Source: qa/xref-r3-QA2-request.md
# QA2 request — Astra Round 3 authoring integration and re-emit proof Status: ready for independent QA2 verification. QA1 passed Round 2 in `xref-r2-QA1.md`. Review the three changed tracked authoring files (`parse_docs.py`, `eyecite_extract.py`, `test_eyecite_extract.py`). The reviewed resolver is reused from its tracked neov2 module via a lazy loader, rather than introducing an untracked authoring module. Confirm default-off behavior, source-text preservation, cache behavior, and full-checkout dependency. Evidence: `neov2/xref/out/r3/REBUILD_PLAN.md`, `reemit_evidence.json`, `sample_anchor_additions.diff`, and test logs. The actual 80-case scratch run has zero annotation-caused HTML differences; compared with current pages it has exactly four added anchors on two pages, zero removed anchors, zero text or non-anchor markup differences. Four additions arise from prior U.S.C.A. renderer fixes, not the research field. No full emit, deploy, S3, or CloudFront operation ran. Verify the exact operator-only command and the distinction between bounded sample proof and a future full-corpus gate. `/tmp` cannot hold a full run. Check court-level classifier edge cases and its explicit 570-record unknown bucket. Main tests194 passed; five skipped fixture tests subsequently passed; only two optional image-library tests remain skipped. Worktree: `/tmp/neochart-astra-r3`; changes will be committed/pushed wiki-style. Please record the verdict in `neov2/qa/xref-r3-QA2.md`. Production publication remains operator-gated regardless of this code review.