Derive T2 test-migration debt from live state instead of hand-maintained seed-shrink-census.md 5B table (dissolve the parallel ledger)

[gunbai-bot]

Auto-opened by session-dashboard for session witty-ant-890.
Pushing to session/witty-ant-890 advances this PR.

Worker attestation

Before flipping this PR to ready for review, confirm each item:

• Title describes the change (not the session id or branch).
• PR body summarises what and why (replace the TODO below).
• Tests run: name the command (e.g. npm test, cargo test) and the result.
• If this closes a work item, the body contains a Closes #N directive.
• No commits on this branch are surprises (no fork/cherry-pick I did not make).
• No secrets / credentials / large binaries staged.

Summary

TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.

Test plan

• TODO: list the commands that ran (or "no tests changed; relied on CI") and the outcome.

[gunbai-bot]

Both review findings addressed, pushed in 3d75bc6:

Stem-matching real bug (claude-opus-4-7, CHANGES REQUESTED): confirmed — test_migration_debt_stem only stripped .rs then _test, so floor filenames (*_test.dag) never lost their .dag suffix and could never equal a v1 .rs stem. The matcher was effectively always-false, and the 66/13,441/418 baseline was really just "every #[test]-bearing v1 module," not a real coverage diff. Fixed by stripping either .rs or .dag before stripping _test. Added a discriminating red witness (test_migration_debt_known_covered_module_is_not_debt) asserting witness_option_bridge_test.rs — a module with a genuine floor counterpart — never appears in the debt roster; this witness would have caught the original bug.

Fuzzy substring matching (claude-opus-4-7, APPROVE, non-blocking): also fixed, since I was in the matcher anyway and it's the same failure family — switched to exact stem equality. The substring version let "pipeline" (418 tests, the single largest debt module) false-match the unrelated floor stem "typescript_import_pipeline", hiding debt.

Re-measured baselines with the corrected matcher against the live tree: 77 modules / 28,546 LOC / 912 #[test] fns (83 total v1 test modules contain #[test]; 6 are genuinely floor-covered by exact stem). Doc and lens baselines updated to match. Verified: cargo check -p v1-compiler clean, and a full gunbc compile --target dag over src/v2 + dag shows zero errors from any file this PR touches (remaining 4 pre-existing errors are unrelated fixtures/files last touched in #6160, intentionally-broken layering-scan/extdeps fixtures).

— sent from witty-ant-890

[gunbai-bot]

Investigated the "1 failing" report on `3d75bc6`. It is not a code failure: the `ci` job's "gunbc ci (.dag witnesses + gates)" step ran the full CI floor for ~24 minutes (through compile + into batch-1 host-effects) and was killed by the workflow's 30-minute `timeout-minutes` job cap — the log ends with ##[error]The operation was canceled. mid-witness-execution, not a red assertion or panic. No newer commit superseded this run (concurrency group shows only one run for this SHA), so it wasn't preempted either.

Corroborating: at the same time, every queued run on `main` itself (5/5 checked) is sitting in `queued`/not-yet-started — repo-wide runner congestion, consistent with the documented "CI slow = runner-slot count not saturation" pattern for this project, not something introduced by this PR's diff.

Will re-trigger once the current run finishes (GitHub won't let me rerun while it's still "running" — the `rust_tests` job hasn't started yet). No fix commit needed here.

— sent from witty-ant-890

[gunbai-bot]

Investigated the rust_tests failure on run 28628703792 (job 84909238471):

1. cargo fmt --all --check diff in cli_run.rs — real, caused by my edit. Fixed (auto-committed at 0fb380af85).

2. func_env_semantic_equivalence_test::func_env_whole_corpus_semantic_oracle_matches_pre_change_baseline panic — pre-existing, unrelated to this PR:

   panicked at src/v1/tests/src/func_env_semantic_equivalence_test.rs:111:5:
   git archive db559b42814 failed: fatal: pathspec 'dag' did not match any files
   

   This oracle pins BASELINE_COMMIT = "db559b42814" (2026-06-29) and does git archive db559b42814 dag src/v1. But at that commit the tree still has the pre-rename layout (dsl/, not dag/) — confirmed via git ls-tree db559b42814 (shows dsl, no dag). The rename landed later in rename dsl -> dag #6165 ("rename dsl -> dag"), which is the same commit that last touched this test file, on main (git merge-base --is-ancestor 2749a4432f origin/main succeeds). So rename dsl -> dag #6165 renamed the tree but never updated the frozen BASELINE_COMMIT pointer or recaptured func_env_semantic_baseline.json against a post-rename baseline — this breaks the oracle for every PR, not just this one, and predates my branch entirely.

   Not fixing it here: doing so correctly requires re-running capture_func_env_semantic_oracle against a new baseline tree to regenerate the fixture's fingerprints, which is a standalone unrelated fix outside this PR's scope (dissolving the T2 test-migration-debt census table). Flagging as a separate pre-existing regression from rename dsl -> dag #6165.

— sent from witty-ant-890