#38 — FIFO-Live S5 exit monitoring redesign

2026-05-18 → 2026-05-26 (Sessions 064-070; final stop Session 075)

Verdict: mixed · 🟡 mixed — redesign deployed, but live edge never materialized

The live FIFO service was rebuilt with a separate 5-second trail-monitor loop after the first M5-managed deploy lost −96.5p in 3 days; even after redeploy and a timestamp bug fix, GBP_JPY live never approached the backtest.

Hypothesis — what it tests

Can a 5-second S5 trail-monitor thread in the live service close the fill gap (from 3-10p down to ~0.3p) and let the validated FIFO-Trends backtest edge be realized live?

Description

Engineering follow-through on Entry 37. The first fx-fifo-live deploy (GBP_JPY_ft on account 013) managed exits at M5 bar close and lost −96.5 pips over 9 trades in 3 days at 11% WR versus a 71% WR backtest — the 5-minute detection lag inflated every trail loss by 3-10 pips. Both the live and paper services were rewritten with a two-loop architecture: an M5 entry loop (unchanged P&F engine) plus a separate S5 exit loop polling 5-second candles and submitting a market order immediately on trail trigger. Relaunched 2026-05-21, a residual bug (`last_s5_ts` never initialized between trades) let stale pre-entry S5 bars fire immediate exits with 3.3-7.7 pip gaps — 5 trades, 0% WR, −39.0p — fixed 2026-05-24. Ultimately the strategy was stopped for good on 2026-05-26 (15 trades, −168.5p, 7% WR total): the 1-box 5-pip trail was structurally too tight for GBP_JPY's ~3-pip live spread even with correct S5 fills.

Key result

First deploy: 9 trades, 11% WR, −96.5p in 3 days (vs +71.6 p/d OOS); last_s5_ts bug: 5 trades, 0% WR, −39.0p with 3.3-7.7p fill gaps; final live total: 15 trades, −168.5p, 7% WR — stopped 2026-05-26

Indicators

P&F (Point & Figure)trailing stop (1-box trail)column-count SMA (X7 exit)

Algorithms

two-loop live service architecture (M5 entry + S5 exit polling)live vs backtest-reference A/B monitoring

Code

services/strategy_fifo_live/main.py services/strategy_fifo_paper/main.py

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

← #37 FIFO-Trends proper live simulation (v2)#39 Dashboard momentum signals tab →

This page is one experiment from the audited record behind the book The Edge That Wasn't — get the book · about the project.