Window-only check (step 2 of plan-final.md), 2026-10-15 10:04 Halifax.
Locked plan: plan-final.md (SHA-256 0e3f8f114ed44eade7e211317217f23e27c1075008e561c2b07ea11645823121); lock: lock.json (SHA-256 3c7f70973247a327067a1817766b0dae3e0d72154f1de7785df7213ff271b23d); code: scattering_side.py (SHA-256 2a20b872d594a53b50f7b124e700c58e0e4efe12377324ac17105818f9e94634), the code whose step-1 checks all passed (2026-10-15 09:57 Halifax).
Old starts 16 and 17 (seeds 15 and 16), 10 moments each by the medium pilot's rule and draws: 20 moments. Base runs made afresh by the medium pilot's own function, keeping the states at T and R and no gap growth. At each moment, from the state at T, the first copy only, 200 units: the untouched window, the raised pull (K x 1.005 = 1.9094999999999998), arm F (c_f = 0.6), arm F' (c_f = 0.4) and arm U's search. Nothing after R was run or looked at.
No labels are applied to these numbers, and they never enter the pilot's readings. The levels, the rule and its thresholds stay as written whatever this shows.

1. Arm U's search (c found so that C at R matches the raised pull's within a relative 1%)
   matched at 10 of 20 moments; the plan asks for 18 or more: NOT met
   runs per moment: 1 run: 3, 4 runs: 1, 5 runs: 2, 6 runs: 3, 7 runs: 2, 8 runs: 9
   accepted c: 0.99213 to 0.99937 (median 0.99526), a narrowing of 0.06% to 0.79%
   each group's own mismatch at the accepted c (relative to the raised pull's R_g): A -0.2286 to +0.4252, B -0.6202 to +0.6890
   unmatched: start 16, moment 1 (T = 2045): the secant points to c = 1.002746, not below 1 (closest run c = 0.99527, f = +0.00623)
   unmatched: start 16, moment 3 (T = 2705): the secant points to c = 1.004937, not below 1 (closest run c = 0.99527, f = -0.00420)
   unmatched: start 16, moment 6 (T = 3695): no run within 1% in 8 runs (closest run c = 0.99481, f = +0.00131)
   unmatched: start 16, moment 9 (T = 4685): no run within 1% in 8 runs (closest run c = 0.99563, f = +0.00056)
   unmatched: start 17, moment 1 (T = 2195): no run within 1% in 8 runs (closest run c = 0.99678, f = +0.00226)
   unmatched: start 17, moment 2 (T = 2525): no run within 1% in 8 runs (closest run c = 0.99593, f = +0.00097)
   unmatched: start 17, moment 3 (T = 2855): no run within 1% in 8 runs (closest run c = 0.99331, f = -0.00163)
   unmatched: start 17, moment 5 (T = 3515): no run within 1% in 8 runs (closest run c = 0.99936, f = +0.00035)
   unmatched: start 17, moment 6 (T = 3845): no run within 1% in 8 runs (closest run c = 0.99577, f = +0.00131)
   unmatched: start 17, moment 9 (T = 4835): no run within 1% in 8 runs (closest run c = 0.99553, f = +0.00113)

2. Comparisons B (arm F, c_f = 0.6) and B' (arm F', c_f = 0.4): a first look
   C at R (the average of the two groups' coherence), averaged over each start's 10 moments:
     start 16: untouched 0.02518, raised pull 0.04561, arm U 0.04600, arm F (0.6) 0.02787, arm F' (0.4) 0.04106
     start 17: untouched 0.02459, raised pull 0.04544, arm U 0.04574, arm F (0.6) 0.02570, arm F' (0.4) 0.03251
   B1, the gain at R as a share of the raised pull's gain (each averaged over moments and starts, then divided):
     arm F (0.6): start 16 +0.132, start 17 +0.053; both +0.092
     arm F' (0.4): start 16 +0.777, start 17 +0.380; both +0.577
     (the raised pull's gain at R: start 16 +0.02043, start 17 +0.02085)
   B2, the faster half's own coherence at R, arm minus untouched (the two groups' average):
     arm F (0.6): start 16 +0.00789, start 17 +0.00526; both +0.00657
     arm F' (0.4): start 16 +0.07435, start 17 +0.03780; both +0.05607
     (untouched: the faster half's own coherence at R averages 0.01880)
   Rough error bars projected to the pilot's 12 starts (two starts can't estimate the spread well; for orientation only):
     arm F (0.6): B1 0.016 from the two starts, 0.046 from the 20 moments scaled to 120; B2 0.00054 and 0.00100
     arm F' (0.4): B1 0.081 from the two starts, 0.058 from the 20 moments scaled to 120; B2 0.00746 and 0.00299
     For scale: the margins are 0.25 (B1) and 0.005 (B2). A reading of zero lies, with 2 bars either side, inside its margin only if the bar is under 0.125 (B1) or 0.0025 (B2).

3. The rhythm rule's numbers, read over (T + 150, R]: each direction of each group at each moment (80 per arm), and each group at each moment (40 per arm)
   untouched      rhythms  0 of 80; group-moments: pair 0, any 0, steady 0 of 40; L median 1.10 (largest 1.35); S median 0.0098 (largest 0.0278); |Omega*| of rhythms: none
   raised pull    rhythms  0 of 80; group-moments: pair 0, any 0, steady 0 of 40; L median 1.15 (largest 1.35); S median 0.0096 (largest 0.0279); |Omega*| of rhythms: none
   arm U          rhythms  0 of 80; group-moments: pair 0, any 0, steady 0 of 40; L median 1.15 (largest 1.30); S median 0.0097 (largest 0.0266); |Omega*| of rhythms: none
   arm F (0.6)    rhythms  0 of 80; group-moments: pair 0, any 0, steady 0 of 40; L median 1.12 (largest 1.42); S median 0.0179 (largest 0.0518); |Omega*| of rhythms: none
   arm F' (0.4)   rhythms 68 of 80; group-moments: pair 28, any 40, steady 40 of 40; L median 1.92 (largest 2.63); S median 0.1220 (largest 0.1548); |Omega*| of rhythms: 0.964 (0.948 to 0.987)
   Shares, arm minus untouched (start 16, start 17; both):
     raised pull    E-pair +0.000, +0.000; +0.000; E-any +0.000, +0.000; +0.000; E-steady +0.000, +0.000; +0.000
     arm U          E-pair +0.000, +0.000; +0.000; E-any +0.000, +0.000; +0.000; E-steady +0.000, +0.000; +0.000
     arm F (0.6)    E-pair +0.000, +0.000; +0.000; E-any +0.000, +0.000; +0.000; E-steady +0.000, +0.000; +0.000
     arm F' (0.4)   E-pair +0.600, +0.800; +0.700; E-any +1.000, +1.000; +1.000; E-steady +1.000, +1.000; +1.000
   Untouched held a rhythm at 0 of 40 group-moments (the plan expects none), and steady turning (S >= 0.057) at 0.
   The most natural rates in use any interval holds, per direction (the rule's denominator): untouched [20], raised pull [20], arm U [20], arm F (0.6) [33], arm F' (0.4) [49]
   arm F' (0.4): slower-half oscillators turning inside a rhythm's interval: median 17, largest 30

4. Notes
   The untouched window on one copy against the base run (two copies): coherence identical to the last digit at 0 of 20 moments (largest difference 2.4e-14); the state at R identical at 0 of 20.
   Fresh base runs against the medium calibration's saved ones (same starts, same function): start 16: {'z': True, 'trace': True, 'kept_t': True, 'kept_u': True, 'kept_f': True}; start 17: {'z': True, 'trace': True, 'kept_t': True, 'kept_u': True, 'kept_f': True}
   Saved: 21.7 MB in generated-artifacts/scattering-side/window-check/ (the plan said about 12 MB; every pulled oscillator's turning rate and every unit's readings are kept).
   Compute: 2 pieces, 5.5 minutes of foreground time (7 workers at a time). Machine: INTEL(R) XEON(R) PLATINUM 8581C CPU @ 2.30GHz.
   Every piece ran under this plan, lock and code: yes.
   The cause test's locked files unchanged: True.
   Moments: start 16: T = 2045, 2375, 2705, 3035, 3365, 3695, 4025, 4355, 4685, 5015; start 17: T = 2195, 2525, 2855, 3185, 3515, 3845, 4175, 4505, 4835, 5165