Does trading the gap fill make money?

The fills are real. The money is not. On small gaps the prior close was reached on 72.4% of 2,540 trades, and the rule lost $27,659 after costs over eighteen years: the 1,839 fills made $454,066, the 534 stops lost $453,999, and the 167 days that did neither cost $27,726. On big gaps a quarter of the gap back made $23,132, of which 2026 alone is $34,876; the other eighteen years together lost $11,743, and 646 of 5,000 coin-flip twins beat it on the years before 2023. The stop at 0.4% instead of 0.5% makes the two nets $3,078 and $8,513 better and changes no verdict. On the months the stat was cut on, small gaps filled 34 of 39 times, 87.2%, for $18,762: the stat reproduces, and eighteen years say those were the good months.

Every gap day in its band is a trade: 2,540 small gaps and 2,248 big ones, 4,788 together (the gap page counted 4,777 gap days by its own definition); no signal was dropped for a target behind the price. Stopped includes the 2 trades that touched the stop and the target on one bar, taken as a stop. The white rows are the setup as written; runs 3 to 5 are the runner's half and the tighter stop a reader asked about.

Run 1 is the shape of the whole setup: 1,839 fills worth $454,066 against 534 stops worth $453,999, which cancel to within $67, and the 167 days that reached neither by 4:00 are the bill. The average fill made $247 and the average stop lost $850: the stop sits 0.5% from the entry and the prior close is under 0.4% away, so one stop costs about three and a half fills.

Before 2023, run 1 was up in 6 of 15 years and run 2 in 6; the engine scores on those years. The white rows, 2023 to 2026, are the years it sets aside as seen, declared here by the examiner since the stat was cut on 2026, and they do not decide the verdict. Run 1 lost $32,994 over 2023 to 2025, the three years the stat was sold on, and made $4,560 in 2026 to August 14. Run 2's +$23,132 is 2026's +$34,876 and -$11,743 from the other eighteen years together. Run 1 went 176 months without a new equity high, ending October 2022; run 2 168, ending February 2022. Median year: -$808 and -$1,758.

The twins take the same trades on the same days with the same costs and flip a coin for the side. On the years before 2023, run 1 is beaten by 267 of 5,000, run 2 by 646: neither clears one in twenty. Move the 0.4% cut ten percent either way and run 1 keeps none of its result and changes sign; run 2 keeps 84%. No run passed; the two that came close are the big-gap quarter, whose profit is 2026.

The stat behind the setup was cut on six months, March 25 to September 24, 2026; the file ends 2026-08-14, so this is 20 of those 26 weeks. The shared version says the prior close was hit before the stop 85.7% of the time on small gaps. Here it is 34 of 39, 87.2%, for $18,762 after costs. The months reproduce. The eighteen years above are the same rule on every other month.

What this does not say. The runner: the shared version takes most of the position off at a quarter of the gap and lets a runner go to half, two contracts with two targets. The engine runs one contract to one target, so the two-contract version is the sum of run 2 and run 3, +$14,595 with the 0.5% stop, not a managed pair. Their gap is read off the 9:30 open print and this one off the 9:29 close, a few ticks a day. The years from 2023 are declared seen by the examiner, since the stat was cut on 2026; nothing about how the setup was arrived at was sworn. MNQ is the same points at a tenth of the dollars; no other market was run. Roll days are in, as on the gap page.

How it was done. The setup, its five runs and seven predictions were written down before anything ran (SETUP 10, September 28, 2026; six of the seven held, and the miss is the second half of the finding: the declaration said the big-gap quarter would lose, and it made +$23,132). Family Y, the gap as written, in families3.py v0.5, run by engine_score.py under Protocol v1.1 through run_setup10-2026-09-28.py v0.1 on the one-minute NQ file: 6,180,656 bars, 2008-01-02 to 2026-08-14, none excluded. The reference is the last prior session's 4:00 close, so Monday reads Friday; the gap is the 9:29 bar's close against it; the fill is the 9:30 bar's open, plus a tick; $14.20 a round trip, commission and a tick each way; a bar that touches both is a stop; flat by the 4:00 stamp. The five reports and records are the record for this page, setup10_1_small_close_050.txt to setup10_5_big_quarter_040.txt with their .record.json, and the months are read by window_y-2026-09-28.py, whose print is window_y-2026-09-28.txt; every number here is read from them.

Raw market data: FirstRate Data (firstratedata.com). A measurement on a file, not advice.