86 claims graded on 749,127,548 NQ trades. 9 hold up (5 supported, 4 descriptive), 2 run backwards, 32 don't hold, and 43 can't be settled with the data that exists.
Regraded Oct 2026 by one written rule after an independent review; 13 labels weakened, none strengthened. See Method. Scope: NQ, 2018 to 2026, 60-minute ±0.10 ATR horizon.
Key Insight: no claim both beats its baseline and pays for its trading costs.
What to carry into your own notes, each with the number behind it.
| Conclusion | The number |
|---|---|
| Delta describes the present, it doesn't forecast | A 1-minute bar's delta explains 43.6% of that bar's move (R²) and almost none of the next bar's |
| Absorption adds nothing to direction | 49.9% with it, 49.7% without, on 4,423 events. The registered rule agrees: 51.1% against 51.1% on 33,281 |
| The footprint canon as a whole is flat | Pooled -0.12 pts [-0.45, +0.22], 472,284 events, 19 claims, I² 19% |
| Levels don't clearly beat random prices the same distance away | Naked POCs fill 85% of the time, random prices at the same distance 84%. The interval (+1.9 [-0.5, +4.3]) can't rule out a small edge |
| A hit rate without a baseline isn't evidence | The 80% rule's fans weren't lying about what they saw. They just never saw the comparison |
| Confluence doesn't clearly help | "Full house" setups (CX-16) were +3.8 [-2.8, +10.5] against their baseline on development and -6.8 on the holdout |
| The taught setups are too rare to learn from experience | Several confluences fire 4-10 times a year. Three years of screen time is ~15 observations |
| The signals are volatility detectors, not direction signals | MFE/MAE shift correlation -0.86; reward:risk 0.966 without vs 0.956 with |
| A wider target doesn't help | All 18 stop/target shapes lose money. Expectancy gets worse as the target widens |
| The only forecast that held up works one tick ahead | Queue imbalance calls the next tick 65.6% of the time. OFI explains 70.1% of 10-second price changes, but over the same 10 seconds, so that's a data check, like delta, not a forecast |
| Almost no practitioner claim survives | OPEX compression and dealer gamma looked like passes but don't survive the study's own multiple-testing rule. Wide-IB extremes holding (OI-05) is the only practitioner claim Supported after correction, narrow IBs ran backwards (OI-04), and the holdout couldn't confirm either |
| Two claims run backwards | Narrow IB -11.9 pts, sweeps -3.6 pts (smaller on the holdout) |
| "Inconclusive" is the most common outcome for levels and profile | 43 of 86. Means unresolved, not disproven |
Every claim that beat its baseline or ran backwards. Supported means it beat the baseline by 3+ points and survived multiple-testing correction. It does not mean tradeable, see the double bind below.
| ID | Claim | Verdict | Effect | 95% CI | n | Units |
|---|---|---|---|---|---|---|
| MR-01 | Order-flow imbalance at the top of the book explains most short-horizon price change. | Supported | +20.1 | [+17.9, +22.2] | 40 | see primary test |
| MR-06 | Queue imbalance at the best bid and ask predicts the next mid-price move. | Supported | +15.6 | [+13.5, +17.8] | 40 | see primary test |
| MR-02 | Delta explains same-bar returns, but predicts next-bar returns only weakly. | Supported (descriptive) | +43.6 | -- | 739,638 | R², same bar (1 min) |
| MR-05 | Liquidity (price move per contract of imbalance) varies sharply by time of day and regime. | Supported (descriptive) | -- | -- | 147,931 | ticks per 1k contracts |
| MR-08 | Intraday volume and volatility are U-shaped, with spikes at the 8:30 and 10:00 ET releases. | Supported (descriptive) | -- | -- | 1,858 | - |
| OI-03 | The initial balance is broken on at least one side on most days. | Supported | +46.9 | [+46.0, +47.6] | 1,818 | rate difference |
| OI-05 | A wide IB means its extremes hold for the day. | Supported | +4.8 | [+2.9, +6.8] | 1,809 | rate difference |
| RL-02 | On trend days, price holds one side of VWAP after 10:30 ET. | Supported (descriptive) | +11.6 | [+7.9, +15.3] | 1,818 | rate difference |
| ME-04 | Profiles built from 1-min bars are close enough to profiles built from ticks. | Supported | +1.6 | [+1.5, +1.6] | 1,818 | see primary test |
| OI-04 | A narrow IB leads to a trend day or a large range extension. | Reversed | -11.9 | [-14.8, -9.0] | 1,809 | rate difference |
| OB-03 | Sweeps through 3+ price levels continue over the next minute. | Reversed | -3.6 | [-4.1, -3.1] | 160,196 | see primary test |
Removed in the Oct 2026 regrade: OI-07 and CF-04 (intervals cross zero), CF-02 (q about 0.13) and RL-03 (q about 0.08, and its Reversed label came from a post-holdout rerun). All four are now Inconclusive or Not supported. RL-02 is descriptive because "trend day" is decided at the end of the day. Of these rows, only MR-01, MR-06, OB-03, OI-03 and ME-04 replicated on the holdout.