feat(ask): Phase 3.5b guardrails — verifier + telemetry + grounding 강화

Phase 3.5a(classifier+refusal gate+grounding) 위에 4개 Item 추가:

Item 0: ask_events telemetry 배선
- AskEvent ORM 모델 + record_ask_event() — ask_events INSERT 완성
- defense_layers에 input_snapshot(query, chunks, answer) 저장
- refused/normal 두 경로 모두 telemetry 호출

Item 3: evidence 간 numeric conflict detection
- 동일 단위 다른 숫자 → weak flag
- "이상/이하/초과/미만" threshold 표현 → skip (FP 방지)

Item 4: fabricated_number normalization 개선
- 단위 접미사 건/원 추가, 범위 표현(10~20%) 양쪽 추출
- bare number 2자리 이상만 (1자리 FP 제거)

Item 1: exaone semantic verifier (판단권 잠금 배선)
- verifier_service.py — 3s timeout, circuit breaker, severity 3단계
- direct_negation만 strong, numeric/intent→medium, 나머지→weak
- verifier strong 단독 refuse 금지 — grounding과 교차 필수
- 6-tier re-gate (4라운드 리뷰 확정)
- grounding strong 2+ OR max_score<0.2 → verifier skip

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Hyungi Ahn
2026-04-10 09:49:56 +09:00
parent a0e1717206
commit b2306c3afd
9 changed files with 533 additions and 20 deletions

View File

@@ -26,6 +26,8 @@ class AIConfig(BaseModel):
rerank: AIModelConfig
# Phase 3.5a: exaone classifier (optional — 없으면 score-only gate)
classifier: AIModelConfig | None = None
# Phase 3.5b: exaone verifier (optional — 없으면 grounding-only)
verifier: AIModelConfig | None = None
class Settings(BaseModel):
@@ -86,6 +88,11 @@ def load_settings() -> Settings:
if "classifier" in ai_raw.get("models", {})
else None
),
verifier=(
AIModelConfig(**ai_raw["models"]["verifier"])
if "verifier" in ai_raw.get("models", {})
else None
),
)
if "nas" in raw:

View File

@@ -114,6 +114,10 @@ async def _run_migrations(conn) -> None:
for version, name, path in pending:
sql = path.read_text(encoding="utf-8")
_validate_sql_content(name, sql)
if "schema_migrations" in sql.lower():
raise ValueError(
f"Migration {name} must not modify schema_migrations table"
)
logger.info(f"[migration] {name} 실행 중...")
# raw driver SQL 사용 — text() 의 :name bind parameter 해석으로
# SQL 주석/literal 에 콜론이 들어가면 InvalidRequestError 발생.