perf: pre-compute query phrases once per rankAnswerEvidence call

[Copilot]

Summary

• phraseScore rebuilt the same token n-gram array for every scored result. With 30–50+ results per query, phrase construction was O(N × tokens²); this moves it to O(tokens²) by computing once and passing down.

Before:

// Called N times — phrases rebuilt from same tokens each time
function phraseScore(tokens: string[], haystack: string) {
  const phrases: string[] = [];
  for (let size = Math.min(4, tokens.length); size >= 2; size -= 1) { ... }
  ...
}

After:

// Built once in rankAnswerEvidence, passed to each answerEvidenceScore call
function buildQueryPhrases(tokens: string[]): string[] { ... }
function phraseScore(phrases: string[], haystack: string) { ... }

const phrases = buildQueryPhrases(tokens); // once per query
results.map(result => answerEvidenceScore(tokens, phrases, result, queryClass));

Verification

• npm run verify:cheap
• npm run verify:ui when UI, routing, styling, browser behavior, reduced-motion, or forced-colors behavior changed
• npm run verify:release before release or handoff confidence claims
• npm run format:check
• npm run check:production-readiness when clinical workflow, privacy, environment, Supabase, source governance, or deployment behavior changed

Clinical Governance Preflight

Complete this section when the change touches ingestion, answer generation, search/ranking, source rendering, document access, privacy, production env, or clinical output.

• Source-backed claims still require linked source verification before clinical use
• No patient-identifiable document workflow was introduced or expanded without explicit governance approval
• Supabase target remains Clinical KB Database (sjrfecxgysukkwxsowpy)
• Service-role keys and private document access remain server-only
• Demo/synthetic content remains clearly separated from real clinical sources
• Source metadata, review status, and outdated/unknown-source behavior remain conservative
• Deployment classification/TGA SaMD impact was checked when clinical decision-support behavior changed

Notes

• No change in ranking behaviour — identical outputs, less work per query.
• Remaining 9 identified improvements: Array.from(queryTerms) × 4 allocations per result in buildMatchExplanation; linear Array.includes membership checks in evidence.ts; DELETE+INSERT vs upsert in replaceSharedCacheRow; structuredClone on every cache read; redundant combined renormalization in resultTexts; eager buildClinicalTextSearchQuery in normalizedCacheQuery; no explicit response compression config; imageMatchScore re-splits source on each token; linear cache-key scan in invalidateRagCachesForOwner.