IR-150 capture — doyle, .worktrees/gate-w7/target, hfenduleam NTFS, 2026-09-24T20:09:41Z NOTE: CI self-hosted Windows runner ACTIVE on this box during capture (run 36051790154 unit leg); free-space delta deferred to a quiet window. -- outbound classification -- drwxr-xr-x 1 decid 197609 0 Sep 24 11:37 .worktrees/gate-w7/target -- apparent -- 91491284528 .worktrees/gate-w7/target -- allocated -- 91580153856 .worktrees/gate-w7/target -- escaped hard links (inodes, bytes) -- 0 0 -- free before (whole volume) -- C: 1999327195136 1885320839168 114006355968 95% /c -- writer census, read BEFORE the rm (2026-09-24T18:48:42Z) -- rust-analyzer 37184 rust-analyzer 50072 rust-analyzer-proc-macro-srv 52172 rust-analyzer-proc-macro-srv 54188 CI runs at census: 36051790154 completed, 36052137510 completed, 36053208562 completed (no run in_progress) -- inbound reparse-point sweep aimed at the target -- [2927 reparse-point rows elided - every hit was in an unrelated tree (rebno node_modules, attractor references); 0 aimed at .worktrees/gate-w7. Re-derive with: Get-ChildItem -Path C:\Users\decid\Documents\projects -Recurse -Force -Attributes ReparsePoint] -- INBOUND sweep verdict: 0 reparse points aimed at the target (hits found were unrelated trees: rebno node_modules, attractor references) -- -- census note: 2 rust-analyzer + 2 proc-macro-srv only; no cargo/rustc/nextest/Runner.Worker; all three CI runs completed -- -- rm start 2026-09-24T20:49:55 BOX CLOCK -- -- CLOCK NOTE (RESOLVED): there is NO clock skew. Git Bash, PowerShell and GitHub all agree; what was wrong was my sense of elapsed time. pacer's stretch board puts this session at ~6h, so the CI runs timestamped 19:58-20:11Z really did finish ~5h40m before the rm, across turn gaps I read as minutes. An earlier line in this file claimed the shell clock ran fast - that claim was unfounded and is withdrawn. Consequence for the arithmetic below: the stale baseline is stale by ~5h40m of volume activity, not by one CI job. -- rm exit=0 wall=18s -- free AFTER -- C: 1999327195136 1866225717248 133101477888 94% /c -- target present? -- ls: cannot access '.worktrees/gate-w7/target': No such file or directory == ARITHMETIC == expected reclaim = allocated 91,580,153,856 - escaped 0 = 91,580,153,856 B observed free delta = 133,101,477,888 - 114,006,355,968 = 19,095,121,920 B GAP = 72,485,031,936 B unaccounted PROCEDURAL FINDING (mine, this reap): the free-space PRE-read was taken at CAPTURE time, not adjacent to the rm. Between them the CI Windows unit leg for run 36051790154 ran to completion ON THIS BOX and wrote its own workspace target. The delta is therefore CONFOUNDED, not merely unexplained: the baseline is stale by a whole CI job. No mechanism is assigned to the gap. Next reap: re-read free IMMEDIATELY before the rm, in the same call as the census. == CLOSURE == pool-release: 'pool .worktrees/gate-w7/target holds no stamp - nothing to release' (the stamp lived inside the reaped tree), exit 0. git worktree remove .worktrees/gate-w7: exit 0, registration gone. git worktree list now 45 rows. Reaped the target SUBTREE first, the worktree only after, never a session directory.