MODEL ADOPTION #987: REJECT — LEDGER BENCH: Apply filter: Quarterly phase = MANIPULATION — improves win rate from 31.6% — E8 Intelligence Research

DECISION: REJECT ELEMENT: The +0.6pp win-rate improvement from n=1012 is statistically fragile — a 0.6pp delta over a 31.6% baseline yields a Cohen's h of ~0.013, far below the 0.2 threshold for a "small" effect. The confidence of 62 is not sufficient to override the risk of phase-dependent volatility compression. MANIPULATION quarters often coincide with lower ATR and wider spreads, which would erode the already thin edge from win-rate alone. The filter also rejects trades in non-MANIPULATION quarters — the backtest does not isolate whether the filter's benefit is due to avoiding bad trades or merely shifting the trade set to a different volatility regime. Without a controlled A/B on PnL per trade (not just win rate), the edge is unproven. EXPECTED IMPACT: Negligible to negative. Win-rate might tick up 0.3–0.5pp in MANIPULATION quarters, but average win size likely drops 1–2% due to lower volatility, and the filter will reject 40–50% of all signals (since only 1 of 4 quarters is MAN Author: Andrew Stewart Caldin, Independent Researcher, UK. Part of the E8 Intelligence Research series. Platform: e8intelligence.com