diff --git a/docs/superpowers/specs/2026-04-15-oos-generalization-enhancement-design.md b/docs/superpowers/specs/2026-04-15-oos-generalization-enhancement-design.md index 5eb6304a0..c833a37a8 100644 --- a/docs/superpowers/specs/2026-04-15-oos-generalization-enhancement-design.md +++ b/docs/superpowers/specs/2026-04-15-oos-generalization-enhancement-design.md @@ -307,9 +307,10 @@ All 10 components deploy together as a single release. | Metric | Current | Target | |--------|---------|--------| -| OOS Sharpe (min across windows) | ~0 | > 5.0 sustained | +| **val/OOS Sharpe gap** | **val=45, OOS≈0 (∞ gap)** | **< 15% gap** | +| OOS Sharpe (min across windows) | ~0 | > 10.0 sustained | +| val_Sharpe (post-generalization) | 45 (overfitted) | 15-25 (genuine edge) | | Training Sharpe stability | -1.2 to +0.9 | > 1.0 sustained | -| val/OOS gap | val=45, OOS≈0 | gap < 2× | | Active features (w_s1 non-zero columns) | 42/42 | < 20/42 | | Branch correlation (max pairwise cos_sim) | unmeasured | < 0.1 | | Atom utilization | 20-68% | > 70% |