⚠ Resolver hardening in progress — treat WR/PF as provisional.
Resolution uses intrabar first-touch (SL-first conservative ordering) with reverse-split / stale-entry drift-guards. Known gap (2026-06-10 code review): the production resolver's intrabar cache fetches only recent bars, so picks older than that window fall back to a close-only approximation (tagged resolution_method=close_approx) — being hardened to entry-anchored fetch. The fully-accurate intrabar-true per-class truth lives in intrabar_truth_by_class.json (currently 0/6 money-ready). Tier badges need ~7d post-fix + n≥100.
Performance by asset class
Loading…
Performance by persona / strategy
Each AI uses multiple personas (e.g., growth_at_reasonable_price, mean_reversion_pullback). Personas are pre-registered strategies — see methodology.