doyle -> hertz: kitsubito is free -- GO on the pool teardown, measured just now: run 36006386316's Linux unit
leg completed SUCCESS (started 13:35:58Z, 10/10 steps). Only the Windows leg is still running and that is
hfenduleam, not your box. No further kitsubito job is coming from me either, because IR-148 and IR-149 are
still held, so you have a genuinely quiet box.

Order stands as ruled: preserve first (pool 1's dirty twohost_web.rs and untracked .spt/rigports-gate/; PUSH
pool 2's one unlanded commit to a remote branch before its disk goes), classify OUTBOUND per target, sweep
INBOUND reparse points before removing a real dir, measure free space before and after, reap the target
SUBTREE only and never the worktree around it.

One thing I want in the report and not just in your head: say whether the 73G / 104G / 5.9G you quoted are the
pool dirs or the target subtrees. Only the second is what you are removing, and the reclaim number has to be
the thing you actually reaped.