A directory that was emptied for good is not a directory that might refill

[melbinjp]

paths.py skips a trailing-slash directory claim when the parent is tracked, because git stores files and not directories, so an empty directory is indistinguishable from an absent one. The reason is real - PostHog/posthog-python's RELEASING.md says changesets live in .sampo/changesets/, and every release empties it as the bot consumes them. The sentence says where sampo add PUTS files.

The guard's own comment then claims the tracked-parent requirement "keeps the case this must still catch: a documented directory whose entire tree really did go".

It keeps none of them. A removed subdirectory of a live tree has a tracked parent too.

stacklok/toolhive   docs/arch/06-registry-system.md:864   `pkg/container/verifier/`
                    7095e8e1  "Remove /verifier in favour of one coming from toolhive-core"

pkg/container is alive and documented. verifier is gone. The tool said nothing - two lines under a finding on the same paragraph that it did report.

Measured

Every --show-skips run over the 47 clones of sweep batches 10 and 11, captured to disk rather than sampled:

directory claims silenced by the guard	38
never tracked at all - the guard is right	31
populated and wholly emptied	7
emptied and refilled in this corpus	0

The 7 are in four repositories: toolhive x2, hive x2, onyx x2, openmed x1. The churning shape the guard exists for does not appear in these 47, though it is real - composio/.changeset has it, and the test suite builds it.

The fix

The skip keeps its reason and gains a receipt, which is the soundness rule the rest of this verifier already runs on: a path is called broken only when the repository can be shown to have HAD it and dropped it.

Git.emptied_and_stayed replays the paths oldest first and counts how many times the live set falls to empty. Once means it stayed empty. More than once is a lifecycle and stays skipped.

The replay direction is not incidental

The first version of the measurement walked newest-first, set a flag on the first D, and called it a refill at the next A. An A seen later in a newest-first walk is older, so every directory that was created before it was deleted scored as churn - which is all of them. That run reported:

38 directory claim(s) silenced by the tracked-parent guard
  never tracked        31
  churns                7

Zero emptied. Nothing about that output looks wrong, and it is the exact opposite of the truth. It was caught only because one case had been verified by hand first, which is the argument for checking a new measurement against a known answer before believing its totals.

Tests

The posthog case moved to tests/test_emptied_directory.py and now builds a real fill-consume-fill-consume history. The old fixture committed one file and deleted it once, which is not what a changeset directory does - so it asserted the posthog verdict over a history posthog does not have. Also checked against the live repository: emptied_and_stayed returns None for .sampo/changesets there.

215 tests. Verified end to end on the real toolhive clone: 13 findings to 15, both new ones naming the commit that removed the subsystem.