Submitted with 15 minutes to spare, which for me counts as early.

Submitted with 15 minutes to spare, which for me counts as early.

❔ no checkable claim

README.md and BRIEF.md are finally written, and eval/cases.jsonl now holds 30 hand-labelled sentences checked against 8 real commits.

Every named file, function, and integration appears in the diff.

• README.md (README.md)
• BRIEF.md (BRIEF.md)
• eval/cases.jsonl (eval/cases.jsonl)

Half the adversarial cases exist just to catch the model naming a real file in the wrong range or a near-miss filename.

❔ no checkable claim

In cdl/llm.py the extraction still runs through forced tool_use, and cdl/grounding/check.py's claim_vs_diff is the only thing allowed to decide SUPPORTED or UNSUPPORTED.

Every named file, function, and integration appears in the diff.

• cdl/llm.py (cdl/llm.py)
• cdl/grounding/check.py (cdl/grounding/check.py)
• BRIEF.md:8 (tool_use) — **Offline: `make test` — 65 tests pass** at submission. They cover the diff parser (hunk-header line numbers), every verdict rule (file / symbol / integration matching, alias expansion for compound names, evidence ranking and the 3-per-entity cap, the one-UNVERIFIABLE gate boundary), every staleness rule (removed file, renamed file, net-removed symbol; a *moved* symbol and a body edit are not stale), the substring guard, `tool_use` schema validation and the length-truncation fallback, the publish gate, the approval mid-check, corrections, dashboard rendering, and the eval harness itself.
• README.md:58 (tool_use) — | **Anthropic** | Drafting and entity extraction only, via forced `tool_use` JSON schemas. | Claude Sonnet; never decides a verdict. |
• cdl/drafter.py:5 (claim_vs_diff) — must not decide support outside ``grounding.check.claim_vs_diff``.
• cdl/eval/run_eval.py:17 (claim_vs_diff) — from ..grounding.check import check_staleness, claim_vs_diff
• cdl/grounding/check.py:156 (claim_vs_diff) — def claim_vs_diff(claim: Claim, diff: DiffContext) -> Verdict:
• cdl/drafter.py:29 (SUPPORTED) — marker = {"SUPPORTED": "✅", "UNSUPPORTED": "⛔", "UNVERIFIABLE": "❔"}[verdict.status]
• cdl/eval/run_eval.py:23 (SUPPORTED) — LABELS = ["SUPPORTED", "UNSUPPORTED", "UNVERIFIABLE"]
• cdl/grounding/check.py:198 (SUPPORTED) — status="SUPPORTED",
• cdl/drafter.py:29 (UNSUPPORTED) — marker = {"SUPPORTED": "✅", "UNSUPPORTED": "⛔", "UNVERIFIABLE": "❔"}[verdict.status]
• cdl/eval/run_eval.py:23 (UNSUPPORTED) — LABELS = ["SUPPORTED", "UNSUPPORTED", "UNVERIFIABLE"]
• cdl/grounding/check.py:190 (UNSUPPORTED) — status="UNSUPPORTED",