Fresh-gap calibration (prelock_checks_plan.md, part 1): old starts 1 to 18, untouched only
made 2026-10-13 10:33 Halifax; not a locked plan; no new start; no change to any copy

Reading, with the revised exactness check (V1-revised, adopted after V1 as preregistered failed; prelock_amendment.md): FAIL
Reading with V1 as preregistered: NO READING
What the reading leads to: the fresh gap is not the primary outcome. The carried gap is not put in its place automatically (amendment 2): the alternatives, with their measurement-history biases, go to Colson for review, and any new outcome gets its own check on old starts before any lock.

V1 as preregistered, every re-run reproduces its saved run exactly: NOT met (6 of 18 starts; coherence in 6, the gap in 6, the settling in 18)
   first difference from the saved run, by start (gap; coherence): 1: t = 1185, t = 2151; 2: t = 2065, t = 3396; 3: t = 1680, t = 2258; 4: t = 2300, t = 3479; 5: t = 2440, t = 3150; 6: t = 2625, t = 3047; 7: t = 320, t = 1834; 8: t = 625, t = 1527; 9: t = 1440, t = 1845; 10: t = 1020, t = 1615; 11: t = 1765, t = 2812; 12: t = 1985, t = 2694; 13: none, none; 14: none, none; 15: none, none; 16: none, none; 17: none, none; 18: none, none
V1-revised (not the preregistered check), every re-run agrees to the last digit with the cause test's unchanged base-run code run in one go on the same machine, in one numerical environment throughout: met (18 of 18 agree; one environment: True)
   the environment: INTEL(R) XEON(R) PLATINUM 8581C CPU @ 2.30GHz, Python 3.11.14, glibc 2.40, numpy 2.4.3 (X86_V3, X86_V4, AVX512_ICL, AVX512_SPR); one boot id in every full record: True; pieces 1, 2, 3, 4 ran before the environment was recorded, and every later record shows the container running without a restart since 07:40:54 Halifax, before piece 1: True
   bitwise agreement on one machine shows reproducibility, not numerical accuracy; no step-size check was made (prelock_amendment.md)
V2 every fresh gap measured at all 60 put-backs, every number finite: met
C1 convergence: 22.6% lined up within the window (needs 90%); median alignment inf units (needs at most 100): NOT met
C2 stability: direction noise 0.00196 against the natural spread of W 0.00664, ratio 0.30 (needs at most 0.5): met
C3 agreement: pooled within-start r(G, W) = 0.873 ± 0.007 (needs at least 0.8): met
C4 interpretability: kappa = 0.579 ± 0.024 (needs 0.75 to 1.25, error bar at most 0.25): NOT met

Also reported (no criterion):
  b_M with the fresh gap -0.454 ± 0.045; with the carried gap -0.784 ± 0.057
  mean growth: fresh 0.0137, carried 0.0138; fresh minus carried -0.00010 ± 0.00020
  share lined up by 25 units 0%, 50 units 0%, 100 units 0%, 150 units 3%, 200 units 7%, 300 units 23%
  median overlap at 5: 0.014, 10: 0.028, 25: 0.086, 50: 0.198, 100: 0.352, 150: 0.456, 200: 0.538, 300: 0.666
  mean log growth per put-back, fresh minus carried, at 5: -0.0504, 10: -0.0314, 25: -0.0095, 50: +0.0091, 100: +0.0046, 150: +0.0027, 200: -0.0013, 300: -0.0004
  summed over the first 100 units -0.036, over the last 200 +0.007
  first 100 units: fresh minus carried -0.00036 ± 0.00033, r = 0.418
  last 200 units: fresh minus carried +0.00003 ± 0.00016, r = 0.914
  old 1-6: kappa 0.599, r 0.875, lined up 24%, median alignment inf
  old 7-12: kappa 0.521, r 0.882, lined up 19%, median alignment inf
  old 13-18: kappa 0.647, r 0.862, lined up 25%, median alignment inf

Seen before this check was complete (prelock_amendment.md; dryruns/calibration_dryrun.txt): the dry run on old start 1, fresh copies made at t = 200 and 250, outside these moments, gave G = 0.0049, 0.0055 and 0.0071 against W = 0.0028 and 0.0024, and overlaps of 0.098 and 0.255 after 300 units.

Compute: 208.7 CPU minutes (696 s per start); 11 pieces
Parts (in generated-artifacts/cause-prelock-2026-10-13/calibration): 18 of 18 match the run log
The cause test's locked files unchanged: True; its folder and the exploratory folder unchanged: True