Step-1 code checks, 2026-10-15 09:57 Halifax. Equality and mechanics only: no outcome of any change in the model is read, shown or kept. Old start 1 (seed 0) and pilot start 103 (seed 102); the rule on made-up phase histories. Locked plan: plan-final.md; lock: lock.json.

0. The code's settings are the lock's, and plan-final.md is the locked plan: PASSED
     settings equal after a JSON round trip; plan SHA-256 0e3f8f114ed44eade7e211317217f23e27c1075008e561c2b07ea11645823121
1. Arm U with c = 1 and arm F with c_f = 1 match the untouched run bit for bit (old start 1, from t = 100: 20 units changed and 20 restored, both copies): PASSED
     arm U, c = 1: {'rates': True, 'coherence': True, 'gap': True, 'states': True, 'turning': True}
     arm F, c_f = 1: {'rates': True, 'coherence': True, 'gap': True, 'states': True, 'turning': True}
     this file's untouched arm against the medium pilot's own branch(): {'coherence': True, 'gap': True, 'pull': True, 'states': True}
2. After R the rates are the originals bit for bit, and the run continues exactly as the unchanged model's step from the same state (coherence, gap growth and end state, to the last digit): PASSED
     raised pull: {'rates_after_R_original': True, 'coherence': True, 'gap': True, 'states': True, 'change_took_effect': True, 'pull_as_scheduled': True}
     arm U (c = 0.99527): {'rates_after_R_original': True, 'coherence': True, 'gap': True, 'states': True, 'change_took_effect': True, 'pull_as_scheduled': True}
     arm F (0.6): {'rates_after_R_original': True, 'coherence': True, 'gap': True, 'states': True, 'change_took_effect': True, 'pull_as_scheduled': True}
     arm F' (0.4): {'rates_after_R_original': True, 'coherence': True, 'gap': True, 'states': True, 'change_took_effect': True, 'pull_as_scheduled': True}
3. The pulled and free oscillators and both halves are unchanged, the rates' mean stays 0 (within 1e-12), and every rank and sign is kept: PASSED
     arm U, c = 0.99527: {'pulled_set_is_nudge_tests': True, 'halves_unchanged': True, 'directions_unchanged': True, 'mean_near_zero': True, 'ranks_kept': True, 'signs_kept': True} (mean -3.847e-13)
     arm U, c = 0.95: {'pulled_set_is_nudge_tests': True, 'halves_unchanged': True, 'directions_unchanged': True, 'mean_near_zero': True, 'ranks_kept': True, 'signs_kept': True} (mean -3.665e-13)
     arm F, c_f = 0.6: {'pulled_set_is_nudge_tests': True, 'halves_unchanged': True, 'directions_unchanged': True, 'mean_near_zero': True, 'ranks_kept': True, 'signs_kept': True} (mean -3.861e-13)
     arm F', c_f = 0.4: {'pulled_set_is_nudge_tests': True, 'halves_unchanged': True, 'directions_unchanged': True, 'mean_near_zero': True, 'ranks_kept': True, 'signs_kept': True} (mean -3.870e-13)
4. In arms F and F', the slower half's and free turners' rates are unchanged bit for bit, and each faster rate is sign(omega) (b + c_f (|omega| - b)) to rounding: PASSED
     c_f = 0.6: {'slower_half_and_free_unchanged': True, 'formula_to_rounding': True} (largest difference 1.78e-15)
     c_f = 0.4: {'slower_half_and_free_unchanged': True, 'formula_to_rounding': True} (largest difference 1.78e-15)
5. Arm U's 200-unit window matches its exact mirror (K and eps raised by 1 / c, steps of 0.025 c) within 1e-9 in every phase and each group's coherence, every unit: PASSED
     c = 0.99527: largest phase difference 5.69e-11 rad, largest coherence difference 2.78e-14 (within 1e-9); phase difference by time: t = 1: 2.8e-13, t = 10: 2.8e-12, t = 50: 1.4e-11, t = 100: 2.8e-11, t = 150: 4.3e-11, t = 200: 5.7e-11
     c = 0.95: largest phase difference 4.73e-10 rad, largest coherence difference 3.72e-13 (within 1e-9); phase difference by time: t = 1: 7.2e-14, t = 10: 7.1e-13, t = 50: 8.5e-12, t = 100: 2.1e-11, t = 150: 1.3e-10, t = 200: 7.2e-11
6. The raised-pull arm reproduces the medium pilot's saved arrays for pilot start 103, moment 1 (T = 2,055), bit for bit: coherence every unit and gap growth, 700 units: PASSED
     {'coherence': True, 'gap': True}; T = 2055 (saved 2055), size 0.005; largest coherence difference 0.0e+00
7. The untouched rerun reproduces the pilot's base run for start 103 bit for bit over the same 700 units (coherence every unit and gap growth): PASSED
     {'coherence': True, 'gap': True}; T = 2055 (saved 2055), size 0.005; largest coherence difference 0.0e+00
8. Turning rates from readings every unit equal those from readings every step (40 per unit) within 1e-12, over (10, 20] of each short arm; every reading present and finite, and the rule runs end to end (values not shown): PASSED
     untouched: largest difference 2.5e-14; readings {'shapes': True, 'finite': True, 'sums_add_up': True, 'rule_runs': True}
     u1: largest difference 2.5e-14; readings {'shapes': True, 'finite': True, 'sums_add_up': True, 'rule_runs': True}
     f1: largest difference 2.5e-14; readings {'shapes': True, 'finite': True, 'sums_add_up': True, 'rule_runs': True}
     raised: largest difference 2.1e-14; readings {'shapes': True, 'finite': True, 'sums_add_up': True, 'rule_runs': True}
     u: largest difference 2.8e-14; readings {'shapes': True, 'finite': True, 'sums_add_up': True, 'rule_runs': True}
     f06: largest difference 1.6e-14; readings {'shapes': True, 'finite': True, 'sums_add_up': True, 'rule_runs': True}
     f04: largest difference 1.2e-14; readings {'shapes': True, 'finite': True, 'sums_add_up': True, 'rule_runs': True}
     pilot_raised (700 units, read (150, 200]): readings {'shapes': True, 'finite': True, 'sums_add_up': True, 'rule_runs': True}
     pilot_untouched (700 units, read (150, 200]): readings {'shapes': True, 'finite': True, 'sums_add_up': True, 'rule_runs': True}
9. The rule on made-up phase histories, no model run: a planted pair found as a pair at the planted rate (within 0.03), a one-way cluster in that direction only, and six imitations all failing: PASSED
     planted pair (expected: a pair at +0.96 and -0.96): forward L 5.00 (100 of 20), S 0.1032, Omega* +0.960, rhythm yes; backward L 5.00 (100 of 20), S 0.0983, Omega* -0.960, rhythm yes: as expected
     one-way cluster (expected: a rhythm forward only, at +0.96): forward L 5.00 (100 of 20), S 0.0994, Omega* +0.960, rhythm yes; backward L 1.00 (20 of 20), S 0.0063, Omega* -0.966, rhythm no: as expected
     random phases (expected: no rhythm either way): forward L 1.00 (20 of 20), S 0.0044, Omega* +0.966, rhythm no; backward L 1.00 (20 of 20), S 0.0094, Omega* -0.966, rhythm no: as expected
     brief alignment (expected: no rhythm either way): forward L 1.15 (23 of 20), S 0.0269, Omega* +0.988, rhythm no; backward L 1.25 (25 of 20), S 0.0259, Omega* -1.006, rhythm no: as expected
     fixed lean (expected: no rhythm either way): forward L 1.10 (22 of 20), S 0.0108, Omega* +0.988, rhythm no; backward L 1.10 (22 of 20), S 0.0055, Omega* -0.953, rhythm no: as expected
     beat (expected: no rhythm either way): forward L 1.00 (20 of 20), S 0.0093, Omega* +0.966, rhythm no; backward L 1.00 (20 of 20), S 0.0140, Omega* -0.966, rhythm no: as expected
     crowded rates, unlocked (expected: no rhythm either way): forward L 1.00 (49 of 49), S 0.0086, Omega* +0.965, rhythm no; backward L 1.00 (49 of 49), S 0.0141, Omega* -0.965, rhythm no: as expected
     cluster at 0.5 (expected: no rhythm either way): forward L 0.85 (17 of 20), S 0.0080, Omega* +1.128, rhythm no; backward L 0.85 (17 of 20), S 0.0058, Omega* -1.128, rhythm no: as expected
10. The cause test's locked files unchanged: PASSED
     none changed

Reported only (not checks):
  The untouched short arm run with one copy against the first copy of the run with two: identical to the last digit False (largest coherence difference 2.7e-16; end state identical False). Step 2 runs one copy.
  The most natural rates any interval holds, per direction (forward, backward): {'untouched': [20, 20], 'F (0.6)': [33, 33], "F' (0.4)": [49, 49]}; the plan states 20, 33 and 49: matches.
  The faster half's median and largest |natural rate|: {'untouched': [2.18, 19.72], 'F (0.6)': [1.69, 12.21], "F' (0.4)": [1.44, 8.46]}; the plan states {'untouched': [2.18, 19.72], 'F (0.6)': [1.69, 12.21], "F' (0.4)": [1.44, 8.46]}: matches.
  The medium pilot's numerical environment (its run log) and this run's agree: True.
  Free turners whose rate falls to 20 or below in arm U (they stay free, oscillator by oscillator): {'0.99527': 2, '0.95': 8}. b = 0.9502092363816494.
  Time: 0.0205 s per copy-unit with 8 tasks on 7 workers; projected for step 2: a base run about 252 s, a window moment with six 200-unit runs about 25 s on one worker. Task seconds: {'short_c': 6.5, 'short_b': 6.9, 'made_up': 1.1, 'short_a': 11.0, 'mirror_2': 11.5, 'mirror_1': 11.8, 'pilot_untouched': 16.6, 'pilot_raised': 16.7, 'base_run_to_t_100': 9.6}.

All checks passed: yes
Run under plan-final.md SHA-256 0e3f8f114ed44eade7e211317217f23e27c1075008e561c2b07ea11645823121, lock.json SHA-256 3c7f70973247a327067a1817766b0dae3e0d72154f1de7785df7213ff271b23d and scattering_side.py SHA-256 2a20b872d594a53b50f7b124e700c58e0e4efe12377324ac17105818f9e94634; 29 s in all.