docs: note hazard-analysis citations drift after the rag.ts decomposition — search by symbol

docs/clinical-hazard-analysis.md

@@ -11,7 +11,7 @@
 
 Findings were produced by a blind multi-agent fan-out — one analyst per harm pathway, each unaware of the others — followed by orchestrator-level dedupe and **adversarial verification of every finding against the live code** before it entered this document. Where an agent's claim did not survive verification it was corrected or dropped (examples: model-supplied citation identity fields are inert, so citation-ID spoofing via document text is _blocked_, not a hazard; `indexing_quality.issues` reaches the prompt raw but is built from deterministic strings and is not attacker-controllable; the audit's H4 ward-note "fix" was verified to live in **dead code**).
 
-The 2026-07-01 repository audit ([`docs/audit/repo-audit-2026-07-01.md`](audit/repo-audit-2026-07-01.md)) findings **H1–H4** are used as evidence that these pathways are real: each was a confirmed, reproduced defect on exactly one of the pathways below. Their remediations (audit "Remediation — 2026-07-02", merged via PR #123) are the **current controls** named here — so each control is cited at its present location, and each is re-examined for residual gaps.
+The 2026-07-01 repository audit ([`docs/audit/repo-audit-2026-07-01.md`](audit/repo-audit-2026-07-01.md)) findings **H1–H4** are used as evidence that these pathways are real: each was a confirmed, reproduced defect on exactly one of the pathways below. Their remediations (audit "Remediation — 2026-07-02", merged via PR #123) are the **current controls** named here, each re-examined for residual gaps. The `file:line` citations were refreshed after the `rag.ts` monolith was decomposed into the `src/lib/rag/*` directory (PR #998) — treat the line numbers as accurate as-of that refresh and **search by the named symbol** if a later refactor has shifted one (`docs:check-links` validates the file path, not the line).
 
 ## 2. Risk framework