How we grade ourselves
Most trading tools publish their wins. We built a system that grades every recommendation — and every refusal — against the do-nothing alternative, in public, with receipts.
We built this scoreboard before knowing what the score would be. The ledger below started counting on July 29, 2026 — whatever it says in a quarter, it says in public.
Check your ticker
We grade the regime detector too
Every verdict stamps the market regime the detector saw at issue. Twenty trading days later we also record whether that regime persisted or transitioned — and score the verdict separately in each stratum. If the detector is informative, verdicts should survive persistence better than transitions. A verdict on the detector itself is only issued once both strata reach n≥5 matured checks.
| Regime at issue | Matured | Regime persisted | Regime transitioned | Detector verdict |
|---|---|---|---|---|
| strong uptrend | 349 | 109 of 219 verdicts held | 27 of 130 verdicts survived | informative |
| moderate trend | 231 | 23 of 39 verdicts held | 109 of 192 verdicts survived | informative |
| strong downtrend | 63 | 19 of 34 verdicts held | 8 of 29 verdicts survived | informative |
| range-bound | 61 | 12 of 20 verdicts held | 17 of 41 verdicts survived | informative |
18 matured receipts carry no regime stamp (enrolled before stamping shipped) — excluded, not hidden.
Meanwhile on our structural board: ▲31 JPMPL climbed to rank 60 (Strong Downtrend) this week — the screener hunts engine-friendly structure, and its picks are graded against a control group.
Our stock screener is graded too
Our internal screener hunts engine-friendly structure — and enrolls its top picks AND a control group of mid-ranked names into the same forward ledger. Matured so far: picks showed a validated out-of-sample edge in 10 of 11 re-races, controls in 10 of 10 — a gap of -9.1 pts, published whichever way it lands.
Every recommendation becomes a receipt
When our optimizer recommends a configuration — or refuses and says "keep your baseline" — that decision is enrolled in a forward ledger automatically. Twenty trading days later, we grade it: did the chosen config actually beat keeping the baseline on the exact same forward bars? Then again at 60 days and a quarter. The grade measures the decision, not the stock — a recommendation on a stock that fell can still be right if it lost less than the baseline would have.
We grade our refusals too
When the optimizer finds no configuration that honestly beats what you already have, it says so — and that refusal is enrolled and graded like any other decision. A refusal scores as correct when the baseline wins the forward window. Nobody else grades their own no’s; we think it’s the most important grade on the page.
The ledger is hash-chained
Each receipt carries a SHA-256 hash of its contents chained to the previous receipt, so the history cannot be quietly edited, backdated, or pruned after the fact. A broken chain is detectable by anyone. The same construction protects our public regime-forecast calibration ledger.
Counts, never percentages — pooled across users
Matured receipts pool across all users by ticker, engine, and configuration (within tolerance). We report them as counts — "beat baseline in 7 of 9 matured checks" — never as polished percentages, and we claim nothing below five matured receipts. Small samples get labeled as small samples.
The gates that judge results are versioned
The thresholds behind every verdict — what counts as validated, what triggers an overfit warning, when a drawdown is flagged — live in an append-only registry, currently v1. Any change must be pre-registered: declared with a future effective date and the receipt evidence motivating it, then graded only on receipts that mature AFTER the change. We never retro-fit a threshold to the ledger that motivated it and call it verified.
Live positions are never touched
Gates and thresholds apply at decision time only. Positions snapshot their full parameters at entry, and no system change — gate version, memory update, anything — reaches back into an open position. Improvements affect the next decision, presented as evidence before you act, and strategy-level changes always end at a human decision.
All results are hypothetical backtests and forward paper-grades, not live trading or advice. The one promise this page makes: the grades will be published whether they flatter us or not.