Scenario
Track record
Every signal this project has published, date-stamped, with what actually happened next. The record is generated from the signal log — nobody hand-picks rows. Newest signals show as "pending" until their 2-day window closes. Provenance is labelled per record from actual capture latency: 39 of the 63 forward-test sessions (61% of records) were captured by the running pipeline within 3 days and are tagged live; the 24 earliest sessions, 2026-07-01 to 2026-08-04, were rebuilt by a later pipeline-gap repair pass and are tagged reconstructed.
What this record shows, net of a 0.10% round-trip cost. No probability floor has been reliably profitable (numbers load from the daily edge audit). Every row is net of a 0.10% round-trip cost and of modelled execution slippage: wins are recorded at +2.979% rather than +3.000%, and losses at −3.049% rather than −3.000%, because a stop that gaps through the level fills at the open. The signals are published because the record is real, not because it is profitable.
What the model actually predicts
Scenario track record
out-of-sample exam replay of frozen models
Live published signals — 3pct_2d
| Date | Ticker | Probability | Outcome (next 2 trading days) | Result |
|---|
WIN and DEFEAT are the first +3%/-3% barrier touched. MISS means neither barrier was touched: Profit means the t+2 close was at or above entry, Loss means it was below. Outcomes are computed from public price data by the same script that publishes them.