A sample report. A real screen from this site’s record, shown in full — the numbers a plan adds to every screen are included. On a paid plan your own screens look exactly like this, with every trade to download; the free screen shows the verdict, whether each gate passed, and the line written by the gate that decided it.
Not this one — the edge is too small to tell apart from noise.
On NQ, trade the reading alone (20 EMA) at 10:00, 12:00, 14:00. It acts only when the reading is in its top or bottom 40% of the last 250 sessions. Risk 0.8 ATR against a 2.0 ATR target, flat at the close (16:00 ET).
Decided by the edge vs. noise gate, on its edge vs. sample size check
3,010 trades. The smallest edge this sample can resolve is +0.061 R; the idea reported +0.015 R, inside the noise (t = 0.62, needs 2.49).
It averaged +0.015 R a trade (about 1.25 NQ points, $25 a contract). Over 3,010 trades, an edge has to be at least +0.061 R (about 4.75 NQ points, $95 a contract) before it stands out from luck — about four times the edge, or about 17 times as many trades (roughly 50,000) at the same edge.
R is the stop distance: +0.010 R means each trade made 1% of what it risked. Points and dollars are at NQ’s recent median ATR, on one full-size contract; every trade was sized on its own session’s.
Noise shrinks with the square root of the trade count, so every sample has a smallest edge it can tell apart from luck. This check puts that edge beside the one the idea measured. A small edge stops here even over thousands of trades — it is not that there were too few trades, it is that the edge is too small for the trades there were.
The edge vs. sample size gate ends this one: across roughly three thousand trades, the per-trade advantage your specification produced is smaller than what a sample that size can separate from chance, so the result is rejected rather than left open. This is not a shortage of trades — more of the same would not rescue an edge this thin. The checks that ran afterwards, including the one Deflated Sharpe pass inside the same block, do not revise that; they are diagnostics on an already-rejected specification. The thing worth your attention next is the entry condition itself, since a mid-band skip on a single moving average at fixed clock times is a weak filter. Consider what would need to change to produce a larger per-trade edge, not a longer sample.
105,161 specifications screened here to date; about 1 in 187 clears the full stack.
The numbers
Deepest -$85,942
Top: what one contract made, cumulatively, with the dashed line marking its running high. Bottom: how far under that high the account sat, which is the part you would have had to hold through. The curve is sampled to keep the page light, so read every figure off the four numbers with it — they are measured on every fill.
Per year, per contract
$4,475
One contract, no compounding, costs already charged.
Maximum drawdown
-$85,942
Peak to trough on the same single contract.
Average trade
+0.015 R
Against an average round-trip cost of 0.012 R.
Trades
3,010
1.34 per session over 8.9 years.
About $448 a year on one MNQ: a tenth of the full-size figure, costs scaled down with it. A micro's commission is a larger share of its smaller tick, so treat this as a ceiling.
Every figure is for a single contract and is net of commission and slippage charged per fill at this contract’s measured cost. Sessions run 2018-01-02 to 2026-09-07. The most recent completed quarter is held back from every research path and is not included here.
What it did
Its average trade was better than 36% of the 32,112 specifications in the record with a measured average trade.
Three trades it took
Times are fills, in ET: every decision fills at the open of the next half-hour, so a 10:00 decision shows as 10:30.
- Tue, Mar 31, 2026, 10:30 ET — bought NQ. Was stopped out in the 10:30 half-hour (−1.01 R).
- Tue, Mar 31, 2026, 12:30 ET — bought NQ. Reached its target in the 12:30 half-hour (+2.49 R).
- Tue, Mar 31, 2026, 14:30 ET — bought NQ. Was held to the flat time and closed there (+0.82 R).
The gate stack
- Passed
01Causality
Could the signal have read a bar it could not have seen live?
- Causalityby constructionPassed
Built only from series on the menu, each rebuilt on truncated data and matching on every shared session before it is offered.
- Failed
02Edge vs. noisedecided the verdict
Is the average trade clearly above zero: over enough trades, not carried by a handful of sessions, and after counting every idea screened here?
- Edge vs. sample sizeFailed
3,010 trades. The smallest edge this sample can resolve is +0.061 R; the idea reported +0.015 R, inside the noise (t = 0.62, needs 2.49).
What it measuredWhat 3,010 trades can tell from noise+0.015 R+0.061 RIt averaged +0.015 R a trade (about 1.25 NQ points, $25 a contract). Over 3,010 trades, an edge has to be at least +0.061 R (about 4.75 NQ points, $95 a contract) before it stands out from luck — about four times the edge, or about 17 times as many trades (roughly 50,000) at the same edge.
R is the stop distance: +0.010 R means each trade made 1% of what it risked. Points and dollars are at NQ’s recent median ATR, on one full-size contract; every trade was sized on its own session’s.
Noise shrinks with the square root of the trade count, so every sample has a smallest edge it can tell apart from luck. This check puts that edge beside the one the idea measured. A small edge stops here even over thousands of trades — it is not that there were too few trades, it is that the edge is too small for the trades there were.
- SignificanceFailed
Average trade +0.015 R; 95% interval −0.033 to +0.064 R; p = 0.27.
- ConcentrationFailed
Remove the best 17 of 1,708 sessions and the result turns into a loss (the bar is 35% of it surviving). 3 sessions carry half of it.
- SelectionFailed
p = 0.27 against a critical value of 0.01 once the 105,162 specifications screened here are accounted for (false-discovery rate 0.05).
- Deflated Sharpereported onlyPassed
Sharpe +0.17 annualised (skew −0.06, kurtosis 6.4). Deflated for every trial in the record it reads 0.00: a Sharpe this size needs more sessions than exist to stand on its own.
- Failed
03Drift nullafter the verdict — decides nothing
Does it beat random entries with the same trade count, hold and long/short mix?
- Drift nullFailed
+0.020 R above matched random entries; z = +0.65.
- Failed
04Walk-forwardafter the verdict — decides nothing
Does it hold on sessions the selection never saw?
- Walk-forwardFailed
Out of sample: −$9.94 a trade over 4 folds (2,411 trades); average trade −0.002 R against +0.004 R for a random entry on the same sessions.
- Walk-forward efficiencyreported onlyFailed
Walk-forward efficiency −0.24 against a bar of 0.30; 25% of 4 folds hold (median −0.43, worst −1.07).
- Cross-validation pathsreported onlyFailed
Across 5 out-of-sample paths: −$3.44 a trade (spread $5.06), from −$12.12 to +$0.59; 80% of paths under water. The single walk-forward path read −$9.94.
- Edge decayreported onlyPassed
Stable: trend −0.002 R a year (t = −0.20); first half +0.027 R, second half +0.004 R.
- Failed
05Mirror testafter the verdict — decides nothing
Does the opposite trade, taken on the same bars, do clearly worse?
- Mirror testFailed
Forward +0.015 R against inverse +0.007 R on 2,149 paired entries: a paired edge of +0.002 R, t = +0.06 against a bar of 1.0.
- Failed
06Costafter the verdict — decides nothing
Does the edge clear real per-fill commission and slippage, with room to spare?
- CostFailed
At three times the measured cost the average trade is −0.015 R.
- Cost under stressreported onlyPassed
Entries land at the 60th percentile of tape stress, worth +0.03 ticks of extra cost.
- Not run
07Era splitafter the verdict — decides nothing
Did it also pay in the years before the sample it was judged on? Reported on survivors; it never decides a verdict.
- Era splitNot run
01Causality
PassedCould the signal have read a bar it could not have seen live?
- Causalityby constructionPassed
Built only from series on the menu, each rebuilt on truncated data and matching on every shared session before it is offered.
02Edge vs. noisedecided the verdict
FailedIs the average trade clearly above zero: over enough trades, not carried by a handful of sessions, and after counting every idea screened here?
- Edge vs. sample sizeFailed
3,010 trades. The smallest edge this sample can resolve is +0.061 R; the idea reported +0.015 R, inside the noise (t = 0.62, needs 2.49).
- SignificanceFailed
Average trade +0.015 R; 95% interval −0.033 to +0.064 R; p = 0.27.
- ConcentrationFailed
Remove the best 17 of 1,708 sessions and the result turns into a loss (the bar is 35% of it surviving). 3 sessions carry half of it.
- SelectionFailed
p = 0.27 against a critical value of 0.01 once the 105,162 specifications screened here are accounted for (false-discovery rate 0.05).
- Deflated Sharpereported onlyPassed
Sharpe +0.17 annualised (skew −0.06, kurtosis 6.4). Deflated for every trial in the record it reads 0.00: a Sharpe this size needs more sessions than exist to stand on its own.
03Drift nullafter the verdict — decides nothing
FailedDoes it beat random entries with the same trade count, hold and long/short mix?
- Drift nullFailed
+0.020 R above matched random entries; z = +0.65.
04Walk-forwardafter the verdict — decides nothing
FailedDoes it hold on sessions the selection never saw?
- Walk-forwardFailed
Out of sample: −$9.94 a trade over 4 folds (2,411 trades); average trade −0.002 R against +0.004 R for a random entry on the same sessions.
- Walk-forward efficiencyreported onlyFailed
Walk-forward efficiency −0.24 against a bar of 0.30; 25% of 4 folds hold (median −0.43, worst −1.07).
- Cross-validation pathsreported onlyFailed
Across 5 out-of-sample paths: −$3.44 a trade (spread $5.06), from −$12.12 to +$0.59; 80% of paths under water. The single walk-forward path read −$9.94.
- Edge decayreported onlyPassed
Stable: trend −0.002 R a year (t = −0.20); first half +0.027 R, second half +0.004 R.
05Mirror testafter the verdict — decides nothing
FailedDoes the opposite trade, taken on the same bars, do clearly worse?
- Mirror testFailed
Forward +0.015 R against inverse +0.007 R on 2,149 paired entries: a paired edge of +0.002 R, t = +0.06 against a bar of 1.0.
06Costafter the verdict — decides nothing
FailedDoes the edge clear real per-fill commission and slippage, with room to spare?
- CostFailed
At three times the measured cost the average trade is −0.015 R.
- Cost under stressreported onlyPassed
Entries land at the 60th percentile of tape stress, worth +0.03 ticks of extra cost.
07Era splitafter the verdict — decides nothing
Not runDid it also pay in the years before the sample it was judged on? Reported on survivors; it never decides a verdict.
- Era splitNot run
Year by year
| Year | Result | Trades | |
|---|---|---|---|
| 2018 | $2,966 | 367 | |
| 2019 | -$2,606 | 367 | |
| 2020 | -$13,893 | 350 | |
| 2021 | $31,049 | 349 | |
| 2022 | $68,129 | 361 | |
| 2023 | -$9,794 | 384 | |
| 2024 | -$9,757 | 372 | |
| 2025 | -$30,073 | 369 | |
| 2026 | $862 | 91 |
What this means
This specification did not pass screening. It stopped at the edge vs. noise gate, on the edge vs. sample size check. 3,010 trades. The smallest edge this sample can resolve is +0.061 R; the idea reported +0.015 R, inside the noise (t = 0.62, needs 2.49).
Noise shrinks with the square root of the trade count, so every sample has a smallest edge it can tell apart from luck. This check puts that edge beside the one the idea measured. A small edge stops here even over thousands of trades — it is not that there were too few trades, it is that the edge is too small for the trades there were.
The checks after it still ran, and the stack below shows them greyed out. Once one gate has failed they describe a specification the data has already ruled on, so none of them can change the verdict — a pass among them is not a pass for the idea.
This is a complete answer, and it is worth having before anyone builds anything. For context: 105,161 cells screened here to date, about 1 in 187 clears the full stack. What this run bought you is a clear reading of where the idea stands, and the gate above is the place to look for the next version.
Notes from the run
- The gates are measured on the 2,239 sessions from 2018-01-02, the sample their pass lines were set on. The era label also reads the 1,289 earlier sessions, back to 2013-01-02.
- Exit mix: stop 56%, flat 28%, target 16%.
- Worst stretch under water: 1,181 calendar days, 96% of the sample spent below a prior peak.