strict-N at emit: 22,066 · 17 locks verified · every campaign pre-registered before evaluation. Negative results are published at equal prominence; this annex never contains a deployment recommendation.
locked 2026-08-03 · sha256[:8] a7a6ab40 · CLOSED
NO SELECTION JUSTIFIED — Is any parameter setting in this family better than the already-validated center?
| configs evaluated | 681 (0 errors) |
| stability regions | 1 region / 647 configs — landscape flat |
| PBO (overfit probability) | 0.854 — ranking is noise |
| holdout confirmation | rho +0.37 < 0.7 bar — IS ranking did not persist |
Uniformly profitable in-sample, distinguishable nowhere: picking a “best” config over the family center is validating luck. The center (already validated on its own layers) stands as the only defensible representative.
locked 2026-08-09 · sha256[:8] a399ec61 · CLOSED
NO DURABLE REGION — Does ANY parameter region make this logic work on MES?
| configs evaluated | 329 (0 errors) |
| best config found | +$905 over 4.5y — economically nil |
| stability regions | 0 — zero of 38 candidates passed L9 |
| PBO (overfit probability) | 0.759 — ranking is noise |
| family center on holdout | −$4,045 (PF 0.75) — loses both sides of the firewall |
65 of 329 configs profitable in-sample, all hugging the widest-gap grid edge where the strategy simply trades less. No stability region exists; nothing earned holdout validation. The pre-committed “no durable region” branch fired — the edge, as parameterized, does not transfer to MES.
locked 2026-08-09 · sha256[:8] 1fad5764 · CLOSED
NO DURABLE REGION — Does ANY parameter region make this logic work on MYM?
| configs evaluated | 270 (0 errors) |
| best config found | +$636 over 4.5y — economically nil |
| stability regions | 0 — zero of 4 candidates passed L9 |
| PBO (overfit probability) | 0.343 — loss-ordering persists; nothing persists as profit |
| family center on holdout | −$12 (PF 1.00) — breakeven — costs consume the signal |
9 of 270 configs profitable in-sample — isolated points, not regions; every one surrounded by losers. The center lands at breakeven on holdout: gross signal ≈ cost drag, confirming the pre-registered warning that $3.49/ct ≈ 7 Dow points at $0.50/pt may preclude any edge at this size. The pre-committed “no durable region” branch fired.
locked 2026-08-10 · sha256[:8] b4f4f609 · CLOSED
NO DURABLE REGION — Does ANY parameter region make this logic work on MGC?
| configs evaluated | 255 (0 errors) |
| best config found | −$1,152 over 13y |
| stability regions | 0 — zero of 0 candidates passed L9 |
| PBO (overfit probability) | 0.553 — ranking within a uniformly losing landscape |
| family center on holdout | +$9,420 (PF 1.31) — pre-recorded regime windfall, not searchable skill |
0 of 255 configs profitable over 13 years — the most decisive negative in the program; S3 had no candidates to scan. The center's holdout WIN (+$9,420) is the gold 2024-25 run the lock pre-recorded as unsellable-as-skill: it lost for 13 years, then rode a once-in-a-generation regime. The pre-committed “no durable region” branch fired.
locked 2026-08-10 · sha256[:8] 000cc348 · CLOSED
NO DURABLE REGION — Does ANY parameter region make this logic work on MCL?
| configs evaluated | 458 (0 errors) |
| best config found | +$7,050 over 2.4y |
| stability regions | 8 L9 region(s), none durable on holdout (R1) |
| PBO (overfit probability) | 0.314 — within-regime ranking persists; the regime didn't |
| family center on holdout | −$2,714 (PF 0.83) — IS winner, holdout loser — regime flip |
The full overfitting arc: 425 of 458 configs profitable in-sample, 163 L9-eligible, one 156-config region — and EVERY sampled member lost or broke even on holdout (R1 false for all 8 regions; member DSRs 0.00–0.01). In-sample ranking persisted (rho +0.90) as an ordering of LOSSES: the 2021-23 crude regime did not survive into 2024+. The holdout firewall did exactly its job. The pre-committed “no durable region” branch fired.