feat(interop): classify third-party gap rows; ratchet SPAR-DEFECT count exactly (#427)

[avrabe]

Closes #427 (a)+(b) — the two "cheap suggestions", plus the separate-count assertion. Does not fix the three underlying parser defects, and does not decide whether the conflict-marker files stay in the corpus (both are separate maintainer moves the issue lays out).

The defect #427 measured

test-data/interop/baseline/third-party-gaps.txt ratchets 14 vendored models spar fails to parse. Read as a number that looks like 14 units of debt. #427 opened all 14 against the source and measured that it is 3:

class	n	is spar wrong?
UPSTREAM-INVALID — committed git-conflict markers	7	no
AADLV1 — unnamed connections (OSATE rejects them too, #421)	2	no
MALFORMED-V2 — missing properties / implementation keyword	2	no
SPAR-DEFECT — EMV2-qualified owners (2) + two-token reference(virtual processor) (1)	3	yes

Bundling three genuine defects with eleven files no parser accepts is how "14" invites being driven to zero by relaxing the parser onto invalid input, or absorbing a real regression into the mixed pile.

What this PR changes

• third-party-gaps.txt — a CLASS token on every row (path\tCLASS\tfirst-diagnostic), header rewritten to define the four classes.
• crates/spar-cli/tests/third_party_corpora.rs — two tests:
  • every_baseline_entry_is_classified — reds any row missing a recognised class token, so a new gap must be classified before it can be ratcheted in (the moment someone actually looks).
  • spar_defect_count_matches_ratchet — asserts SPAR-DEFECT == MAX_SPAR_DEFECT (3) two-sided, same shape as MAX_TOO_PERMISSIVE in three_way_conformance.rs: above the floor reds a new too-strict rejection; below reds until the constant is walked back. The other three classes stay in the file as documentation but are not counted as debt.
• artifacts/{requirements,verification}.yaml — REQ-PLUGFEST-CORPUS-CLASSIFY-001 (traces-to REQ-PLUGFEST-007) + TEST-PLUGFEST-CORPUS-CLASSIFY with executable fields.steps.

Non-vacuity (executed, not asserted in prose)

Three baseline mutations, each reddening a distinct assertion, run before land:

1. reclassify one SPAR-DEFECT row → UPSTREAM-INVALID: count drops → below-floor branch reds;
2. write an unrecognised class token: classifier reds;
3. append a fabricated SPAR-DEFECT row: count rises → above-floor branch reds.

Restoring the file greens all four tests. MAX_SPAR_DEFECT is a fixed literal, so neither bound is vacuous.

Verification

• cargo test -p spar --test third_party_corpora → 4 passed
• cargo fmt --all --check clean; cargo clippy --workspace --all-targets -- -D warnings clean
• rivet validate → 0 broken cross-refs, new REQ visible (its lone missing: design-decision warning is the repo-wide baseline shape)
• Independent clean-room pass re-derived every row's class from file content, re-ran all three non-vacuity mutations, and confirmed no over-claim in the artifact prose.

🤖 Generated with Claude Code

---

Generated by Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[github-actions]

Rivet verification gate

✅ 20/20 passed

	count
Passed	20
Failed	0
Skipped (no steps)	0

Filter: (and (= type "feature") (or (has-tag "v093") (has-tag "v0100")))

Failed artifacts

(none)

_{Updated automatically by tools/post_verification_comment.py. Source of truth: artifacts/verification.yaml.}