test(dl): pin the free_mnd_to_vec Suspend(None) panic message (#312)

[tsondru]

Closes #312.

What changed

• catgraph-dl/tests/free_monad_bijections.rs (+20/−1, git diff --stat 23e8a0a..HEAD): new test free_mnd_to_vec_panics_on_bare_suspend_none — #[should_panic(expected = "non-canonical Free value")] on a bare Free::<ListEndo<u32>, ()>::suspend(None) — plus its module-doc taxonomy item; the taxonomy header's stale hand-maintained count ("Six", against 7 items at base 23e8a0a and 8 after this branch's addition — round-1 reviewer's git show trace) dropped in favor of the per-item claim.
• catgraph-dl/tests/THEOREM_MAP.md (+9): a "Not a law test (dl: documented Suspend(None) panic contract has zero should_panic pins #312)" note excluding the pin from the free_monad_bijections::* row, mirroring the file's own recursion-depth precedent (an engineering pin carries no paper anchor).
• catgraph-dl/CHANGELOG.md (+3): one bullet under [Unreleased] / Changed referencing dl: documented Suspend(None) panic contract has zero should_panic pins #312, matching the four precedent audit-sweep test commits (git show --stat a43f6e1 646881f 68be9f4 24b5524 — each carries one).

Falsification

Impl round plus 2 executing review rounds, all perturbations in cp -r copies:

Perturbation (list_endo.rs::free_mnd_to_vec)	Compiles?	Named test	Observed
panic text "non-canonical" → "malformed"	yes	RED	panic message: "…malformed Free value." vs expected substring: "non-canonical Free value"
panic! → return (items, Default::default()), signature unchanged	no	n/a	E0277: the trait bound Z: Default is not satisfied — the issue's adjusted-scope premise ("a silent-return regression cannot compile today") measured, not assumed
signature widened to Z: Default, silent return	yes	RED	test did not panic as expected — the pin catches the future-signature-change path the issue names

In every red row the new test was the only red of the file's 8 (review round 1, full-file reruns: 7 passed / 1 failed).

Review record

2 rounds (rust-v2:falsifying-reviewer, sonnet), one per commit (git log --oneline 23e8a0a..74a5c58 = 2 commits): findings 1 → 0 — tallies read off each round's reviewer report. The one finding (stale "Six consolidated tests" header over an 8-item list, drifted since two commits before this branch) fixed by dropping the count, verified true in round 2 by an 8-item ↔ 8-#[test]-fn bijection check. Round 1 also flagged a methodology hazard for the standing rules: a target-dir shared between a pristine tree and a perturbed copy silently reused a stale test binary — the per-agent --target-dir rule should be read as covering sequential pristine-vs-copy comparisons inside one review, not only concurrent agents.

Gates (final tree 74a5c58)

• cargo test --workspace -- --nocapture — 147 test result: lines, all ok, none FAILED (log kept as the measured-claims input)
• cargo nextest run -p catgraph-dl — 103 passed / 0 failed at 6c414bb (the tip commit is one //! line; its tier gates — fmt, clippy — clean at 74a5c58)
• cargo clippy -p catgraph-dl --all-targets -- -D warnings clean on default, --no-default-features, --features ad, --features serde, --all-features (at 6c414bb; default lane re-run at 74a5c58)
• RUSTDOCFLAGS="-D warnings" cargo doc -p catgraph-dl --no-deps clean; cargo fmt --all --check clean
• scripts/check_measured_claims.py — "guard ok: 44 citation site(s) across 20 key(s) match 24 measured fact(s)"
• check_audit_counts.py not run: the diff touches none of the four audit docs it checks

What the row got wrong about the tree

Nothing: all six premises AGREE (premise-verify herder report /tmp/audit-reports/G9-T3/premise-verify.md — zero should_panic crate-wide, panic site/message/rustdoc at the cited lines, no coverage added by the interim G9-T2 landing).

🤖 Generated with Claude Code