Log: logs/edge_discovery_paper.jsonl | 61 records | as-of 2026-09-23. Judge gate: >= 20 matured snapshots.
Realized = EW return of logged holdings from signal-date price to the price hold_days later; benchmark = EW market over the same window.
| hid | snapshots | matured | pending | mean ret | hit | vs market | excess hit | status |
|---|---|---|---|---|---|---|---|---|
| H2 | 16 | 0 | 7 | - | - | - | - | INSUFFICIENT (0 matured) |
| H5 | 16 | 0 | 16 | - | - | - | - | INSUFFICIENT (0 matured) |
| H6 | 15 | 0 | 15 | - | - | - | - | INSUFFICIENT (0 matured) |
| H10 | 14 | 0 | 0 | - | - | - | - | INSUFFICIENT (0 matured) |
Reading
- The forward log only started 2026-09-22; snapshots mature when the hold window elapses AND a price exists at signal+hold. Until >= 20 mature, no candidate is judged (a running record, not a verdict).
- H10 (the placebo survivor) is the priority; H2/H5/H6 are the placebo-rejected controls (expected to fade / track momentum).
- Keep the paper_tracker scheduled: it is the only source of genuinely unseen H10 data.