Methodology

Each top-level selector is a canonical strategy identity; linked tracker ledgers are nested and appended only after validated historical cutoffs.

Every appended segment carries explicit lineage and accounting metadata; overlapping canonical rows are never overwritten.

Daily PnL is ordered by evidence date. Cumulative PnL is the arithmetic running sum of known rows only; unknown rows never update equity.

Multi-horizon metrics use known evidence sessions, excluding unknown rows before taking the last 30 and last 10 sessions. Green-calendar rate counts flat and inactive known sessions as not green; green-active rate uses only sessions with a trade or nonzero PnL.

The descriptive recency blend is 50% lifetime, 30% last-30-session, and 20% last-10-session PnL per session. Its overlapping windows are not independent validation and the blend cannot select or promote a strategy.

Daily drawdown is cumulative equity minus the running peak, with the initial peak fixed at zero.

Non-finite source values are normalized to null; strict JSON serialization rejects non-finite numeric literals.

Source hashes are SHA-256 digests of the exact files consumed by each strategy export.

Apply one-position capacity independently to each configured sleeve.

Restrict both sleeves to their common eligible calendar.

Concatenate candidates and apply global one-position capacity.

Apply the causal HIGH_VOL gate; trailing health state is diagnostic only and never an entry filter.

Historical and selected-current-regime results are research evidence, not forecasts.

Static export has no broker connection, credentials, mutation endpoint, or order path.

Passing any future monitoring gate does not change authorization in this dashboard.

Raw freshness remains available as date-lag metadata, but monitoring posture is authoritative in the UI: current forward tracker, current via an exact tracker proxy, or frozen research evidence. A frozen cutoff is not an alpha-decay or profitability judgment.