dump (wins/losses/winrate/ avg_win/avg_loss/return/maxDD/Sharpe) per ogni top-K × 4 fold expanding-window WFA. Usato per identificare i winner robusti vs i lucky-shot overfit. - scripts/compare_winners.py: cross-run comparison di 5 winner candidate (BTC 1h + ETH 1h + BTC 5m + ETH 5m) sui medesimi 4 fold, con totali cumulativi. Risultati WFA freezati: - validation-btc-100-001.json: BTC 1h baseline (undertrading=10) - validation-btc-100-001-thr3.json: BTC 1h con threshold=3 (rilassato per strategie ultra-selettive) - validation-btc-100-5m-thr3.json: BTC 5m con threshold=3 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-16 21:48:55 +00:00
Adriano Dal Pastro	8b767da5e7	feat(strategy): swap paper a 238e4812 (BTC) + c04dff7086 (ETH) — phase1-100 winners Risultati phase1-{btc,eth}-100-001 (1h) + phase1-{btc,eth}-100-5m-001 (5m, 770K bars) con 100-agent GA × 10 generazioni × fitness v2 hardened + WFA 0.7 split. BTC (238e4812, meteorologist Gen 7 winner del run 1h): - IS f=0.2604, Sharpe=0.437, 25 trades - OOS: f=0.4184, Sharpe=1.296, +22.15% return fold 3, DD 7.4% - WFA 4-fold: cum_ret +31.65%, 58.6% winrate, 2 fold Sharpe 2+ ETH (c04dff7086, mean-reversion oversold del run 5m): - IS f=0.1881, OOS f=0.1789 (no overfit), Sharpe_OOS=0.171 - WFA 4-fold: cum_ret +23.83%, 77.8% winrate, 2 fold @ 100% winrate - Primo ETH winner che sopravvive hard-kill v2 in tutta la storia progetto Archive: btc_9cf506b8.json (precedente winner hardened-001), eth_facd6af85d5d.json. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-16 21:48:42 +00:00
Adriano Dal Pastro	1c0058ec3b	feat(strategy): swap BTC paper strategy a 9cf506b8 (hardened-001 robust winner) Sostituisce btc_fb63e851 (in produzione dal 13 maggio, net +$9.9k su 7y BTC, Sharpe medio -0.20) con btc_9cf506b8, vincitore del run phase1-hardened-001 (K=50 G=15 7y, fitness-v2 con hard-kill su fees_eat_alpha + negative_net_pnl). Performance comparate su 7y BTC continui: - btc_fb63e851: Net +$9,951 Sharpe medio -0.20 4/8 anni positivi - btc_9cf506b8: Net +$30,538 Sharpe medio +0.31 5/8 anni positivi Best year 2021: +$18,938 (vs prod +$6,835) Best Sharpe annuale: 2023 +1.27 (vs prod 2024 +1.16) Zero adversarial findings su 7y continui. Performance cross-asset (test su ETH 6.8y): - btc_fb63e851 su ETH: -$120 (pseudo-flat, nessun segnale) - btc_9cf506b8 su ETH: +$2,059 (4/7 anni positivi) La strategia generalizza out-of-asset, indicatore di robustezza non-ETH-overfit. DSL: physicist style, lookback 150 bar (~6 giorni). Triple-AND condition su realized_vol + atr_pct + sma_pct per entry (long o short). Exit su sma_pct=0 o vol collasso (<0.1%). Selettivo: 502 trade in 7y = 1.4 trade/giorno medio. btc_fb63e851 archiviato in strategies/archive/ per consultazione futura. Il glob loader `btc_*.json` e' non-ricorsivo, quindi archive/ non viene caricato automaticamente. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-16 13:21:37 +00:00
Adriano Dal Pastro	220e510d5e	fix(orchestrator): escludi prompt_library dal config_dict JSON PromptLibrary e' un frozen dataclass non JSON-serializable. Quando passata in cfg.prompt_library e poi spread via **cfg.__dict__ in config_dict, faceva fallire repo.create_run() con TypeError al primo run dopo refactor. Fix: filtra cfg.__dict__ escludendo prompt_library, e salva separatamente la lista degli stili (prompt_library_styles) per reproducibility. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-16 13:21:08 +00:00
Adriano Dal Pastro	9742df3a1f	fix(fitness): hardening anti-overfit post-7y incident — 3 correzioni Incident: extended run elite (Sharpe IS 1.93) net-negativo su 7y continuous (fees=101% del gross). Multi-fold validation NON sufficiente: ogni fold restarta equity, mascherando accumulo fees compound. Correzioni: 1) Default --start esteso a 2018-09-01 (7.3 anni) - Copre bear 2018-19, halving 2020, COVID, ATH 2021, winter 2022, ETF rally 2024, regime corrente. - Una finestra corta (2y) lasciava il GA libero di overfit single-regime. 2) fees_eat_alpha promosso a hard-kill in fitness v2 - Da soft-penalty 0.4x a hard-kill 0 fitness. - Una strategia con fees > 50% del gross non e' recuperabile via selection: il prodotto del GA non puo' deployare con quel cost burden. 3) Nuovo finding negative_net_pnl (HIGH, hard-kill) - Fires quando sum(trade.net_pnl) < 0 sul training window. - Cattura: gross negativo (no edge direzionale) E gross positivo ma fees > gross (edge insufficiente). - Sintesi del net-after-fees su finestra continua come "deal-breaker" non negoziabile via soft penalty. CLI: - --fitness-hard-kill <csv> per override esplicito. - Default v2: no_trades,degenerate,undertrading,fees_eat_alpha,negative_net_pnl. Test: - 252 pass (251 + 1 nuovo: test_negative_net_pnl_fires_on_negative_gross). - Test e2e WFA aggiornato: passa fitness_hard_kill_findings=('no_trades',) perche' il fixture sintetico non produce strategie profittevoli. - test_no_findings_on_reasonable_strategy rinominato: test_rsi_mean_reversion_loses_money_on_synthetic_data (riflette semantica reale: RSI mean-rev su synthetic ohlcv ha net negativo). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-16 11:07:40 +00:00
Adriano Dal Pastro	21b5cf1eae	feat(validation): add validation data for phase1-extended-001 with detailed results	2026-05-16 10:55:35 +00:00
Adriano Dal Pastro	a29748e3d8	perf(backtest): vectorize engine + parallel LLM propose + multi-fold validation tool - backtest/engine.py: state machine numpy invece di pd.iterrows() - 16.8x speedup su 2y (470ms -> 28ms), 11.3x su 7y (1744ms -> 154ms) - 7 parity test parametrici vs reference iterrows assicurano equivalenza - orchestrator/run.py + run_phase1.py: --llm-concurrency N - ThreadPoolExecutor parallelizza hypothesis_agent.propose() per generazione - 5-8x speedup wall time GA loop (OpenRouter qwen-2.5 regge 6-10 concorrenti) - default 1 = backward-compat sequenziale - scripts/validate_run.py: validation multi-fold standalone - prende run_id + top-K + N-folds expanding-window su dataset esteso (7y) - rivela overfitter mascherati da fitness IS alta (vedi phase1-extended-001: elite IS Sharpe 1.93 collassa OOS a -1.00) - ranking per robust_score = min(fitness_oos) su tutti i fold Test 250/250 pass. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-16 10:53:48 +00:00
Adriano Dal Pastro	fa11cca2bc	docs(readme): allinea a stato finale post-refactor sessione 2026-05-15 Aggiornato per riflettere i 12 commit di refactor della sessione: - Layout: 3 servizi Docker (strategy-crypto-paper + strategy-crypto-gui + multi-swarm-core-gui), GUI split (core GA / strategy paper) - Test: 238 attesi (era 186) - Architettura prompt: SYSTEM compositor v3.2 documentato (9 sezioni iniettate da scaffold core + contenuto strategy via prompts.json) - Grammar: 8 indicatori (5 base + 3 _pct: atr_pct, sma_pct, macd_pct) - Input MarketSummary: 14 metriche (5 base + 4 regime + 5 geometria) - 7 stili cognitive in prompts.json (era 6, +military_strategist e psychologist), con focus_metrics + lookback consigliato + archetipo - Deploy: 2 URL distinti (/multi_swarm_core_gui + /strategy_crypto_gui), Traefik replacepathregex (NON stripprefix per evitare doppio root_path) - Note operative: chown 1000:1000 anche per strategies/ post git mv - 3 regression guard test ASCII/archetype/lookback documentati Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-15 22:05:06 +00:00
Adriano Dal Pastro	6526c6e6e3	refactor(prompts): strategy_crypto v3.2 — consolidamento + 3 invariant tests Patch di consolidamento post-diagnosi v3.1, divisa in REGRESSIONI ripristinate + MIGLIORAMENTI opportunistici + INVARIANTI permanenti. Regressioni v3.0 -> v3.1 ripristinate: 1. ASCII-strict: caratteri circa-uguale (U+2248) sostituiti con 'vicino a' in 3 stili (historian, engineer, psychologist). Critico per encoding robustness. 2. 'Archetipo dominante: <metafora>': chiusa identitaria reintrodotta in tutte le 7 directive. Ancora semantica resistente a mutate_prompt_llm. 3. 'Lookback consigliato: X-Y barre': hint range differenziato per stile (physicist 150-300, biologist 80-200, historian 200-500, meteorologist 50-150, engineer 60-120, military 100-200, psychologist 50-120) per orientare il parametro evoluto lookback_window del genoma. Miglioramenti opportunistici: 4. Voce attiva rinforzata: +verbi generativi ('costruisci', 'combina', 'cattura', 'diagnostica', 'preferisci') in tutte le directive 5. anti_patterns 5 -> 7 voci: aggiunti (6) chattering crossover same-type same-lookback, (7) soglie hard senza isteresi entry/exit 6. output_priorities 4 -> 5 voci: aggiunta in cima (#1) 'coerenza con lente cognitiva' (fondamento del design swarm) 7. domain_warnings: +frase 'seasonality > 0 non significa significativa, gate a 0.05' 8. NEW _design_invariants metadata: documenta gli invarianti che future versioni DEVONO preservare (utile per chi edita + mutate_prompt_llm) NEW invariant tests (regression guards permanenti): - test_strategy_crypto_directives_ascii_safe - test_strategy_crypto_directives_have_archetype_marker - test_strategy_crypto_directives_have_lookback_hint Statistiche v3.2: - directive: 800-950 char (era 545-614 in v3.1, troppo snellite) - physicist 890, biologist 867, historian 887, meteorologist 884, engineer 904, military_strategist 898, psychologist 909 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-15 21:54:13 +00:00