deployah -> doyle: STEP 7 TEARDOWN, PRESERVE-FIRST MESSAGE. Nothing reaped. This is the gated proposal; the reap is a SEPARATE message after your confirmation.

OWNER: me, deployah. Cascade section 7 names the shape rig as mine and I am not delegating it.

THE OBJECT
  path      C:\Users\decid\Documents\projects\spt-core\.worktrees\shape-0680
  what      git worktree, DETACHED at 25e6001585ed0cd495a72c7634dded9dd207bdae — the dead r2 sha
  pool      its own target\, lane label shape-0680-r2, POOL-OWNER.json owner_tree = the worktree
            itself (claimed from its own tree, so the identity it wrote is the lane's)
  size      Length-sum 12,447,389,555 B = 11.59 GiB over 13,760 files
            of which target\ is 12,422,722,724 B = 11.57 GiB over 12,852 files (99.8%)
  ⚠ THAT FIGURE IS AN OVERSTATEMENT, stated as such: a Length-sum counts every hardlink entry, and
    a cargo pool is full of them. It RANKS the object; it does not predict the reclaim. The reclaim
    is a before/after free-space measurement and I will report it as one, not as this number.

CLASSIFICATION, in the order the rule requires
  OUTBOUND   Get-Item -Force: attributes = Directory, ReparsePoint = FALSE, Target empty.
             A REAL directory, not a junction — so it is deleted as a tree, not unlinked as a link.
  INBOUND    Reparse-point sweep, depth 3 under the project root: NONE aimed at it. Checked BEFORE
             proposing removal of a real directory, which is the order that matters.
  LOCKED     Not locked. (wan289 is the locked worktree in this repo; shape-0680 is not.)

WHAT IS AT RISK: NOTHING. Measured three ways, not assumed:
  1. Untracked files outside target\ : ZERO.
  2. Tracked dirty: ONE file, docs-site/src/cli/reference.md, flagged M. It is NOT modified.
     git diff is EMPTY (0 bytes). Byte comparison against HEAD: working copy 184,288 bytes, blob
     184,288 bytes, CR=0 and LF=4537 on BOTH, IDENTICAL RAW. The M is a stale stat-cache entry;
     git's 'LF will be replaced by CRLF' line is autocrlf's PREDICTION about a future checkout,
     not a description of the current file. I preserved a copy anyway before concluding that
     (.spt/preserved/golden-272-r4-drive/shape-0680-preserve/, sha256
     8525a194e570777b22f6c3e593f54726337a21952a1f934e5558e18a5fec9aa6) — cheap, and it means the
     claim does not rest only on my reading.
  3. Commits: 25e60015 IS AN ANCESTOR of origin/main (a2f335f8). Every commit in that worktree is
     reachable from main. Nothing unique lives there.

LANE STATE — the pool rules' own test, which is git identity, not the holder process:
  lane_base recorded = e44444136dc4eacf07516531ee3b8604f933498c, and that base is now an ANCESTOR
  of origin/main. The lane's work has LANDED, so this is a finished lane, not a live one. The
  recorded holder_pid 56464 is ADVISORY by the same rule and I am not resting the case on it.

RESTORE COST, stated so you can price the decision
  git worktree add --detach .worktrees/shape-0680 25e6001585ed0cd495a72c7634dded9dd207bdae
  ~seconds for the checkout. The POOL is the real cost: a cold full rebuild of that lane, which on
  this box is the 6-minute fixture build plus a full debug compile — call it the 15-25 minute class,
  and it is exactly the cost you told me not to spend before r4. r4 is now shipped, so that
  objection is discharged.

PROPOSED SEQUENCE — and one rule tension I want you to rule on rather than resolve myself
  AGENTS says: reap the TARGET SUBTREE, never a session or worktree directory around it. Section 7
  says: tear down the shape rig worktree. Those pull in different directions here. My proposal
  satisfies both and gets 99.8% of the space in the first step:
    1. pool-release the lane, using the PREBUILT xtask at ./target/debug/xtask.exe rather than a
       cargo run — pool-release REBUILDS xtask into the pool it just reaped (2.8 GB came back in
       hertz-lane4 that way, silently). Test-Path the target afterwards to prove it stayed empty.
    2. Reap the TARGET SUBTREE only: .worktrees\shape-0680\target — 11.57 of the 11.59 GiB.
    3. THEN git worktree remove the now ~25 MB worktree — the sanctioned verb for a worktree, not a
       blind rm of a directory. If it refuses (a handle pins it, e.g. rust-analyzer), I stop and
       report rather than falling through to rm -rf: a refusal HALTS, and I read git status before
       anything dies.
  Free space before, on the record: 127,939,584,000 B = 119.15 GiB. I will measure it again after
  and report the DELTA as the reclaim.

BUILDER CENSUS AT THIS MESSAGE, per your instruction: total 0, USER-rooted 0, three-state
classifier so an unresolvable read cannot pass as clean; both runners busy=False. The peers' lanes
are NOT yet up. If they come up before you confirm, I re-census and hold rather than racing them.

NOT DOING WITHOUT A SEPARATE CONFIRMATION: any of steps 1-3, and the runner reap, which stays
second in the order you set.