Fluidity Replication: results (2026-10-14 18:29 Halifax).
A toy model: structure inside the model, not evidence about cells. Histories, not memory: differences between routes are between intervention histories and the states they leave.
Fresh starts 115 to 134: 20 starts, 10 moments each, 200 planned moments. Every number comes from these starts; the pilot's values appear only beside them, for comparison.
Reproduced before reading on: untouched states at R from the base runs against the report's descriptions (coherence, slower half, faster half) (largest difference 3.1e-16); coherence from the saved states against the base runs' record at T and R (largest difference 1.4e-17); the moment files' untouched coherence against the base runs (largest difference 0.0e+00); the three contributions adding up to the group's coherence (largest difference 2.1e-17); the recorded parts against Check 3's calculation (largest difference 2.7e-16) and adding up (largest 5.3e-16).
Failed checks: none

Primary verdicts. Each hypothesis is judged on its own, with no combined verdict. Smallest effect of interest 0.10; lower bound = gap - 3 bars; upper bound = gap + 2 bars. PASS: lower bound above zero and gap at least 0.10. FAIL: upper bound below 0.10. Otherwise INCONCLUSIVE.
  Hypothesis 1, coherence distribution (the share gap): PASS (clearly above zero and at least 0.10)
    slower half's share of the extra at R: medium 0.999 +- 0.003; push 0.688 +- 0.019
    gap (medium minus push): +0.311 +- 0.018; lower bound +0.258, upper bound +0.346
    like-for-like moments: 198 of 200 (2 not; more than 20 would make it inconclusive); starts without any: none
  Hypothesis 2, persistence (the persistence gap): PASS (clearly above zero and at least 0.10)
    coefficients over (t, t + 100]: slower half +0.318 +- 0.031 (+10.2 bars); faster half +0.074 +- 0.039 (+1.9 bars)
    gap (slower minus faster): +0.243 +- 0.038; lower bound +0.130, upper bound +0.319

Beside hypothesis 1 (reported; they don't decide its verdict)
  from the exact contributions: medium 1.120 +- 0.014; push 0.499 +- 0.000; gap +0.621 +- 0.014 (+43.4 bars); clearly opposite in sign: no
  conservative, all 200 planned moments (the push counted like the medium where not like for like): medium 0.999 +- 0.003; push 0.689 +- 0.018; gap +0.310 +- 0.018 (+17.7 bars)
  the medium's raised-minus-lowered coherence difference at R, averaged over moments: like-for-like moments +0.02654 (198); others -0.00261 (2)
  directions not built: 2, with reasons:
    start 124, moment 9, raised: group A: no move up to a quarter turn reaches 0.033 x R
    start 126, moment 7, raised: sham outside its band (1.0000, 1.0507 x R)

Secondary: delayed perturbation growth, by the pilot's own analysis function (descriptions, never PASS or FAIL), each beside the pilot's value
  the medium's S (later growth against untouched): +0.00238 +- 0.00030 (+7.9 bars); pilot +0.00303 +- 0.00036 (+8.4 bars): same sign and clear
  the medium's P: -0.00212 +- 0.00065 (-3.2 bars); pilot -0.00237 +- 0.00070 (-3.4 bars): same sign and clear
  D_raised: -0.00450 +- 0.00078 (-5.8 bars); pilot -0.00541 +- 0.00088 (-6.1 bars): same sign and clear
  D_lowered: +0.00025 +- 0.00065 (+0.4 bars); pilot +0.00066 +- 0.00067 (+1.0 bars): same sign, not clear
  the pushes against their shams: S: +0.00201 +- 0.00027 (+7.3 bars); pilot +0.00186 +- 0.00070 (+2.7 bars): same sign and clear
  the pushes against their shams: P: -0.00245 +- 0.00047 (-5.2 bars); pilot -0.00265 +- 0.00038 (-6.9 bars): same sign and clear
  the pushes against untouched: S: +0.00123 +- 0.00028 (+4.4 bars); pilot +0.00127 +- 0.00050 (+2.6 bars): same sign and clear
  the pushes against untouched: P: -0.00140 +- 0.00055 (-2.5 bars); pilot -0.00115 +- 0.00045 (-2.5 bars): same sign, not clear
  the shams against untouched, raised: +0.00183 +- 0.00050 (+3.7 bars); pilot +0.00209 +- 0.00052 (+4.0 bars): same sign and clear
  the shams against untouched, lowered: +0.00027 +- 0.00042 (+0.6 bars); pilot +0.00092 +- 0.00051 (+1.8 bars): same sign, not clear
  medium minus push, like for like: S difference: +0.00117 +- 0.00032 (+3.7 bars); pilot +0.00181 +- 0.00051 (+3.6 bars): same sign and clear
  medium minus push, like for like: P difference: -0.00076 +- 0.00035 (-2.2 bars); pilot -0.00126 +- 0.00040 (-3.1 bars): same sign, not clear
  persistence over (R + 200, R + 500]: medium: +0.37837 +- 0.02162 (+17.5 bars); pilot +0.34996 +- 0.01783 (+19.6 bars): same sign and clear
  persistence: push: +0.32448 +- 0.01994 (+16.3 bars); pilot +0.29815 +- 0.03147 (+9.5 bars): same sign and clear
  persistence: difference (medium minus push): +0.05389 +- 0.02708 (+2.0 bars); pilot +0.05182 +- 0.03232 (+1.6 bars): same sign, not clear
  lasting: the medium's last block: +0.00574 +- 0.00063 (+9.2 bars); pilot +0.00696 +- 0.00066 (+10.5 bars): same sign and clear
  lasting: the medium's decline: +0.00834 +- 0.00059 (+14.0 bars); pilot +0.00665 +- 0.00099 (+6.7 bars): same sign and clear
  lasting: the push's decline: +0.00568 +- 0.00083 (+6.9 bars); pilot +0.00471 +- 0.00080 (+5.9 bars): same sign and clear
  shapes: medium same direction (pilot same direction); pushes against shams non-monotonic (pilot non-monotonic); pushes against untouched mixed (pilot no clear effect)
  the medium's S ruled out at half the benchmark: no (pilot no)
  readings: differs, same direction (pilot: differs, same direction)

Exploratory: does slower-half concentration explain the route difference in persistence? (exploratory, not a test; the re-weighting assumes each half's extra fades the same way whatever the other half holds)
  (R+0, R+100]: route difference +0.210 +- 0.024 (+8.7 bars); persistence medium 0.875, push 0.665, push re-weighted 2.494 +- 63.157; share of the difference closed +8.70 +- 274.40; slower half kept: medium 0.876, push 1.485 (difference -0.608 +- 0.034): consistent with composition
  (R+100, R+200]: route difference +0.137 +- 0.027 (+5.0 bars); persistence medium 0.685, push 0.548, push re-weighted 2.219 +- 64.426; share of the difference closed +12.17 +- 450.52; slower half kept: medium 0.698, push 1.243 (difference -0.545 +- 0.039): consistent with composition
  (R+200, R+300]: route difference +0.098 +- 0.033 (+2.9 bars); persistence medium 0.530, push 0.432, push re-weighted 1.795 +- 57.028; share of the difference closed +13.96 +- 908.50; slower half kept: medium 0.553, push 0.998 (difference -0.446 +- 0.061): nothing clear to explain
  (R+300, R+400]: route difference +0.066 +- 0.032 (+2.1 bars); persistence medium 0.389, push 0.323, push re-weighted 1.411 +- 50.352; share of the difference closed +16.40 +- 1099.66; slower half kept: medium 0.417, push 0.762 (difference -0.345 +- 0.056): nothing clear to explain
  (R+400, R+500]: route difference -0.002 +- 0.032 (-0.1 bars); persistence medium 0.216, push 0.218, push re-weighted 1.044 +- 33.721; share of the difference closed -352.62 +- 2217.51; slower half kept: medium 0.241, push 0.531 (difference -0.289 +- 0.058): nothing clear to explain
  (R+200, R+500]: route difference +0.054 +- 0.027 (+2.0 bars); persistence medium 0.378, push 0.324, push re-weighted 1.417 +- 47.002; share of the difference closed +20.27 +- 1161.75; slower half kept: medium 0.404, push 0.764 (difference -0.360 +- 0.049): nothing clear to explain

Also reported (descriptions)
  where natural order sits (Check 3's question 1): more in the slower half than an even spread; shares: slower half +0.848 +- 0.028 (+29.9 bars); faster half +0.150 +- 0.026 (+5.8 bars); free turners +0.002 +- 0.006 (+0.3 bars); slower half minus an even spread (0.484): +0.364 +- 0.028 (+12.8 bars)
  the natural share beside the routes (question 2): natural 0.942 +- 0.016; medium 0.999 +- 0.003; push 0.688 +- 0.019: between the two; from states at R alone +0.947 +- 0.024 (+38.9 bars), at T alone +0.955 +- 0.024 (+39.5 bars)
  hypothesis 2's coefficients, block by block (slower half; faster half; free turners; slower minus faster):
    (t+0, t+100]: +0.318 +- 0.031 (+10.2 bars); +0.074 +- 0.039 (+1.9 bars); +0.108 +- 0.138 (+0.8 bars); +0.243 +- 0.038 (+6.5 bars)
    (t+100, t+200]: +0.219 +- 0.029 (+7.6 bars); +0.055 +- 0.034 (+1.6 bars); +0.079 +- 0.132 (+0.6 bars); +0.165 +- 0.034 (+4.8 bars)
    (t+200, t+300]: +0.166 +- 0.023 (+7.3 bars); +0.032 +- 0.037 (+0.9 bars); +0.012 +- 0.117 (+0.1 bars); +0.134 +- 0.039 (+3.4 bars)
    (t+300, t+400]: +0.088 +- 0.019 (+4.6 bars); -0.014 +- 0.029 (-0.5 bars); +0.042 +- 0.134 (+0.3 bars); +0.103 +- 0.032 (+3.2 bars)
    (t+400, t+500]: -0.001 +- 0.020 (-0.0 bars); -0.042 +- 0.034 (-1.2 bars); +0.106 +- 0.155 (+0.7 bars); +0.041 +- 0.041 (+1.0 bars)
  share of a coherence excess still there (natural slope; medium; push):
    (0, 100] after t or R: +0.290 +- 0.031 (+9.3 bars); +0.875 +- 0.027 (+31.9 bars); +0.665 +- 0.018 (+37.4 bars)
    (100, 200] after t or R: +0.201 +- 0.029 (+6.9 bars); +0.685 +- 0.027 (+25.0 bars); +0.548 +- 0.018 (+30.1 bars)
    (200, 300] after t or R: +0.145 +- 0.022 (+6.6 bars); +0.530 +- 0.025 (+21.2 bars); +0.432 +- 0.026 (+16.3 bars)
    (300, 400] after t or R: +0.072 +- 0.017 (+4.1 bars); +0.389 +- 0.026 (+14.8 bars); +0.323 +- 0.023 (+14.1 bars)
    (400, 500] after t or R: -0.015 +- 0.016 (-0.9 bars); +0.216 +- 0.021 (+10.0 bars); +0.218 +- 0.023 (+9.6 bars)
  arms separately against untouched (slower half's share of the change): medium raised +0.997 +- 0.003 (+310.4 bars); medium lowered +1.003 +- 0.004 (+229.4 bars); push raised +0.586 +- 0.011 (+52.4 bars); push lowered +1.327 +- 0.101 (+13.1 bars)
  the medium over all 200 moments: slower half's share +0.999 +- 0.003 (+305.1 bars)
  states at R (coherence; slower half; faster half; |z2|; |z3|; |z4|; A-B timing):
    untouched: 0.0239; 0.0464; 0.0192; 0.0116; 0.0127; 0.0124; 0.8389
    medium raised: 0.0433; 0.0896; 0.0193; 0.0118; 0.0125; 0.0126; 0.4713
    medium lowered: 0.0171; 0.0312; 0.0192; 0.0116; 0.0127; 0.0128; 1.2312
    push raised: 0.0436; 0.0657; 0.0327; 0.0121; 0.0128; 0.0125; 0.8410
    push lowered: 0.0171; 0.0397; 0.0209; 0.0116; 0.0128; 0.0124; 0.8419
    sham raised: 0.0239; 0.0466; 0.0192; 0.0119; 0.0128; 0.0123; 0.8418
    sham lowered: 0.0239; 0.0466; 0.0192; 0.0117; 0.0128; 0.0125; 0.8414
  phase distance between the medium's and the push's states at R: raised +0.705 +- 0.018 (+38.4 bars); lowered +0.419 +- 0.013 (+32.5 bars)

Limits
  A toy model at one pull, size and window, with one push method. Natural states within a start share one trajectory; the start is the unit throughout.
  A half's own coherence responds only weakly to small amounts of order near the random-phase floor; the exact contributions beside hypothesis 1 don't have that limit.
  The exploratory reading isn't a controlled comparison: no arm changes the composition with everything else fixed, and the routes also differ in A-B timing and phase arrangement at R.

Compute: 18 pieces, 46.6 minutes of foreground time. The report was computed twice, with identical numbers.
Plan, code, model files, the pilot's freeze and records as locked (2026-10-14 17:34 Halifax); Colson's go recorded 2026-10-14 17:40 Halifax; the cause test's locked files unchanged: yes.