MODEL ADOPTION #3995: REJECT — LEDGER BENCH: Apply filter: Min confirmations >= 5 — improves win rate from 32.5% to 34.0% — E8 Intelligence Research

DECISION: REJECT ELEMENT: No single rule survives — the +1.5pp win-rate gain is statistically fragile (n=1813, p≈0.08) and the filter contradicts E8's core timing logic, which relies on early confirmation counts as a *dynamic* signal, not a static threshold. EXPECTED IMPACT: Win-rate gain likely evaporates in live conditions because the filter removes high-volatility breakout entries that carry the model's fat right tail (TP2 payoffs). Net PnL impact: -1% to +0.5% monthly, with higher variance due to missed momentum. RISK: Overfits to recent ranging regime; in fast trends, confirmations lag, causing systematic late entries or missed reversals — degrading the FADE polarity edge that depends on entering *before* full confirmation. TEST: Must pass a 12-month historical replay with regime segmentation (trend vs. range). Specifically: bucket all signals by confirmation count at entry (0–4 vs 5+), then require bucket B (5+) to show ≥10% higher R-multiple per trade *and* lower max dra Author: Andrew Stewart Caldin, Independent Researcher, UK. Part of the E8 Intelligence Research series. Platform: e8intelligence.com