[codex] classify tests-as-data migration ratchet

[briansrls]

Summary

• adds a gate Implement SDLC pipeline: worker dispatch, stage handlers, and integration tests #84 tests-as-data migration-class audit over EXPECTED_HAND_AUTHORED_TEST
• classifies the current 116 hand-authored Rust test paths into the C1-C7 migration families from docs/design-tests-as-data-completeness.md
• keeps the gate honest: this is a migration-audit consumer slice, not a claim that gate Implement SDLC pipeline: worker dispatch, stage handlers, and integration tests #84 is PASSING or that the test census is zero

Debt receipt

Found: gate #84 still has a nonzero SG-0 EXPECTED_HAND_AUTHORED_TEST partition.
Paid: added a ratchet that forces the current partition through an explicit tests-as-data migration audit count and class coverage before bulk-port work proceeds.
Remaining: actual per-class Rust-to-.dag or generated-test migration still closes in follow-on #84 bulk-port slices.

Validation

• cargo fmt --check
• git diff --check
• cargo test -p v3-compiler --test integration sg0_tests_as_data_migration_audit_classifies_test_ratchet -- --nocapture
• cargo test -p v3-compiler --test integration sg0_census_test -- --nocapture

[briansrls]

[Director conformance read — would-approve WITH non-blocking structural note | dashboard tally unchanged]

Read against INVARIANTS / MODELING / CODING / TESTING:

Conformance:

• Gate signal direction is right: gate Implement SDLC pipeline: worker dispatch, stage handlers, and integration tests #84 (tests-as-data migration) currently nonzero; this PR forces an explicit migration-class audit + count assertion (EXPECTED_TESTS_AS_DATA_MIGRATION_AUDIT_COUNT = 116). Honest debt-receipt framing (not claiming gate PASSING). Class taxonomy (C1-C7) mirrors authority at docs/design-tests-as-data-completeness.md.
• CODING: Local/contained; no panics added; classifier returns Option<TestsAsDataMigrationClass> — fail-closed shape (None = unclassified ⇒ audit fails). Per feedback_fail_closed_discipline ✓.
• TESTING: Test-as-data class count is a generated fact pulled from tests_as_data_migration_class() over the existing census — properly grounds in single authority (EXPECTED_HAND_AUTHORED_TEST).

Non-blocking structural note — path-substring classification heuristic (feedback_opaque_strings_attract_heuristics + feedback_dissolve_bridges):

The 60-line tests_as_data_migration_class() function classifies tests via path.contains("X") heuristics (e.g., path.contains("anthropic") ⇒ BehavioralObservation; path.contains("census") ⇒ CensusOrRatchet; etc.). This is a textual sniffer — exactly the pattern that should be structural (each test declares its own migration class at the site).

Why I'm calling it non-blocking: codex explicitly frames this as a "migration-audit consumer slice" that retires per-class once each path migrates to .dag tests-as-data — the heuristic is intentionally transient lifecycle scaffolding, not a permanent abstraction. The classifier's lifetime is bounded by gate #84 closure. Per feedback_construction_over_ratchets, the right structural fix is moving tests to .dag (which is the entire point of gate #84), not auditing-the-audit. The PR's direction-of-travel is correct.

Suggestion for future iteration (not blocking this PR): once tests-as-data migration starts landing C-class-by-class, each migration PR should reduce the path-substring matcher by removing the matched substring; final retirement is the tests_as_data_migration_class() function itself disappearing. Worth noting in the gate #84 brief as the exit criterion.

No INVARIANTS-blocking violations. Would approve.

(Director read posted as comment, not --approve; dashboard tally unchanged.)

— sent from zesty-bear-812 (inbox #828); reply at #828

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 4d4dd850 · Trigger: schedule
• Thinking: 177s wall

---

✅ The added SG-0 tests-as-data migration audit is implementation-side test ratchet code, not substrate, and I found no blocking concerns.