Riley Coleman's Free 5-Step Checklist — Full Test Report

What we granted, and what we actually tested

Granted, not measured: that Riley Coleman trades profitably. We never audited his P&L and this test cannot. He is on record in our own files as a genuine, honest live trader — he narrates real trades, shows losses on camera, and says out loud when months have gone badly. The question here is never the man.

Tested: only the part of the method that fits in a PDF — the written checklist, handed to a machine. One step set of six was mechanized. Four components were deliberately excluded and named in writing before the first bar loaded (§1 below).

§1 · The locked pre-registration

Committed 8a8540e on 2026-06-29, before the probe was run — the wager you cannot walk back once the data arrives.

The falsifiable question

Does a purely mechanical fade of an "unhealthy" opening drive into an early-session extreme — entered on a momentum-reclaim trigger in the 09:40–10:30 ET window, stopped beyond the extreme, with a fixed-R target — produce positive net-of-cost expectancy, robustly across the sample?

• Yes looks like: net expectancy > 0 after a 0.05R friction haircut, and positive in both sample halves, and not driven solely by the top-5 days, and beating the "fade every drive" benchmark.
• No looks like: any one of those four failing.
• Ambiguous residue: gross-positive but net ≈ 0 (friction-bound), or exactly one check failing near threshold.

Our stated prior: ~30%

Written down before results: about a 30% chance the mechanical skeleton clears net-positive robustly. Recorded so that a positive result would have had to overcome a low prior — and so a negative one could not be rationalised away as expected. Grading our own prior: the outcome landed in the 70% we called more likely.

The primary configuration (locked)

Deliberately NOT tested — named in advance

1. The big-picture directional filter — which he repeatedly calls the most important factor. It is not operationalized anywhere in the five steps.
2. Head-and-shoulders confirmation (step 4) — a chart shape, too fuzzy for honest code.
3. Zone construction — pivot clustering and touch counting.
4. Discretionary management — scaling out and trailing. His narrated live 2.4R came from managing, not from a fixed target.

The four goalposts — all four required to proceed

0 of 4. Not an ambiguous middle: the result is gross-negative, not gross-positive-but-friction-bound, and not near threshold. The pre-registered verdict for this exact outcome was kill the skeleton.

§2 · The headline arithmetic

31.3% against 33.3%, both at a 2R target. Two points short of its own break-even, before a single dollar of cost. That is the whole finding in one line. There was nothing here for costs to kill.

§3 · Was it the path? Three stress tests

• Split the 22 years in half — both halves negative: -0.181 R (606 trades) and -0.079 R (606 trades). No lucky era.
• Remove the five best days — it gets worse, -0.138 R, because those extreme days were net contributors against the strategy (-6.2% of net). Not an outlier story.
• Sweep every setting — nine parameterizations (three drive thresholds × three targets). All nine came back negative at realistic cost. No magic setting.
• Breadth — the setup found trades in 17 calendar years and never fired at all in 6 of them. Of the 17 it traded, 13 lost money at realistic cost.

§4 · The two controls — and the row that made us sit back

Control B1 — fade every morning drive, with his signature "unhealthy move" filter removed entirely. It traded 4,050 times, more than three times as often, and came out roughly break-even before costs: +0.009 R gross (38.3% win), -0.042 R after realistic cost.

Requiring the big stretched drive — his single most-emphasized signal — makes expectancy worse: -0.080 R against +0.009 R. The mechanized version of the filter selected worse entries, not better.

Control RC — random timing. Replace the momentum-reclaim trigger with entry at a random qualifying minute: -0.155 R at realistic cost, against the real trigger's -0.130 R. The trigger adds a little. It does not add an edge.

§5 · The best objection, pre-registered and answered

"You tested one fund." Fair — so we wrote a prediction down (SENSITIVITY_PREREG.md, locked before the run: still negative on at least three of the four) and pointed the same locked machine, without changing a line of code, at three more index funds.

Three of four lost before a single cent; all four lose after costs. The prediction holds. We are not saying it lost on all four — IWM came out gross-positive, a coin-flip whisper at profit factor 1.01, and it is recorded here rather than rounded away. It goes negative the moment costs exist. The one market that did not lose gross is also the one furthest from anything Riley trades.

§6 · His own risk-reward terms

Two R was our choice; his framing is a bigger reward-to-risk. So we ran a 3R target, where the naive break-even drops to 25%. Measured win rates at 3R: 25.9%, 26.3%, and 24.9% across the three drive thresholds. Expectancy stays negative anyway (-0.049 R at the primary threshold) — because of the clock. The 12:00 time-stop closes whatever is still open and those exits pay about 1R apiece, so the average winner shrinks to 2.579R and the real break-even moves to about 27.9%. Out of reach from any cell.

§7 · The complete trade ledger — all 1,212 trades

Every trade the machine took, in order. Nothing is filtered, sampled, or summarised. 354 reached the target, 820 were stopped out, and 38 hit the 12:00 time-stop.