The Edge That Wasn't
All 81 experiments › #37

#37 — FIFO-Trends proper live simulation (v2)

2026-05-21 (Session 068)
Verdict: positive · ✅ positive — edge recovered under S5-monitor fill model
A redesigned backtest modeling 5-second exit monitoring (fill ≈ trail level ± 0.3p) showed the original FIFO-Trends configs were never broken — the live service's M5 polling was.

Hypothesis — what it tests

Does S5-resolution exit monitoring (submit market order within ~5s of the trail trigger) recover the FIFO-Trends edge that M5 bar-close management destroyed?

Description

Follow-up to the bar-close fill correction (Entry 36). The insight: a market order fills at the price prevailing at detection time, so fill quality depends on detection frequency. With M5 polling, detection lags up to 5 minutes and fills land at bar close (3-10 pips past the trail); with S5 (5-second) polling, S5 bars are only 0.1-0.5 pip wide, so fills land essentially at the trail level — equivalent to an OANDA broker-side trailing stop. The v2 backtest models this explicitly: trail exits fill at trail level (S5_MONITOR mode), X7 column-count exits fill at M5 bar close, with an M5_MANUAL reference mode showing the broken design. Under the realistic S5 model the original d=1 (5-pip trail) configs recovered strong out-of-sample results, and the conclusion was that the live service needed a separate S5 exit-monitoring loop.

Key result

S5-monitor OOS: GBP_JPY d=1 135.3 p/d (71% WR), USD_JPY 138.9 (70%), EUR_JPY 73.7 (68%), GBP_USD 53.9 (86%); same configs under M5-manual: −94.6 / −59.5 / −49.9 / −19.2 p/d; d=1 optimal for all pairs

Indicators

P&F (Point & Figure)trailing stop (d-box trail)column-count SMA (X7 exit)spread gate (IS P90)

Algorithms

Numba JIT backtest kernelexplicit fill-model simulation (S5_MONITOR vs M5_MANUAL)walk-forward validationtrail-distance grid sweep

Code

research/experiments/fifo_trends/backtest_fifo_pnf_v2.pyresearch/experiments/fifo_trends/results/fifo_pnf_v2_proper_sim.csv

Interactive version (search, filters, figures): the experiment explorer.

← #36 FIFO-Trends bar-close fill correction#38 FIFO-Live S5 exit monitoring redesign →
This page is one experiment from the audited record behind the book The Edge That Wasn't — get the book · about the project.