Detail
–
Strategy Research Pipeline
RESEARCH LAB
Fake alpha dies here. Click anything.
–
Combos scanned
–
Shortlisted
–
Survived gate
–
Killed by gate
Layer 1 SCAN vectorbt – LIVE
→
Layer 2 OPTIMIZE Optuna – LIVE
→
Layer 3 ⚔️ REALITY GATE hftbacktest – LIVE
→
Layer 4 MEMORY GraphRAG not wired FUTURE
→
Layer 5 RESEARCH AutoResearchClaw sandbox only SANDBOX ACTIVE
🛰 AutoResearchClaw — Research Launcher SANDBOX ACTIVE Never touches execution · outputs to reports/
Ready
Sessions to run:
1 session · select presets below
🔬
Microstructure
Fill modeling, queue position, latency
⚙️
Execution Costs
Realistic crypto perpetuals costs
📊
Strategy Compare
Reversion vs breakout vs momentum
🌊
Regime Detection
EMA, HMM, volatility clustering
💰
Funding Carry
Academic + practitioner funding research
☠️
Failure Analysis
Backtest-to-live divergence post-mortems
🐙
GitHub Scout
Recent open-source strategy improvements
✏️
Custom Topic
Type your own research question
Select at least 1 preset to launch
⚔️ Realism Gate — Where Fake Alpha Goes to Die LIVE DATA
0.16%/side · 3bps slip · 1-bar lag · 85% fill –
–
Survived
–
Killed
–
Tested
–
Survival rate
Pass threshold: Sharpe ≥ 1.0 after realistic execution = threshold line on bars –
Gate Rules LIVE
✓PASS: realistic Sharpe≥ 1.0
✗FAIL: degradation from ideal → realistic> 70%
✗BLOCKED: trade count below minimum< 20 trades
✗WARN: statistical reliability threshold< 30 trades
⚙Fill model: fee per side / slippage / latency / fill rate0.16% / 3bps / 1bar / 85%
Loading…
Candidates LIVE DATA
–
SCAN:
–
Strategy Sharpe ▼ Monthly DD WR Trades Status
Loading…
Research Feed LIVE
Active
research-lab — live
Initializing…
Best Optimized Params LIVE DATA
– –
No optimization run yet
⚠ Data Quality LIVE DATA
–
High Sharpe with tiny samples is a con artist in a suit. These candidates are flagged:
–
Rule: <30 trades → statistically meaningless. <20 → blocked from gate. Sharpe is a liar at small N.
🧠 Failure Intelligence PARTIAL — GraphRAG FUTURE
Live
Known Failures (from Ledger)
RSI mean reversion — slippage/fee sensitive
Colosseum EA — 133k gens, zero fitness signal
80%+ WR target — wrong fitness metric
Tiny-sample Sharpe — misleading, discard
ATR breakout — survived gate (Sharpe 3.21 real)
Failure Memory ● LIVE
Similar failures found–
Dominant failure mode–
Risk patterns–
Novelty vs repetition–
Click a candidate row to see failure intelligence
Upgrade Stack STATIC
Click any repo for path, status, integration details
📦 Backtesting Data Inventory LIVE DATA Local disk · gitignored · never touches execution
Loading…
–
Total OHLCV
–
Symbols
–
Timeframes
–
OHLCV files
–
Research studies
–
Backtest results
Loading inventory…
Hard Rules — DO NOT DO STATIC
No Exceptions · Ever
  • ✕Auto-edit production without review
  • ✕Bypass the governor
  • ✕Skip the realism gate before paper
  • ✕Let research tools touch execution
  • ✕Merge experimental paths into live
  • ✕Use org tools as execution engines
  • ✕Optimize only for headline return
  • ✕Trust tiny samples (<30 trades)
  • ✕Promote without operator GO
  • ✕Hide stale data or silent failures
  • ✕Use AutoResearchClaw for live signals
  • ✕Claim improvement without evidence
COMPARE:
Select 2–4 candidates from the table above