Read-only reclaim assessment; NO deletion or build performed.

Only one target directory remains beneath the hertz-named worktrees: .worktrees/hertz-294-sync-stages/target. Prior hertz fixture scripts addressed their local ./target; those other local targets are absent from the current glob. No extra completed hertz pool found to discard wholesale. Root target and other authors' trees are not claimed as mine. In particular .worktrees/293-brain/target remains unused.

Owned pool: Get-Item -Force shows a real directory (LinkType null, Target empty), not an outbound junction. POOL-OWNER.json contains only owner_tree=hertz-294-sync-stages and written_by=spt-poolguard: released record, not a live claim. Read-only size traversal skipped reparse points and encountered none/errors none. Full inventory: owned-pool-size-inventory.json. Its initial allocated_path_sum field comes from GetCompressedFileSizeW and is NOT a distinct physical-reclaim guarantee; candidate sizes below were measured separately with FILE_STANDARD_INFO.AllocationSize and link counts.

CONDITIONAL candidate:209 test-executable PDB sidecars in debug/deps,44,005,740,544 allocated bytes =40.98354GiB, each with NumberOfLinks=1. Exact allowlist and per-file allocation sizes: reclaim-pdb-candidates.json. These are debug symbols, NOT executables, libraries or incremental objects. Removing some/all would sacrifice postmortem source-symbol debugging for those exact test binary builds; regenerate by rebuilding that test if later needed. This retention tradeoff requires Doyle's explicit approval: it is NOT an assertion that all symbols are evidence-free, and NOT an IR-93 product symbol-packaging change. Zero bytes currently deletion-approved.

Protect untouched:
- ALL .exe/.dll and ALL DLL PDBs; root binaries/PDBs and fixture executables.
- ALL rlib/rmeta, .fingerprint, debug/build, debug/incremental and metadata: preserve the warm dependency/cache premise of the4–8GiB estimate. No wholesale target deletion/cargo clean.
- All spt* unit/CLI PDBs, both sync PDBs, daemon_stop_convoy_e2e, contract_e2e, legacy_resident_sweep_e2e and false_promote PDBs. These cover every binary with FAIL/LEAK lines in retained ArmB B1–B3 logs, plus the core binaries.
- resume_no_control_steal_e2e PDB, the planned fixture analogue.
- All raw logs, stage reports, commands, source commits, traceability output, bundles and .spt/preserved evidence. No worktree/session-directory teardown.
- Multi-link/unclassified fixture PDBs listed in the JSON are excluded, not counted as candidates.

Before any approved deletion: Doyle chooses exact allowlist/subset and resolves symbol-retention consent; perform current active-consumer/custody check and an INBOUND reparse-point sweep (not yet performed), re-stat exact files/link identities, measure free space before/after, then report actual reclaimed bytes. No claim that the projected40.98GiB has already been reclaimed or that the41GiB start threshold is met.

Binding allocation now recorded: start>=41GiB free; stop at additional growth>=9GiB OR free<=32GiB. Cold-cache/toolchain changes invalidate it. Doyle's reported C:19,918,860,288 and E:17,297,858,560 bytes are accepted; not remeasured to confirm. Fresh-build authorization still withheld. Isolation remains separate.