fix(#1087): parse the ledger duplicate-key-strict — a +622-line waiver was recorded with a +10-line justification

[avrabe]

Closes #1087.

claims.yaml on main carries a duplicate reason: key inside one waiver mapping of the selector_lines_code ratchet. YAML keeps the last value silently, so the surviving justification for RQ-60-VFPPRESSURE increment 1's +622 line growth of the instruction selector is RQ-59-I64SHIFT's reason for +10 lines.

16b135ba  RQ-59-SUBTRACT          18278
2785ffe9  RQ-59-I64SHIFT          18288   +10   <- the surviving reason
e6a3b27a  RQ-60-VFPPRESSURE inc1  18910  +622   <- the actual growth
0ec9dc7a  RQ-60-VFPPRESSURE inc2  19199  +289

The lane did its job — its commit message and the discarded reason both state the +622 honestly. The file edit turned to: 18288 into to: 18910 and appended a second reason: instead of adding a second waiver entry.

Why the gate could not see it

check_ratchet asks only "is there a waiver with to == value, and is its reason non-empty?" Both true. Neither question distinguishes the right reason from someone else's — and the waiver's documented purpose is "permission is per-growth, never standing." A duplicate key makes it standing and deletes the record of what it was granted for, with the gate green.

Red-first, on the real artifact

Same claims.yaml from main, two checkers:

shipped:  51/51 claims hold.                                          EXIT=0
fixed:    claim_check: duplicate key 'reason': declared at line 1311
          and AGAIN at line 1330 — YAML keeps the LAST one silently,
          so the first value is discarded with no diagnostic.          EXIT=1

Mutation-verified: neutering the detection (needle count asserted == 1 before mutating) fails 2 tests and lets the real defective ledger pass again. Control green, 50 tests OK.

The class

#1059's duplicate-key defect in a different file. status_evidence_check.py already parses duplicate-key-strict — but only artifacts/release-*.yaml. The strict loader was never pointed at claims.yaml, the one file every other claim in the repo is checked against. Swept: claims.yaml had exactly 1 duplicate key (this one), artifacts/**/*.yaml and rivet.yaml have 0.

Changes

• claims.yaml — the mapping is split; both reasons are attached to the values they justify. The record is the deliverable; the numbers were always right.
• scripts/claim_check.py — both YAML load sites parse duplicate-key-strict, with a diagnostic naming the discarded value's line (the failure mode is silence, not error).
• scripts/test_claim_check.py — the exact shipped shape refused; the fixed two-waiver split accepted (a gate that only ever fails is as useless as one that only ever passes); safe_load pinned as the counterfactual that keeps the wrong reason; and the repo's own ledger asserted clean — non-vacuity against the real artifact, not only fixtures.

Found while auditing v0.60's ratchet numbers for the release notes.

🤖 Generated with Claude Code

https://claude.ai/code/session_01YJK5LZZEkV5smCY1jKn18L

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!