| # | Claim | Outcome | Actual check |
|---|---|---|---|
| 1 | 152 tasks present in all seven arms, no duplicates | PASS | 152 tasks, 1064 rows |
| 2 | Sensitivity arm (TypeScript, JS stop rule) covers the same 152 tasks once each | PASS | 152 rows |
| 3 | Summary recomputes identically from raw rows | PASS | summary.json byte-equivalent after recompute |
| 4 | No hidden test line appears verbatim in any diagnostic fed back or any model reply | PASS | 0 attempts with a verbatim hidden-test line |
| 5 | Seeded replay of 49 saved final programs reproduces gate and test outcomes | PASS | 0 mismatches |