Brummel
51c87e5a91
Stage-1 "R-based signal quality" cycle spec (milestone #119/#126/#127/#128/#129, contract C10). Signal quality measured in R on a strategy's unsized bias stream, feed-forward (no equity feedback): bias -> stop-rule -> position-management -> per-trade R-outcomes -> summarize_r fold -> E[R]/SQN. Design adversarially hardened (12-juror refute panel; decisions logged on #117): - R is stop-defined and LATCHED at entry (frozen R-denominator, never re-read). - Stop-outs are NOT capped at -1R (honest loss tail; gap-through -> R < -1). - Fills one-cycle-lagged (SimBroker no-look-ahead discipline; RED test mandatory). - First stop is volatility-scaled (VolStop = k * EMA(|d price|)), not constant (a constant makes R just rescaled pips). - position-management emits a dense per-cycle R-record (trade ledger + R-equity + window-end handling), tapped by a Recorder, folded post-run by summarize_r (a post-run fold, NOT an in-graph node). - Sizer = load-bearing flat-1R seam; R computed size-invariantly. Veto = doc seam, not a runtime node (DCE deletes an identity). - exposure -> bias rename lands first (compiler-driven). Grounding-check PASS (11 assumptions ratified by green tests; auto-signed under /boss). refs #117