The planted check passes 10 of 10 on Wikimedia; the planted run's base arm is left out of it¶

The ten runs planned above (jobs 211277 to 211286, pinned at a43f7af), rq1-partition-wikimedia-*-n1850-plant0.5.json. All ten pass, every half's lower bound above zero (least +0.036), per-half estimates from +0.087 to +0.143, projected pass probability at least 0.9994 per run (0.9999 on average). By the rule fixed before the runs, check 5 is registered for Wikimedia as for OpenStack, at a half.

Three planted runs fail non-degeneracy, on the base arm only (partitions 3, 8 and 11: the base model scores exactly 0 on one half). The plant rewrites half the references that arm is scored on, and the base model never emits the marker; Wikimedia's base rate is about 1% (unplanted runs on the same partitions: 0.8% to 2.2% on every half, every adapter arm above 23%), so on its 160 to 200 changes a half zero is reached by chance. No OpenStack planted run did this. Check 5 never read the planted run's own battery; it now requires the arms its contrast compares, the four adapter arms, to be non-degenerate (planted_convention, tested), and leaves the base arms out, with the reason stated in its docstring. Every OpenStack and Wikimedia planted run passes it. The unplanted runs that H1 reads keep the full battery.

The Wikimedia placebo (job 211287, rq1-placebo-wikimedia.json) completed; its simulation, job 211288, is queued behind it.