The claim this page exists to test: on mass data the system has no magic — and each honest constraint added gates returns upward. Every figure below is recomputed from the data lake at export time; nothing is hardcoded. Data through 2026-07-23.
Start with every name-day of the frozen watchlist universe (183 names as of 2026-07-02) during the discovery window (2026-05-14 → 2026-07-02), then add one gate per row. Forward returns are 5 trading sessions.
Correction (2026-07-25): the baseline row previously aggregated every measured name-day in the lake, including off-watchlist large-caps that ride along on the broker tape (ANZ, RIO, CSL…). Those stray rows ran 59% win / median +0.9% (n=783) and flattered the published baseline: 35% / median -2.0% (n=6888) restates to 32% / median -3.2% (n=6105). The rows below the baseline were already universe-bound, so the funnel's lift was understated, not overstated. Found in a pre-freeze audit; the universe is now frozen and stated above.
Removes: Nothing — every name-day of the frozen watchlist universe. The honest base rate: most small-cap days lose money.
Removes: Days when only one desk type is in, and flow arriving on ordinary volume.
Removes: Names reacting to news already out — leaves positioning before the news.
Removes: Names that had already run — leaves accumulation into price weakness.
coverage: of 61 hits, 44 had RSI data, 20 of those <40 — rows without RSI data are counted, never silently dropped.
⚠ Calibration analysis — in-sample, found by exploring. Not evidence; the receipts below are the evidence.
Everything before 2026-07-03 is calibration — in-sample, explored, not evidence. The first registered rule ran 2026-07-03 → 2026-08-04; its record stays intact below. The revised rule starts a fresh receipt at zero on 2026-08-04. Why the rule was revised — including a display rule that silently drifted from the measured one — is documented in the supersession note.
In-sample. These hits shaped the rule; they can never count as evidence for it.
| Date | Ticker | RSI | 5d |
|---|---|---|---|
| 2026-05-14 | OR3 | — | -3.5% |
| 2026-05-15 | OR3 | — | -3.5% |
| 2026-05-20 | SNX | — | -20.0% |
| 2026-05-20 | TMX | — | 0.0% |
| 2026-05-21 | ACWRSI<40 | 21 | +14.3% |
| 2026-05-21 | CY5 | 41 | +9.5% |
| 2026-05-22 | TLG | 77 | 0.0% |
| 2026-05-22 | TM1 | 61 | +2.8% |
| 2026-05-25 | PNN | 58 | +20.0% |
| 2026-05-27 | SNM | 62 | +4.3% |
| 2026-05-27 | WBT | 80 | -8.8% |
| 2026-05-29 | DRO | 46 | -18.0% |
| 2026-05-29 | LIN | 49 | +5.3% |
| 2026-05-29 | PNN | 65 | -13.3% |
| 2026-06-02 | AIS | 44 | -20.4% |
| 2026-06-02 | CBE | 73 | -12.5% |
| 2026-06-02 | PNN | 61 | -13.8% |
| 2026-06-02 | SYR | 64 | -12.0% |
| 2026-06-04 | BRN | 73 | -7.9% |
| 2026-06-04 | PNN | 55 | +7.4% |
| 2026-06-08 | ATV | 52 | +55.0% |
| 2026-06-09 | AT4RSI<40 | 17 | -12.0% |
| 2026-06-09 | DYMRSI<40 | 6 | -1.1% |
| 2026-06-09 | LKYRSI<40 | 24 | -12.8% |
| 2026-06-10 | CY5 | 58 | 0.0% |
| 2026-06-10 | DTRRSI<40 | 22 | +29.6% |
| 2026-06-10 | HLSRSI<40 | 31 | +30.2% |
| 2026-06-12 | KLIRSI<40 | 0 | -9.1% |
| 2026-06-17 | ATV | 77 | -28.6% |
| 2026-06-17 | KLI | 40 | 0.0% |
| 2026-06-17 | RCMRSI<40 | 28 | -11.8% |
| 2026-06-23 | KRR | 80 | +10.4% |
| 2026-06-23 | QPMRSI<40 | 29 | -25.0% |
| 2026-06-24 | NEURSI<40 | 39 | +36.8% |
| 2026-06-25 | EVRRSI<40 | 30 | 0.0% |
| 2026-06-25 | MTMRSI<40 | 35 | +14.7% |
| 2026-06-26 | GL1RSI<40 | 28 | +11.4% |
| 2026-06-26 | KRR | 85 | +5.5% |
| 2026-06-29 | GL1RSI<40 | 36 | +11.9% |
| 2026-06-29 | KRR | 85 | +5.6% |
| 2026-06-29 | TLGRSI<40 | 17 | +40.0% |
| 2026-06-30 | QPMRSI<40 | 21 | -30.8% |
RSI<40 subset — of 42 hits, 38 had RSI data, 16 of those <40 · resolved: 50% win, median +5.7% (n=16).
Res + investment banks both net-buying ≥$5k same day, volume ≥2× ADV, no price-sensitive news that day. Registered 2026-07-03; retires at the 2026-08-04 re-registration — record intact.
| Date | Ticker | RSI | 5d |
|---|---|---|---|
| 2026-07-03 | BOA | 59 | 0.0% |
| 2026-07-08 | KTA | 50 | +9.1% |
| 2026-07-16 | EXRRSI<40 | 36 | +8.3% |
| 2026-07-16 | MSB | 94 | -16.3% |
| 2026-07-17 | LRVRSI<40 | 32 | +10.8% |
| 2026-07-20 | BRNRSI<40 | 30 | -11.5% |
| 2026-07-22 | AON | 44 | pending |
| 2026-07-23 | GNM | 69 | pending |
RSI<40 subset — of 8 hits, 8 had RSI data, 3 of those <40 · resolved: 67% win, median +8.3% (n=3).
Res + investment banks both net-buying ≥$5k same day, volume ≥1.5× ADV, no price-sensitive news in the 3 calendar days ending at the hit day. Approved 2026-07-26 for construct validity; registered at the 2026-08-04 freeze. RSI<40 is the single pre-registered subset — no further slices without a new registration.
The receipt starts at zero on 2026-08-04. An empty receipt is the honest state, not a bug.
RSI<40 subset — of 0 hits, 0 had RSI data, 0 of those <40.
Most of what the system produces are categories it classifies as historically underperforming. These are the rejected tiers, measured identically to everything else.
Market makers hedge and warehouse — their flow is plumbing, not positioning.
Retail-only buying days show negative median forward returns at both horizons in this dataset.
The same price dip as the tracked accumulation pattern, but with retail rather than institutional buyers — the weakest measured cell in the lake.
Tier 1 ("confluence + catalyst") is the system's top tier — cross-group flow plus a price-sensitive announcement. Measured, its first fires run 37% win / median -2.6% (n=73) against a universe baseline of 32% / median -3.2%.
Measured against the corrected universe baseline, first fires sit above it on both win rate and median — a modest measured edge, with a median still below zero. The claim is recomputed at every export; if it stops being true, this sentence changes. The full working is in research note 01.
One email per market day. Zero when nothing happened.
Just the letter. No promotion, unsubscribe anytime.