Correct the k=20 description of the probe-then-commit bound

MANIFEST.md

@@ -48,7 +48,7 @@ Numeric headline audit: `docs/paper_combined/CLAIM_INVENTORY.md`.
 | Routing evaluation (13 PDEs, 450 runs) | `experiments/results/paper_combined/routing_evaluation/results.json` | `experiments/scripts/paper_a_routing/evaluate_routing_full.py` |
 | Scaling analysis (5→9→13 PDEs) | `experiments/results/paper_combined/routing_evaluation/scaling_ablation.json` | `experiments/scripts/paper_a_routing/scaling_and_ablation.py` |
 | Feature ablation (nested CV) | `experiments/results/paper_combined/routing_evaluation/scaling_ablation.json` | Same script (Part 2) |
-| Diagnostic bound (33/52 = 63.5% overall; 33/39 = 84.6% for k ∈ {10,50,100}; universal failure at k=20; fails on Burgers1D and Burgers1D-lownu at every defined k) | `experiments/results/paper_combined/routing_evaluation/regret_bound_validation.json` | `experiments/scripts/paper_combined/validate_regret_bound.py` |
+| Diagnostic bound (33/52 = 63.5% overall; 33/39 = 84.6% for k ∈ {10,50,100}; the k=20 rows were never checked rather than failed -- observed regret was only looked up for k in {10,50,100}, so the comparison returned NaN; supplying the regret already recorded in the early-predictor sweep, the bound holds 6/6 at k=20 with mean observed regret 0.00331, in line with the other windows. The genuine failures are Burgers1D and Burgers1D-lownu, which violate the bound at every evaluated k) | `experiments/results/paper_combined/routing_evaluation/regret_bound_validation.json` | `experiments/scripts/paper_combined/validate_regret_bound.py` |
 | Meta-router (GBR+PDE and RF+PDE-top5: 84.6%, 11/13 held-out PDEs; full RF+PDE: 61.5%) | `experiments/results/paper_combined/meta_router/results.json` | `experiments/scripts/paper_combined/evaluate_meta_router.py` |
 | Held-out generalization (3 PDEs) | `experiments/results/paper_combined/routing_evaluation/holdout.json` | `experiments/scripts/paper_a_routing/holdout_validation.py` |
 | Family-macro accuracy (LOPO aggregated by 8 families): physics-final 81.2% / GBR 68.8% / RF 54.2% / Ridge 45.8% | `experiments/results/paper_combined/routing_evaluation/results.json` | `experiments/scripts/paper_a_routing/family_holdout_analysis.py` |