GOGA methodology
How leaderboard rank_score is computed and where the underlying data comes from. This is a transparency note (L4–L5), not marketing.
How GOGA works — signal to verdict
Every strategy runs through a deterministic, reproducible pipeline. No step is a black box — each one is auditable, and the same inputs always produce the same verdict.
Deterministic safety gate
A rule-based gate — not an LLM — decides whether a strategy may touch real money. Leverage above the live cap or weak out-of-sample behaviour is held to paper. Same inputs → same verdict.
Rank score — what the leaderboard weighs
The composite rank_score blends risk-adjusted return with downside protection — not raw PnL. Illustrative weighting (authoritative formula below):
Why this is more honest than a typical backtest
GOGA
- Walk-forward / out-of-sample check — overfit strategies are flagged, not promoted.
- Full-span candles per coin, so PnL isn't a lucky 3-week window.
- Overfitting passports — strategies that decay over time are marked decaying.
- Leverage cap enforced in the gate, not just in a disclaimer.
- Source labels separate simulated PnL from funded accounts.
- Deterministic gate — reproducible, auditable verdicts.
Typical "we made +400%" pitch
- In-sample only — curve fit to the past, fragile live.
- Cherry-picked short history.
- No overfit check; decay hidden.
- Headline leverage with the risk in fine print.
- Paper and real results blended together.
- Opaque "secret sauce" you can't audit.
Past performance never guarantees future results — but a method you can inspect beats a number you can't.
GOGA safety layers · public read-only
Safety & governance snapshot
Aggregated from AI OS /api/v1/trading/safety/*. No user accounts, ledger, confirmations, or execution controls on this surface.
Layer status
- Trust Ledger active
- Action Ladder active
- Paper-to-Live active
- Strategy Passport active
Action Ladder
- Read
- 1
- Explain
- 2
- Simulate
- 2
- CONFIGURE
- 1
- LIVE_EXECUTE
- 1
- PAPER_EXECUTE
- 1
- SPEND
- 1
Strategy Passport
- Passports
- 10
- Decaying
- 0
- backtested · 10
Rank score (composite)
The public leaderboard default sort uses a composite rank_score in the 0–100 range (display may show decimals). The portal applies the same weighting as documented in the GOGA relay contract:
rank_score ≔ clamp( Sharpe × 0.4 + pnl_pct × 0.3 + (1 − |max_drawdown_pct|/100) × 0.3 , 0, 100 ) — normalized inputs per relay contract; weights may change in methodology v2.
Weights are subject to change when methodology v2 ships; the tooltip on the leaderboard links here for the current description.
Data sources
- AI OS read relay — strategy cards and detail payloads are served from GET /api/v1/portal/goga/leaderboard and GET /api/v1/portal/goga/strategy/{id}. The portal does not call exchanges or custody providers directly.
- Source labels — each row carries a source discriminator (for example backtest_db, analyzer snapshot, or leaderboard_snapshot) so simulated PnL is not confused with funded accounts.
- Timestamps — as_of and data-quality fields on each card describe snapshot freshness; when upstream is unavailable, the UI shows an explicit empty or degraded state.
Кратко (RU)
Рейтинг rank_score — композитный показатель (Sharpe, PnL%, просадка). Данные приходят из read-only API AI OS; портал не исполняет сделки и не хранит средства пользователей.