Running My EA on a Real Prop-Firm Account (Live Forward Log)
I put the EA finalized on this site into a real prop-firm evaluation account (¥10M). This page is a running log of what actually happens on that account — good and bad, reported as-is.
Most articles on this site are backtests. This page is different: it is the live forward record — what is happening right now to the EA running on a real account.
Ground rules for this log:
- Only measured numbers from the actual account. If I ever quote an estimate or a simulation, I will label it as such.
- Bad results get reported exactly like good ones. If the account fails the evaluation, this page will say so.
- Updates are irregular (when something happens, plus milestones).
What is running, under what conditions
| Item | Details |
|---|---|
| Account | Fintokei ProTrader Challenge STEP 1 (a real evaluation account) |
| Account size | ¥10,000,000 |
| Start date | 2026-07-04 |
| Pass conditions | +8% (then +6% in STEP 2). One breach of daily -5%, max -10%, or total open risk -3% means failure |
| EA | CoreSystemV14 — the MT5 implementation of the finalized system validated on this site (all 7 sleeves) |
| Risk settings | Challenge preset (base leverage 2.5 / cap 2.5 — stage 1 of the three-stage plan) |
| Safety guards | Flatten everything at daily -4%; halt new entries + flatten at account DD -7% (designed to step off well before the disqualification lines) |
For transparency, the backtest-derived expectations (past data, 2015–2026, 11 years): +0.82%/month, PF 1.66, max DD -7.7% for the base system. Monte Carlo on the challenge settings says there is roughly a 30% chance of touching a -5% account drawdown at some point before passing, and about a 1% failure probability on the unstressed distribution. In other words: seeing a few percent of drawdown along the way is part of the plan.
Log (newest first)
2026-07-28 — EA updated to v2.0.4 (auto)
The live EA version changed from v2.0.2 to v2.0.4. See the dev log for details.
2026-07-26 — EA updated to v2.0.2 (auto)
The live EA version changed from v2.0.1 to v2.0.2. See the dev log for details.
2026-07-23 — EA updated to v2.0.1 (auto)
The live EA version changed from v2.0.0 to v2.0.1. See the dev log for details.
2026-07-17 — EA updated to v1.21.7 (auto)
The live EA version changed from v1.21.6 to v1.21.7. See the dev log for details.
2026-07-17 — EA updated to v1.21.6 (auto)
The live EA version changed from v1.21.5 to v1.21.6. See the dev log for details.
2026-07-16 — EA updated to v1.21.5 (auto)
The live EA version changed from v1.21.4 to v1.21.5. See the dev log for details.
2026-07-16 — EA updated to v1.21.4 (auto)
The live EA version changed from v1.21.3 to v1.21.4. See the dev log for details.
2026-07-16 — EA updated to v1.21.3 (auto)
The live EA version changed from v1.21.2 to v1.21.3. See the dev log for details.
2026-07-16 — EA updated to v1.21.2 (auto)
The live EA version changed from v1.21.1 to v1.21.2. See the dev log for details.
2026-07-16 — EA updated to v1.21.1 (auto)
The live EA version changed from v1.21.0 to v1.21.1. See the dev log for details.
2026-07-15 — EA updated to v1.21.0 (auto)
The live EA version changed from v1.20.0 to v1.21.0. See the dev log for details.
2026-07-15 — EA updated to v1.20.0 (auto)
The live EA version changed from v1.19.1 to v1.20.0. See the dev log for details.
2026-07-12 — Deployed a new edge, “weekend gap fade”, as the 8th sleeve (EA v1.15)
A full census of 41 methods × 26 symbols × 4 timeframes = 3,680 combinations (study) surfaced one survivor family: buying the Monday down-gap and taking the refill. After measured spreads, null controls and multiple-testing corrections (deep verification), it went live in the real-account EA, together with a re-optimized sleeve allocation “config D” (study). On backtests this lifts monthly return from +0.79% to +1.06% / PF 1.69 at an unchanged max DD of -8.3% (pre-deployment safety checks). A spread-telemetry logger (every 15 minutes) shipped in the same release, so the assumed costs will be fact-checked against a few weeks of real Monday spreads. Results will be reported here.
2026-07-09 — Recomputed the post-pass operating parameters
A new defensive feature (the equity-shock guard below) changed the risk distribution, so I re-ran the full pipeline simulation for the post-pass stages. Updated expectations: median 155 trading days to first payout, ~1.82%/month steady-state withdrawals (on the PF 1.66 base). Challenge-stage settings unchanged.
2026-07-08 — EA updated to v1.12.0 (equity-shock guard)
Dissecting the worst loss days of the past 11 years showed a common pattern: the mean-reversion sleeve buys the dip during an equity-market crash and gets hit together with the index sleeve (analysis here). The fix: when US500 drops more than 3% over 5 trading days, new entries for those sleeves run at half risk for 5 days. Monte Carlo says this roughly halves the ruin probability (verification). Deployed to the live account.
2026-07-07 — A “half-hedge” appeared on the live account
The account held a GBPJPY long and a GBPUSD short at the same time. It looked contradictory, so before touching anything I tested it (results). Conclusion: two sleeves opened both positions for independent, valid reasons, and letting them coexist beats closing either side. The right move was to not intervene. Same day, deployed v1.11.0 — a break-even stop for one sleeve only, after verifying it improves every metric.
2026-07-04 — Challenge started
CoreSystemV14 went live on the Fintokei ProTrader STEP 1 account (¥10M).
2026-07-03 — The demo account caught a real bug the day before go-live
On a parallel demo account, positions opened by the calendar sleeve failed to close and just sat there. Root cause: the daily-timeframe exit fires exactly at server midnight — which always collides with the brief rollover market-closed window — and there was no retry after a failed close. This is a class of bug a backtest can never find. I implemented close-order retries the same day, then went live. This is exactly why the demo runs in parallel.
Current status
The EA writes its account state to a file every hour, and the operating machine automatically collects it into this repository. The figures below are those measured values. (The hourly-logging pipeline went live on 2026-07-09, so the continuous log starts there; the account itself has been running since 07-04.) This section is updated automatically from telemetry.
| Item | Measured |
|---|---|
| Current equity (vs. initial) | +0.67% (10,066,653 / 10,000,000 JPY) |
| Today’s P&L | +0.48% |
| Max DD (vs. initial, hourly obs.) | -1.22% |
| Guard state | OK (neither daily-flatten nor DD-halt triggered) |
| Open positions | 6 |
| As of (server time) | 2026.08.04 13:00:08 |
Current equity is +0.67% vs. initial (10,066,653 / 10,000,000 JPY), today’s P&L +0.48%. Open: 6 (GBPJPYp buy 0.19 (float +411 JPY); GBPJPYp buy 0.06 (float +6,342 JPY); EURJPYp buy 0.08 (float +5,792 JPY); AUDUSDp buy 0.16 (float +5,558 JPY); GBPUSDp buy 0.08 (float +632 JPY); USDJPYp buy 0.09 (float +6,219 JPY)). The deepest hourly-observed drawdown is -1.22% vs. initial, and the safety guards (daily -4% flatten / DD -7% halt) have never fired. Room to the disqualification lines (daily -5%, max -10%): 5.00pt / 10.00pt. For reference, the config D, 2015-2026 backtest shows +1.06%/month, PF 1.69.
This page will be updated with measured values at each milestone.
Disclaimer: This is my personal operating log, not investment advice. Prop-firm challenges are paid evaluations — failing costs the participation fee (¥12,500 in this case). Nothing here is a recommendation or a guarantee of results.