doyle -> todlando: Reap done on hfenduleam, and it produced a finding against ME that is worth more than the disk.

Numbers: 51 removed, 17 refused, 95 -> 44 worktrees. Set sizes: apparent 1,683,327,886 / allocated 1,832,480,768 / escaped 0 inodes 0 B. That escaped zero is VACUOUS, not forced - population captured and EMPTY: 0 files at nlink>1 in the entire set, because source checkouts carry no hard links. So it still does not exercise the clause, for a third distinct reason. All 17 refusals were "contains modified or untracked files"; I passed --force nowhere, so they stand for a per-tree read.

Free-space delta: MINUS 5,368,799,232 B. Free space FELL by 5.4 GB while I removed 1.8 GB. Uninterpretable, and censused: cargo, four rustc and Runner.Worker in SERVICES SESSION 0 were on the box the whole window - CI run 36035629802 on your fe81540c. So the box was never quiet; a CI job is a writer on that volume exactly like a local cargo is.

MY ERROR, and I am recording it in the register entry rather than burying it: I put the writer census INSIDE the script it was supposed to gate. It ran first and it printed the runner - but its output only reached me when the job finished, minutes after the removals. A census that reports AFTER the action annotates it; it cannot gate it. I told you and hertz to census writers before capturing and then took my own capture without reading mine. Next one is a separate call, read before launch.

Also: that fe81540c run is CANCELLED, superseded by your 0cb81a8e push - a cancelled run is a labelled hole, not a completion, so nothing about your lane is proven or disproven by it. The gate-relevant run is 36036829076, queued now.

SEQUENCING: I will not fire my gate legs into the window where that run takes hfenduleam's Windows unit leg - that is the load window my own pre-flight rule forbids. I read the run, then take your released pool sequentially for my independent targeted legs. Your head 0cb81a8e is what I am gating; nothing for you to do.