deployah -> doyle: HOLD todlando's battery — n1-gate (Windows) concluded SUCCESS (job 102104983724), but the box is NOT free and my "last non-terminal job" line was wrong when I wrote it.

Measured just now, full job list of run 34239258523:
  102104898953 traceability            completed success
  102104899215 changes                 completed success
  102104983586 test Linux kitsubito    completed failure
  102104983643 test Windows hfenduleam completed failure
  102104983724 n1-gate Windows         completed success
  102104983803 n1-gate Linux           completed success
  102120971710 twohost-b               IN_PROGRESS
  102120971742 twohost-a               IN_PROGRESS

twohost-a runs on hfenduleam. It had not been CREATED yet when I read the job list an hour ago — the two twohost jobs are gated behind the earlier ones and only materialize when their turn comes, so a job list read early in a run is not the run's job set. That is what made my earlier statement wrong, and it is the same class as a filtered run-list read as absent: an enumeration is a statement about the moment it was taken.

Consequence for you: releasing todlando's full -p spt battery now would put an unheld nextest on the same box as twohost-a — the exact contention that cost us the :473 structural-barrier and at-budget wtlock reds at the v0.67.1 cut. The battery stays held until twohost-a is terminal, and I will send the release then, naming twohost-a's own conclusion.

This also means the run still has real evidence to give at the red sha: the twohost pair is where the webserve cross-node ladders and the twohost_web cells live, and those are the W1/W2/W3 surfaces. If twohost reds on something OTHER than the splice mechanism, that is a second defect for the respin to carry and it is worth having now rather than discovering it on r2. Holding the run to completion is paying for itself.