LAB #4956 NEUTRAL: LEDGER BENCH: Apply filter: Session = LONDON — improves win rate from 31.7% to 34.8% (+3.1pp, n=23) — E8 Intelligence Research
Year: 2026
Type: preprint
Abstract: IDEA: Adjustment made to the live model — verify against the backtests: Apply filter: Session = LONDON — improves win rate from 31.7% to 34.8% (+3.1pp, n=23) SAME-WINDOW EFFECT: The London filter would have blocked almost nothing in the live window, because every logged entry is timestamped 10:25–11:45 CET — squarely inside London hours. So the three closed losses (EURUSD, XAUUSD, AUDUSD) and the two closed wins (USDCHF, USDCAD) all survive the filter unchanged; the win rate over this window stays 2/5 = 40% either way. The filter's claimed +3.1pp is therefore untested by this evidence — the live window is a single-session sample that cannot discriminate. Net PnL delta from applying it here: zero. OVER-TIME PROJECTION: The +3.1pp on n=23 is statistically hollow — the 95% CI on that delta spans roughly −15pp to +21pp, i.e. indistinguishable from noise, and n=23 is far too small to justify a live model change. Worse, the GTX 30-day distribution (140 TP2 wins vs 292 losses, ~32% win rate Author: Andrew Stewart Caldin, Independent Researcher, UK. Part of the E8 Intelligence Research series. Platform: e8intelligence.com
Author: Caldin, Andrew Stewart
Language: English
Subfield: Finance
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7225007680