What the lab rejected
Every run the lab has evaluated, counted with every configuration it tried. A run passes only if it clears every check of the candidate bar; 0.0% have so far.
By source
| Family | Configurations | Runs | Passed | Rejected |
|---|---|---|---|---|
| House2,370 configurations | 2,370 | 888 | 0 | 888 |
| Imports17 configurations | 17 | 5 | 0 | 5 |
| Questions13 configurations | 13 | 4 | 0 | 4 |
| AI agents0 configurations | 0 | 0 | 0 | 0 |
| Forks0 configurations | 0 | 0 | 0 | 0 |
| Allocations0 configurations | 0 | 0 | 0 | 0 |
| Arena10 configurations | 10 | 4 | 0 | 4 |
| Pre-registered0 configurations | 0 | 0 | 0 | 0 |
| Baseline2 configurations | 2 | 1 | 0 | 1 |
What each family holds is on How we test.
By the first check that failed
| Check | Runs | Share of rejected |
|---|---|---|
| the fast screen | 240 | 26.6% |
| information ratio | 428 | 47.5% |
| Newey-West t-stat | 232 | 25.7% |
| deflated Sharpe | 1 | 0.1% |
| probability of overfitting | 1 | 0.1% |
A run is listed under the first check it failed, in the bar's order: information ratio, t-stat, deflated Sharpe, overfitting, false discovery, lookahead, LLM cutoff, then the stress checks once they join the bar. The fast screen stops a run before the bar.
147 more runs stopped before a result (an error, a stop or a time limit) and are not counted above.
As of 2026-10-18T02:14:26+00:00, refreshed at most every 60 seconds. Source: Postgres research.trial_ledger joined to research.runs (each run's bar). Aggregates only: no strategy, family or person is named.
Research, not advice. Past results do not promise future returns. StockGaines publishes its method to everyone alike and does not advise anyone on whether a strategy suits them.