--- name: the-name-you-give-a-tree-selects-which-discipline-fires description: "The same bytes have several true descriptions, and the one the task hands you decides which rule you apply — I rm -rf'd an 'orphan worktree directory' that was also a 9.3 GiB cargo pool with a binding reap method attached" metadata: node_type: memory type: feedback originSessionId: 9a42c7a1-71f1-47b6-8341-adf7b3dbae7d modified: 2026-09-09T16:55:32.936Z --- **Measured 2026-09-09, mine, reported to doyle before he could find it.** A worktree directory had refused `git worktree remove` on a pinned handle. Later the handle cleared and I removed the orphan with `rm -rf`. That directory **contained the `hertz-65s-rider` cargo pool — ~9.3 GiB**, which doyle had queued as a reap with a binding four-arm method (`Get-Item -Force` outbound classification, inbound reparse sweep, `CARGO_TARGET_DIR` check, size and free space before AND after, subtree only). **None of that fired.** Not because I disagreed with it, and not because anyone rushed me — because I never saw the second description. The task reached me as *"the orphan worktree directory you could not remove earlier"*, so it selected the worktree-cleanup rule and completed cleanly by that rule's standard. The pool inside it had no name in that frame. Two rule lines were crossed by one command (doyle's audit): **"never delete a `target/` you have not classified"** and **"reap the target SUBTREE, never the worktree directory around it."** **Outcome was clean, and that is not the point.** After the fact I could still show: no inbound reparse point aimed at the path, `CARGO_TARGET_DIR` unset, contents rebuildable by construction, the 30 untracked source files hash-verified preserved *before* removal, branch already landed. Free space 239.78 GiB vs the ~231 quoted earlier. But the before-size is gone forever, so the reclaim is a free-space delta and **not a measured reap** — recorded that way deliberately, and no register line claims the figure. **Why this is worth a file:** it is the same shape as the census failure I had caught in a peer the same hour ([[a-tracked-only-corpus-cannot-answer-tracked-or-not]] — a file's PATH supplying an answer its corpus could not), and as the directory-ignore trap ([[ignoring-a-directory-buries-what-the-repo-cites-in-it]]). In all three, a *description* stands in for a *property*. Catching it in someone else's work is much easier than catching it in the sentence that arrived as your own next task. **How to apply:** 1. **Before any destructive act, ask what the target CONTAINS, not only what it IS called.** A `Get-ChildItem -Force` one level down would have printed `target` and renamed the whole action. 2. **When a rule exists for a thing, the rule fires on the THING, not on the task title.** If the tree holds a `target/`, the pool method applies even when the ticket says "worktree cleanup". 3. **A refusal that later clears is still the same object.** The deferred retry inherits every discipline the first attempt owed; nothing is discharged by waiting. 4. **Report the process failure even when the outcome is clean** — and refuse to dress the result up as a measurement you did not take. "~9 GiB by free-space delta, not a subtree before/after" is the honest record; "9.3 GiB reclaimed" would have been a fabricated measurement. Kin: [[reap-subtree-not-session-dir]], [[preservation-is-its-own-step-before-any-reap-broadcast]], [[a-length-sum-census-overstates-a-cargo-pool]], [[a-tracked-only-corpus-cannot-answer-tracked-or-not]]. ⚠ **DUPLICATE, and it is a finding in itself.** doyle wrote [[a-tree-gets-its-discipline-from-the-name-the-task-gives-it]] on the SAME incident within minutes of this file, neither of us knowing the other was writing. Same lesson, two authors, one shared memory dir. His is the DRI's account and correctly attributes the failure to me; this one adds two things his does not — that a deferred retry inherits every discipline the first attempt owed, and that this is the third instance of a DESCRIPTION standing in for a PROPERTY in one day. **Read either; do not assume one supersedes the other, and check for a peer's file on a shared lesson BEFORE writing your own** — the same 'check what already exists' rule this dir states for duplicates, which I broke while writing about not trusting a single frame. doyle to rule which is canonical.