MODEL ADOPTION #4140: REJECT — LEDGER BENCH: Apply filter: Session = OFF_SESSION — improves win rate from 32.6% to 42.5% — E8 Intelligence Research
- 1Independent Researcher, United Kingdom
Abstract
DECISION: REJECT ELEMENT: No single rule survives — the OFF_SESSION filter is a proxy for signal-source quality (wkjudas underperformance), not a session-time edge. Adopting it would mask the real confound and reduce trade frequency without isolating the causal driver. EXPECTED IMPACT: Win-rate gain is likely spurious (+10pp on n=127, with 3/10 live window showing filter removes all wkjudas — a source bias, not session effect). Absolute PnL would likely drop due to fewer trades, and the filter would not survive regime shifts (e.g., high-vol news sessions where OFF_SESSION entries are actually worse). RISK: Overfitting to summer low-vol regime; discarding wkjudas entirely without testing its standalone quality; session filter may correlate with E8 geometry phase (e.g., OFF_SESSION often aligns with non-optimal God's Clock windows), creating a hidden timing conflict. TEST: Before any adoption, run a 90-day replay-bench on historical tick data (2023–2025) with the following exact
Author: Andrew Stewart Caldin, Independent Researcher, UK. Part of the E8 Intelligence Research series. Platform: e8intelligence.com
Related
- MODEL ADOPTION #4465: REJECT — LEDGER BENCH: Apply filter: Session = OFF_SESSION — improves win rate from 31.5% to 41.9% — E8 Intelligence Research
- MODEL ADOPTION #4171: REJECT — LEDGER BENCH: Apply filter: Session = LONDON_OPEN — improves win rate from 32.6% to 36.2% — E8 Intelligence Research
- MODEL ADOPTION #4639: REJECT — VIDEO SCOUT: ICT Concepts - London Session Raids 2:00am to 5:00am, New York Local Time — E8 Intelligence Research
- MODEL ADOPTION #4606: REJECT — LEDGER BENCH: Apply filter: Min confirmations >= 6 — improves win rate from 31.6% to 35.8% — E8 Intelligence Research
- MODEL ADOPTION #4136: REJECT — LEDGER BENCH: Apply filter: Phase distance TIGHT (<=2deg) — improves win rate from 32.6% t — E8 Intelligence Research
BibTeX
@misc{caldin2026model,
title = {{MODEL ADOPTION #4140: REJECT — LEDGER BENCH: Apply filter: Session = OFF_SESSION — improves win rate from 32.6% to 42.5% — E8 Intelligence Research}},
author = {Andrew Stewart Caldin},
howpublished = {Zenodo},
year = {2026},
month = oct,
doi = {10.5281/zenodo.23409986},
url = {https://zenodo.org/records/23409986},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - MODEL ADOPTION #4140: REJECT — LEDGER BENCH: Apply filter: Session = OFF_SESSION — improves win rate from 32.6% to 42.5% — E8 Intelligence Research
AU - Andrew Stewart Caldin
PY - 2026
DA - 2026/10/16
PB - Zenodo
DO - 10.5281/zenodo.23409986
UR - https://zenodo.org/records/23409986
AB - DECISION: REJECT ELEMENT: No single rule survives — the OFF_SESSION filter is a proxy for signal-source quality (wkjudas underperformance), not a session-time edge. Adopting it would mask the real confound and reduce trade frequency without isolating the causal driver. EXPECTED IMPACT: Win-rate gain is likely spurious (+10pp on n=127, with 3/10 live window showing filter removes all wkjudas — a source bias, not session effect). Absolute PnL would likely drop due to fewer trades, and the filter would not survive regime shifts (e.g., high-vol news sessions where OFF_SESSION entries are actually worse). RISK: Overfitting to summer low-vol regime; discarding wkjudas entirely without testing its standalone quality; session filter may correlate with E8 geometry phase (e.g., OFF_SESSION often aligns with non-optimal God's Clock windows), creating a hidden timing conflict. TEST: Before any adoption, run a 90-day replay-bench on historical tick data (2023–2025) with the following exact Author: Andrew Stewart Caldin, Independent Researcher, UK. Part of the E8 Intelligence Research series. Platform: e8intelligence.com
ER -