hertz -> doyle: WORKTREE CENSUS (read-only; nothing reaped or written except records). Records: .spt/preserved/331/hertz/worktree-census-20260924/. Classified against main 07d04ba5.

TOTAL ALLOCATED IN MY POOLS: 21,082,091,520 B (21.1 GB, du -sB1), all on KITSUBITO; apparent 21,011,576,703. hfenduleam: 0 B (the only hertz target/ there, hertz-S4-foreign-apply, is an empty dir). kitsubito / has 340.7 GB free of 1006 GB, so this is 6% of free space. My read: tidy-up, not a golden-floor IR, but that's your ruling. Escaped links NOT measured; that belongs in the reap capture.

CORRECTION #2 to my own record: my brief said "I hold no pools". FALSE. 3 kitsubito pools are mine, and 2 carry hertz-labelled CLAIMS (POOL-OWNER.json): hertz-lane1 (holder pid 650345, dead) and hertz-w2-proof (pid 579348, dead). Neither has a lane_branch; both trees are detached. hertz-ir136-fixture has an owner stamp only. Your standing note on me ("ZERO unlanded pool claims, owner-stamp-only") is therefore also false for hertz-w2-proof: its lane_base 2c9a06c3 is UNLANDED (below).

KITSUBITO (repo /home/reavus/projects/spt-core/spt-core, plus one in /home/reavus/spt-core-gate-w2):
- hertz-ir136-fixture: detached 52d56248, ancestor of main = LANDED. Pool 6,729,842,688 alloc.
- hertz-lane1: detached 17443615, 1 ahead, but its patch is in main (git cherry '-') = LANDED BY PATCH-ID. Pool 5,789,241,344.
- hertz-w2-proof: detached 2c9a06c3 'test: instrument W2 rig failures', 1 ahead, cherry '+' = UNLANDED. Pool 8,563,007,488.
- hertz-294-sync-stages (gate-w2 repo): diag/294-sync-stages ed8ecd94, 2 unlanded diag commits, remote branch EXISTS. No pool. Same branch also has a tree on hfenduleam.
Method note: kitsubito's clone is VOID for classification. Its origin is the runner work dir (main ref frozen at 68e34ae5, 09-06) and its object store is incomplete ("Could not read 629e33e1"; cherry printed a FALSE 0 beside a fatal). The gate-w2 clone can't fetch over https. I classified all four heads through the GitHub compare API plus git cherry on the full hfenduleam clone.

HFENDULEAM (36 trees, all in the spt-core .worktrees, no pools):
- LANDED, ancestor of main (25): arm1, f19-rig, integ(det), lane4, percell-id, premise, repin, 304, 307-ack-jitter, 307-sync-wall-bound, 307-twohost-barrier, 307-wall-budgets, 309-er-briefing, 318-ci-reuse, ir11-built, ir11-headless-column, ir135-checker, ir136-fixture, ir137-launch-discipline, ir86-disk-delta, ir86-linux-floor, ir86-retire, ir86-windows-floor, relay-069(det), S4-foreign-apply(det).
- LANDED BY PATCH-ID (not ancestors, cherry 0) (7): attach-ttl, census-iter, converge(det), liveresolve-diag, psyche-bound, teardown-bound, 307-evidence.
- UNLANDED (4): twohost-gate (det b6bbaf22, 1 commit), 294-poll-margins-base (c4109a58, 1, no remote), 294-poll-margins (161eb554, 1, no remote), 294-sync-stages (ed8ecd94, 2, remote exists).
Counts from the TSV: 25 ancestor + 7 patch-id + 4 unlanded = 36.
Anomaly, flagged but not diagnosed: 22 of the 36 appear in `git worktree list` as <dir>\.git, i.e. the admin record points at the .git FILE, not the dir. Every one resolves to an existing dir. It may matter for how `git worktree remove` behaves at reap time.

Nothing reaped. Waiting for your order.