Public scorecard
We grade ourselves in public. These are the numbers that decide whether this tool earns its keep — measured against a locked target, no vibes.
Generated 2026-08-02 · 88 labelled outcomes (span 4d). Numbers match the internal SCORECARD, single-sourced.
Goal grades binary, measured against the Oct 31 2026 target
| Goal | Status | Now |
|---|---|---|
| Win recall ≥ 80% | NOT YET | 66.7% (n=27, target 80.0%) |
| BUY precision ≥ 70% | NOT YET | 58.1% (n=31, target 70.0%) |
| Paper portfolio beats holding sealed boxes | PASS | 188.9% vs baseline 0.0% |
Effectiveness tracked on separate axes, never collapsed into one score
| Metric | Value | Sample |
|---|---|---|
| Win recall | 66.7% | n=27 |
| BUY precision | 58.1% | n=31 |
| Verdict calibration (Brier, lower is better) | 0.284 | n=39 |
Paper portfolio $100 hypothetical per BUY_NOW alert, marked to later sold comps
166 positions · return 188.9% vs sealed-box-holder baseline 0.0%. This is a hypothetical, not a track record of real trades.
Latency
Alert median 9.529s (p95 17.079s, n=3002).
See the methodology for how these are computed, and the verdict archive for the calls behind them.