=== stress.c — LEAKED error-sourced temperature vs Mess3 valley ===
M=16 units, 4000 sweeps, window 10000, 5 seeds
optimal log-loss 0.7960 nats | symbol marginals 0.334 0.334 0.332
--- guard: did the test actually run? ---
[1] field mixed? A stress~field corr 0.704 (want <0.99) B corr 1.000 (want 1.00)
[2] uphill moves? A 45644 B 42338 B2 47935 (all must be >0)
[3] C schedule span 0.0003 (must be >0)
[4] generator marginals sane? yes
[5] mean T: A 3.48e-05 vs B2 3.48e-05 (B2 must match A; that is the point)
[6] positive control: TRUE-seeded unit ends at gap +0.0002 (must stay <+0.01)
guard PASSED.
PRIMARY — distinct belief states vs resolution (seed 0 best unit):
arm G=30 100 300 1000 3000
A shared 127 488 1430 3381 5989
B private 180 776 2210 5085 8763
B2 matched 216 955 2597 5539 9187
C global 121 481 1395 3364 6137
TRUE Mess3 396 2399 8924 26394 54821
learn.c coll 5 9 14 20 36
SECONDARY — log-loss gap vs optimal (mean +- sd over 5 seeds):
A shared +0.0051 +- 0.0020 (mean T 0.0000)
B private +0.0049 +- 0.0019 (mean T 0.0000)
B2 matched +0.0069 +- 0.0023 (mean T 0.0000)
C global +0.0102 +- 0.0044 (mean T 0.0002)
SEPARATION B2 - A = +0.0018 +- 0.0042 nats (>0 means TARGETING beat amount)
NOTE: log-loss is SECONDARY by pre-registration — README records that it barely
distinguishes finite/infinite/fractal structure. Judge on the ladder above.
EXIT=0