Builder routing: rebases go to a builder who can't push (0/5 ok), and the picker sends 23 of 23 picks to Scout (5 slots run 1 builder) #7006

Builder routing is wasting today's crew in two ways. Measured on 2026-10-27, 12:00–13:15Z.

1. Approved pulls that conflict get "brought up to date" by a builder who can't push to the branch. The log shows approved pull conflicts: brought up to date first … pull=6956 builder=torque for=scout, and every such run fails with "this pull's branch lives in a fork the agent cannot push to". #6956 and #6970 looped 7 times in 45 minutes (12:29–13:11), and Torque's forge record over 3 hours is 0 ok out of 5. The rebase has to go to the pull's author (the fork owner). If the author is busy, queue it behind the author's current run; don't hand it to someone else.
2. The picker sends nearly everything to one builder. In the last hour builder=scout won 23 of 23 picks, and Scout holds 56 open forge issues against Mason's 7 and Torque's 1. One agent runs one thing at a time, so 5 slots run about 1 builder (active=1 slots=5), and work gets skipped with "scout's run was not queued (ceiling or breaker)". When the preferred builder is busy, the next capable idle builder takes the issue. The decider's preference is a ranking, not an exclusive pick.

Done when:

• An approved conflicted pull's rebase runs as its author, and a test covers that.
• With Scout busy and Mason idle, a startable issue goes to Mason, and a test covers that.
• active tracks min(slots, startable, idle builders) in the tick line.

Live triage (automatic, not written by a person): no pillar named, so it waits behind the Later gate; name a pillar to move it.