Code checks, 2026-10-14 12:19 Halifax. Old start 1 (seed 0), short runs. Equality and mechanics only: no outcome of any change is read, shown or kept (plan.md, revision 2).

0. The code's settings match the plan's numbers: yes
1. This file's base run against the cause test's unchanged base run, both unbroken, 400 units: identical to the last digit: {'coherence': True, 'gap': True, 'settling': True, 'states': True}
2. The same run stopped every 2 seconds (5 times) and gone on from its kept state: identical to one go: {'z': True, 'logs': True, 'trace': True, 'kept_u': True, 'kept_f': True}
3. A branch from t = 100 with the pull changed by zero, in two 50-unit stretches as a medium arm's change and restoration: reproduces the untouched run to the last digit: {'coherence': True, 'gap': True}
4. The pull raised 5% for 50 units, then restored: the restored stretch is the unchanged model run from the same state, to the last digit; the change took effect; the pull in use is as scheduled: {'restored_is_unchanged_model': True, 'change_took_effect': True, 'pull_as_scheduled': True}
5a. A push raising each group's coherence to 1.5 times, and its sham, on the state at t = 120: {'built': True, 'on_target': True, 'sham_in_band': True, 'same_sizes': True, 'gap_kept': True}
5b. A push lowering each group's coherence to 0.67 times, and its sham, on the state at t = 120: {'built': True, 'on_target': True, 'sham_in_band': True, 'same_sizes': True, 'gap_kept': True}
6. The pilot's moment task on a stand-in (T = 100, the pull changed 5% for 20 units, 50 after): runs end to end: {'shapes': True, 'restored': True, 'problem': 'none', 'pushes_built': [True, True], 'built_arms_complete': True, 'descriptions_present': True}
7. The calibration's moment task on the stand-in (20 units, every size both ways, first copy only): runs end to end: {'shapes': True, 'every_number_present': True, 'records': 12}
8. The cause test's locked files unchanged: True

Time: the cause test's base run took 18 s for 400 units after settling; scaled up, a pilot moment takes about 79 s and a calibration moment about 76 s on one worker (more with seven at once).
All checks passed: yes
Run under plan.md SHA-256 b2592cfea01be207022145da51d78b48c9f30dc1ff5dc53c0ae86026abe7f427 and medium_test.py SHA-256 6b75b7801692298dff0000f1e5fbebdae18285a8b1fedce8805f616e286b793e.