bilingual-terminology-review-gate

When to Use

When reviewing A2AD terminology_bilingual lanes for NCLEX or similar multilingual clinical content, where every changed TERM-REC key must be individually validated and accounted for before screen→reviewed promotion. Use when a terminology review verdict must justify every changed key with clinical/linguistic evidence, risk flagging, and standards cross-reference.

Procedure

1. Preparation:

   • Receive payload.json with sourceBundle.files[].content containing changedRecords: [{ key, pair, record }]
   • Establish standards baseline: KMA edition (e.g., 6.1), OpenRN text editions (Health Alterations, Fundamentals, Skills), clinical practice guidelines
   • Parse intentContract.invariants (e.g., "account for all N changed keys exactly once", "missing/duplicated/mistranslated term returns BLOCK")

2. Per-key validation loop (for each TERM-REC key):

   • Semantic equivalence: EN/KO phrase pair for idea-for-idea match; flag if KO adds, drops, or reinterprets concepts
   • Clinical term accuracy: Cross-reference each medical term (stroke, dysphagia, aspiration pneumonia, neurologic, etc.) against KMA preferred/accepted list and OpenRN source text edition/page
   • Numerics and units: Verify numbers (% SpO₂, angles, etc.) and unit names (IV→정맥, not transliteration) match EN exactly
   • Forbidden-word enforcement: Exclude discouraged terms (e.g., 中風 for stroke) if listed in baseline standard
   • Abbreviation expansion: Ensure first-use abbreviations are spelled out in KO; no abbreviation-only translations allowed
   • Nesting context: If a term appears in nested fields (phases[].body, items[].options[].why), confirm all instances use the same approved KO term

3. Evidence tethering:

   • For each clinical term: cite KMA entry (version, edition, retrievedAt, exactMatchCount)
   • Link clinical rationale to OpenRN pages (§ number, edition, page range) or hospital policy document
   • Record diffHash/intentHash agreement for payload integrity

4. Structured findings:

   • Per-key entry: TERM-REC <key> PASS: <field>, <EN term>→<KO term> <status>, <clinical ref>
   • Aggregate: total keys, error keys (expect 0), risks, non-blocking recommendations
   • Example non-blocking flag: "'plain water'→'생수' is colloquial; consider '일반 물' for precision with nectar-thick contrast"

5. Verdict issuance:

   • PASS: All N changed keys accounted for, 0 critical errors, 0 concept mismatches, semantic equivalence confirmed, clinical terms standards-aligned
   • BLOCK: Any key missing, duplicated, mistranslated, abbreviation-only, or unsupported medical term; OR semantic drift affecting clinical safety

6. Risk and next-action:

   • If sourceOnly=true: note "GitHub live diff not performed; re-examine headSha against actual PR if needed"
   • If verdict is PASS: affirm that PASS does not authorize merge, does not count as RN licensure review, does not affect source.verified status
   • Identify parallel lanes (high_risk_safety, content_clinical, evidence_adversarial) still pending and required for promotion

Safety

• Clinical accuracy is binding: Terminology defects mislead clinicians. Never waive confirmed medical-term mismatch or semantic drift.
• Invariant adherence: Contract requirements (e.g., "all N changed keys exactly once") are binding; violation→BLOCK.
• Scope clarity: This lane validates terminology only; it does NOT approve merge, does NOT substitute for RN licensure review, does NOT grant paid-pool eligibility. Document in verdict.
• Standards baseline: Always cite specific KMA edition and OpenRN edition with timestamp. Do not drift to colloquial or local dialect unless explicitly approved.

Verification

• Self-check before PASS:

  • Counted exactly N TERM-REC keys (the contract-declared count); zero missing or duplicated
  • Every key has EN/KO pair with semantic equivalence confirmed
  • Every medical term has KMA reference with exactMatchCount > 0
  • No forbidden terms; all abbreviations expanded in KO
  • All numerics and units match EN exactly
  • Invariants documented and affirmed in verdict

• Post-PASS:

  • Confirm PASS does not auto-promote screen→reviewed; parallel lanes must complete
  • If sourceOnly=true, flag: "GitHub live diff re-validation recommended before final merge"
  • Link PASS verdict to claim task ID (e.g., RNM-20260810-008) for audit trail