The stopping rule is liberal at the K it stops at, so K is fixed from the pilot: 24 for OpenStack¶

The simulation of the registered rule (partition-sensitivity-openstack-stopping.json, commit 2cd4184): 1,000 null studies and 300 per power step at each run spread, the test window's 1,635 planned changes drawn from the v3 pilot, runs on admissible partitions built as the jobs build them, Algorithm 1 from K = 8 to 40. One-sided false positives against nominal 0.0125 (97.5%) and 0.025 (95%), mean runs, share at the cap, and the rate at which a second, independent aggregation differs by more than xi = 0.01:

The excess is largest where the rule stops earliest and gone where nearly every study reaches the cap, which is a fixed K: the rule stops when the runs happen to agree, and the interval narrows with them. Reproducibility fails above beta from spread 0.015 up. Ritzwoller and Romano's guarantee is asymptotic in the number of splits; they recommend a burn-in of at least 10 and a tolerance at which the rule runs more than 500 splits (their section 5). A run here is a GPU job, and the rule stops near 16.

Amended before any test-window run: K is fixed per organization from its development-window pilot by their sizing formula (eq. 5.3), K = 2 v (z / xi)^2, with v at the 90% upper confidence bound of the pilot's per-run standard deviation, clamped to [10, 40] (partitions.runs_needed). The reading takes the first K runs in admissible order; the criterion is reported beside it and decides nothing. OpenStack's 22 pilot runs, all computed before this rule was written (partition-pilot-openstack.json): standard deviation 0.0140, upper bound 0.0177, K = 24. Over the 22, H1 reads +0.0051 [-0.0089, +0.0192] at 97.5%, inconclusive. Runs 23 and 24 (partition seeds 28 and 30, training seeds 23 and 24) are jobs 208898 and 208899, pinned at 205ba2f as the first 22, so the pilot is read once at the registered K. The fixed-K simulation at K = 24 runs next; its coverage decides whether this rule is registered.