A convex exit made money on MES.
So did a coin flip, to the decimal.
A pre-registered, one-shot holdout test of a momentum exit law on MES micro futures. The real strategy produced +1.338 points per trade. A coin-flip twin, run through the identical stop and flatten logic, produced +1.359. Paired t-statistic: -0.01. That is not a rounding artifact — it is the sharpest null in this lab's three-generation record.
The pre-registration
Before this test ran, the exit law was frozen: a stop-only bracket (no trailing stop) combined with a
hard 15:50 ET flatten, replacing Gen-2's trailing-stop exit. The entry gate and stop-sizing law were
inherited verbatim from the prior generation — zero new constants were introduced or calibrated for
this test. That freeze is a git commit, not a claim: 481eced0770, timestamped
2026-07-07T06:10:00 UTC, made before the holdout script was run against a single point in the data it had
not seen.
The success bar was declared in that same commit, before the result existed: a paired t-statistic ≥ 2.0 and expectancy > 0.8 points per trade. The holdout script itself refuses to run twice — once consumed, a second attempt raises an error. One shot means one shot.
The result
123 sessions, 2026-01-01 through 2026-07-03. 108 trades fired on both the real signal and its coin-flip twin — the twin takes the opposite or same direction on a 50/50 basis, fixed in advance by hashing the session date, and runs through the exact same stop and flatten code as the real trade.
| Leg | Trades | Hit rate | Total pts | Expectancy pts/trade |
|---|---|---|---|---|
| Real signal | 108 | 35.2% | +144.50 | +1.338 |
| Coin-flip twin | 108 | 35.2% | +146.75 | +1.359 |
| Pairs | Mean diff pts | SD | SE | t-statistic |
|---|---|---|---|---|
| 108 | -0.021 | 26.01 | 2.503 | -0.01 |
A secondary split (not a second holdout, and not a claim — the pre-registration allowed one run against the whole window): the first 82 sessions carried the total, at +2.631 pts/trade; the final 26 sessions gave almost all of it back, at -2.74 pts/trade. We report this because burying an inconvenient split inside an aggregate is exactly the kind of thing this site exists to not do — not because it changes the verdict, which is set by the paired test on the full window.
Why the coin flip won too
Both legs made money using the identical exit: cut losses near one times ATR on a stop, remove the cap on winners, and hold every trade to the same 15:50 ET flatten regardless of direction. That structure is convex by construction — small losses, some large gains — and in a trending, high-volatility stretch of 2026 it paid out almost identically whether the entry direction came from the momentum signal or from a SHA-256 hash of the date.
In plain terms: the exit shape harvested the regime. The directional opinion — "price that has moved will keep moving" — contributed nothing measurable on top of it. A profitable backtest is not evidence of a working signal if an unsignaled coin flip, run through the same exit, is equally profitable.
What this is not
- Not a live result. Every number above comes from a historical simulation against recorded price data, not real order fills. No trading costs beyond the strategy's own stop/target levels are modeled.
- Not a recommendation. This page describes what one specific, frozen rule set did in one specific, one-shot holdout window. It is not investment advice and not a signal to trade MES or any instrument.
- Not evidence the underlying idea is wrong forever — only that these three specific, pre-registered attempts at it did not clear their own bar, on this data, in this window.
Provenance
Freeze commit: 481eced0770482e98504f1174c7bebd002e84762
Instrument: MES (Micro E-mini S&P 500 futures), 5-minute bars
Holdout window: 2026-01-01 to 2026-07-03 (123 sessions)
Coin-flip twin: direction fixed in advance by SHA-256 hash of session date, run through identical stop/flatten logic
Success bar (declared before the result): paired t ≥ 2.0, expectancy > 0.8 pts/trade