Sweep dangling citations to retired Python tests

[blooop]

Question

~29 comments in rust/ cite test files deleted with the Python retirement (#267) — e.g. flows/lifecycle.rs:2836 → test/unit/test_purge_ownership.py, devlaunch-test-support/src/cache_tree.rs:26 → parity_cases.txt, and docs/rust-rewrite-plan.md:194 (row 30) likewise. The citation mechanism is the repo's stated decision-vs-accident record; pointing it at deleted files quietly inverts its value. Re-point each at the Rust test that now pins the behavior, or mark it historical where none does.

[blooop]

Claimed 2026-08-29: build in progress from an interactive session (opus agent in an isolated worktree, managed as part of closing out map #299). PR to follow. If you are another session seeing this, do not double-build.

[blooop]

breadcrumb: re-ran the sweep against the current tree rather than trusting the inventory, because the tree has moved since this was written. The count is bigger than "~29": 60 citation sites naming 41 distinct Python test files that no longer exist, spread over 23 Rust files, one Cargo.toml, one workflow, three surviving Python test files and the divergence table. Everything named went in a single commit closing #267, so there is one retirement to point at rather than several.

Seam: the citation text itself, plus one new prose guard at the seam the repository already uses for this class of rule (a Python test that reads the tree and asserts a property of what it finds, the way the docs-prose and docs-links guards do). The rule the guard enforces: a token spelled like a Python test file path is a live pointer, so it has to resolve; a retired suite is named without the path spelling and with the retirement beside it. That makes the sweep verifiable instead of a one-off tidy, and it is the reason a second one will not be needed.

Two things the ticket implies that are not true of the tree as it stands: the citation blocks are almost all already past tense ("Ported from", "re-pinned"), so the defect is that a reader cannot tell the file is gone rather than that the sentence lies; and at least one citation is a live claim that is simply false, not merely dangling. The bench workflow tells a reader that a named Python spawn-count guard "keeps the gating role" and lives in ci.yml. It does not live anywhere.

[blooop]

breadcrumb: #511 is up. 66 citations swept: 4 repointed at the guard that actually replaced them, 62 marked as the history they are, and one claim deleted because what it described cannot be run any more.

The rule the sweep settles on is about spelling rather than tense, which is what makes it checkable: a token spelled the way a test file is spelled has to resolve to a file, and a suite that retired is named without the extension beside a sentence saying it retired. A new guard holds every such name to being a file, so this cannot silently recur the next time a tree is deleted. Two of the sixty-six were not merely stale but false, each asserting a live pin that had gone with the harness that held it.

One thing worth carrying forward: the same rot exists one layer over, in the "Ported from devlaunch/<module>.py" headers that name the retired implementation rather than its tests. Those are out of this ticket's scope and read as honest provenance, but they are the same shape, and if a future reader wants them covered the guard is two lines from reaching them.

[blooop]

wf-review: approve (no blocking findings), two non-blocking inline items left on #511 -- a live suffix_verb citation on the row-30 line and an overstated "needs no test" claim in aid/Cargo.toml. Guard proven non-vacuous (red on exactly 33 files at the guard-only commit) and free of false positives across the whole tree.