Wikimedia at K = 20: detectable effects below the SESOI, the null within nominal at the sizing bound¶
# research(2026-10). partition-sensitivity-wikimedia-rp1.0-logprob_per_token-k20.json (TIGRIS
256458, pinned at 642f6d9, RUNS=20), the simulation the K coverage rule queued. Pool, noise
and shrinkage as at K = 10; 1,000 trials and 4,000 null trials a point, 2,848 planned changes.
Detectable effect at one to three cells, in nats per token:
| spread point | sigma_run | detectable, 97.5% | detectable, 95% | null false positive, 97.5% | null false positive, 95% |
|---|---|---|---|---|---|
| pilot lower 90 | 0.0000 | 0.0037 to 0.0041 | 0.0034 to 0.0038 | 0.0035 | 0.00725 |
| pilot estimate | 0.0037 | 0.0050 to 0.0054 | 0.0046 to 0.0050 | 0.00575 | 0.01375 |
| sizing bound 90 | 0.0062 | 0.0067 to 0.0073 | 0.0062 to 0.0067 | 0.0095 | 0.022 |
| pilot upper 99 | 0.0085 | 0.0083 to 0.0091 | 0.0077 to 0.0084 | 0.01325 | 0.02475 |
Every detectable effect is below the SESOI (0.0183), and below its K = 10 value at every spread point. At the 90% sizing bound the null reads false positives within nominal at both levels, as the coverage rule required. At the pilot's 99% upper spread the 97.5% rate, 0.01325, exceeds the nominal 0.0125 by 0.00075, inside one Monte Carlo standard error at 4,000 trials (0.0018). Reproducibility failure is 0 at every point. The simulated null reads bounded in at least 0.9752 of trials at the registered two cells, and 0.94875 at worst (one cell, 97.5%, 90% lower spread).
The registration now quotes the registered metric. registered-decisions.md ("Interval the
gate reads", "Number of runs", "Stated power"), the interval paragraph of outcome-neutral.md, and
the estimand and development-window reading of protocol.md are rewritten on likelihood, from the
likelihood pilots, these simulations (OpenStack at K = 10, Wikimedia at K = 20) and the
k-coverage-<org>-likelihood.json K sources; harvest.py asserts every figure against them. A
claim's new after option counts a small integer such as a K only where its prefix precedes it,
so "K = 10" is not satisfied by every other 10 on the page.