전 저장소 리팩터링과 SSOT 정비

apps/api/app/services/orchestrator.py

@@ -41,13 +41,13 @@ from ..contracts.engine_gateway import (
     StreamTokenEvent,
 )
 from . import guardrail, persona, state_machine
+from .llm_audit import LlmAuditHook, generate_with_audit, record_llm_audit
 from .persona import PersonaCard, PersonaStateContext, TurnMemory
-from .state_machine import SessionState, Stage
+from .state_machine import SessionState
 
 # 평가 훅 타입: U_t(수련생 마스킹 발화) + 내담자응답 + 상태 → 평가 결과(dict)
 # Features evaluator 가 이 시그니처에 맞춰 함수를 주입한다(여기선 호출만).
 EvalHook = Callable[["TurnContext", str], Awaitable[Optional[dict]]]
-LlmAuditHook = Callable[[dict[str, Any]], Awaitable[None]]
 
 
 def turn_evaluation_error_payload(ctx: "TurnContext", error: BaseException | str) -> dict[str, Any]:
@@ -253,20 +253,7 @@ async def run_turn_generate(
     reply = ""
     safety_flagged = ctx.crisis is not None and ctx.crisis.escalate
     for attempt in range(2):
-        started = time.perf_counter()
-        resp = await engine.generate(req)
-        latency_ms = int((time.perf_counter() - started) * 1000)
-        await _record_llm_audit(
-            audit_hook,
-            session_id=ctx.session_id,
-            provider=resp.provider,
-            model=resp.model,
-            tokens_in=resp.tokens_in,
-            tokens_out=resp.tokens_out,
-            cost_usd=resp.cost_usd,
-            inference_geo=resp.inference_geo,
-            latency_ms=latency_ms,
-        )
+        resp = await generate_with_audit(engine, req, audit_hook)
 
         # 5) 출력 가드레일 — 수단 차단 + persona 품질 재생성
         guard = guardrail.sanitize_client_reply(
@@ -451,7 +438,7 @@ async def run_turn_stream(
                 yield StreamEvent("token", {"text": guard.text})
 
         latency_ms = int((time.perf_counter() - started) * 1000)
-        await _record_llm_audit(
+        await record_llm_audit(
             audit_hook,
             session_id=ctx.session_id,
             provider=str(stream_meta.get("provider") or engine.engine_mode),
@@ -493,18 +480,6 @@ async def run_turn_stream(
         yield StreamEvent("error", {"detail": str(e)})
 
 
-async def _record_llm_audit(
-    audit_hook: Optional[LlmAuditHook],
-    **payload: Any,
-) -> None:
-    if audit_hook is None:
-        return
-    try:
-        await audit_hook(payload)
-    except Exception:
-        return
-
-
 def _optional_str(value: Any) -> Optional[str]:
     if value is None:
         return None