Quality
Expectancy per trade
Average realized or replayed return per trade after documented cost assumptions. Public values should be delayed or simulated unless explicitly reviewed.
Metrics catalog
Sanitized measures for public evidence.
Metrics published here must be delayed, aggregate, simulated, replay-based, shadow-mode, or manually summarized. They must not expose account identifiers, open positions, current order intent, credentials, or raw broker responses.
Quality
Win rate and payoff ratio
Win/loss shape by lane, regime, confidence bucket, volatility bucket, and action family where sample size supports summary.
Risk
Drawdown
Maximum drawdown, drawdown duration, or ulcer-style pain metrics, preferably compared against baseline.
Risk adjusted
Sharpe / Sortino / DSR
Risk-adjusted return indicators, including deflated Sharpe or comparable multiple-testing guard when optimization searched many candidates.
Behavior
Action distribution
OPEN, CLOSE, HOLD, and skip distribution movement by target and non-target pockets.
Audit
Metadata completeness
Share of decisions with reconstructable experiment id, gate state, matched pocket, score before/after, final action, and reason/source.
Required fields for public metric artifacts
as_oftimestamp or date.- Source class:
simulation,replay,shadow,live_delayed, ormanual_summary. - Delay/sanitization note.
- Methodology or experiment reference.
- Reviewer/owner when available.
Current publication state
This MVP defines the public metrics catalog. Machine-readable sanitized metric artifacts are tracked as a separate public-showcase slice so the portal can publish them only after validation rules prevent leakage.