Repository navigation
Release rehearsal: crux producer and clean-room measure different commits; nothing makes them share one C #5074
Description
Follow-up to #5062.
The release rehearsal's pick needs both the crux bundle and the clean-room run on the same commit C. The models-nightly producer picks C as main's head at fetch time, or the commit named in its override-commit file. The clean-room run measures whatever head it saw when it started. Nothing schedules the two to agree, so on most nights they measure different commits and pick stays not_measured.
Today the gap is bridged by hand: an operator or agent writes a commit the clean-room already measured into override-commit before the crux run.
Acceptance:
- One side reads the other's C. Either the crux run takes the newest commit the clean-room measured green (when it is an ancestor of main), or the clean-room runs on the crux's recorded C.
- A night where the two still differ is reported as a named not_measured reason in the rehearsal receipt, not as a silent miss.
- A test row fails when the two Cs differ and the rehearsal does not say so.
Refs #5062
Activity
noahgift commentedon Oct 12, 2026
Evidence, branch
fix/5074-pick-names-split-c@ 0230bb1 (no PR; branch-only).pick_cnow adds this line:not_measured: the two sides measured different commits (#5074): models-crux newest <sha> (main~a), cleanroom-cpu newest <sha> (main~b). A side that measured nothing in the walk is printed asnone in the walk.a_split_c_is_namedanda_side_with_nothing_in_the_walk_is_named_none.rehearse.sh --selftestpasses 271/271. A hand mutation that drops the clean-room side's name turnsa_split_c_is_namedred.merge fix/5052 into the #5074 branch: one branch for the rehearsal pick