캘리브레이션 거버넌스 도구 + WBS-7.1/7.2 실증 격차 가시화

캘리브레이션 백로그 → 우선순위 → 검토리포트 → 승인목록 → 결정초안으로
이어지는 임계값 보정 거버넌스 파이프라인을 추가하고, 2026-06-21
비판적 리뷰에서 발견한 두 가지 stale-수치 문제를 도구 차원에서 해소한다.

- registry_health(): 190여 개 임계값의 source별(SPEC_DERIVED/EXPERT_PRIOR/
  PROVISIONAL/CALIBRATED) 분포를 매 실행마다 자동 집계 — 수동 grep 불필요
- live_t5_status(): T+5 적중률을 하드코딩(35.86 리터럴) 대신
  Temp/prediction_accuracy_harness_v2.json에서 항상 최신값으로 읽음
- spec/calibration_registry.yaml: SEMI_CLUSTER_CAP_RISK_OFF 중복 id로
  인한 조용한 무시 버그 수정(SEMI_CLUSTER_CAP_RISK_OFF_MWA로 분리)
- spec/27_bch_calibration_runbook.yaml: current_status_2026_06_21 블록
  신설(단일 진실원천), 기존 05-30 스냅샷은 "역사적, 현재로 인용 금지"로 명시

tools/build_calibration_priority_v1.py

@@ -29,6 +29,41 @@ ROOT   = Path(__file__).resolve().parent.parent
 AFL    = ROOT / "Temp" / "alpha_feedback_loop_v2.json"
 REG    = ROOT / "spec" / "calibration_registry.yaml"
 OUTPUT = ROOT / "Temp" / "calibration_priority_v1.json"
+PREDICTION_ACCURACY = ROOT / "Temp" / "prediction_accuracy_harness_v2.json"
+
+
+def registry_source_breakdown(reg_index: dict[str, dict]) -> dict:
+    """WBS-7.1(2026-06-21) — calibration_registry.yaml 전체의 source별 분포를 매 실행마다
+    집계해 'CALIBRATED 비율이 실제로 몇 %인가'를 사람이 grep으로 직접 세지 않아도
+    항상 최신 상태로 노출한다(2026-06-21 비판적 리뷰 0c절에서 0/190 발견 당시 수동 집계 필요했던 문제 해소)."""
+    counts: dict[str, int] = {"SPEC_DERIVED": 0, "EXPERT_PRIOR": 0, "PROVISIONAL": 0, "CALIBRATED": 0}
+    for entry in reg_index.values():
+        source = str(entry.get("source", "")).upper()
+        if source in counts:
+            counts[source] += 1
+    total = sum(counts.values())
+    return {
+        "total_thresholds": total,
+        "counts": counts,
+        "calibrated_pct": round(100.0 * counts["CALIBRATED"] / total, 2) if total else 0.0,
+        "unvalidated_pct": round(100.0 * (counts["SPEC_DERIVED"] + counts["EXPERT_PRIOR"]) / total, 2) if total else 0.0,
+    }
+
+
+def live_t5_status() -> dict:
+    """WBS-7.2/7.1(2026-06-21) — T+5 수치를 하드코딩하지 않고 항상 최신 산출물에서 읽는다.
+    Temp/prediction_accuracy_harness_v2.json이 없거나 sample=0이면 정직하게 DATA_GATED로 보고한다."""
+    if not PREDICTION_ACCURACY.exists():
+        return {"status": "ARTIFACT_MISSING", "t5_sample": 0, "t5_match_rate_pct": None}
+    data = load_json(PREDICTION_ACCURACY)
+    t5_sample = int(data.get("t5_sample") or 0)
+    t5_rate = data.get("t5_op_rate")
+    return {
+        "status": "DATA_GATED" if t5_sample == 0 else "OK",
+        "as_of_date": data.get("as_of_date"),
+        "t5_sample": t5_sample,
+        "t5_match_rate_pct": t5_rate,
+    }
 
 if sys.stdout.encoding and sys.stdout.encoding.lower() not in ("utf-8", "utf8"):
     sys.stdout = open(sys.stdout.fileno(), mode="w", encoding="utf-8", buffering=1)
@@ -90,6 +125,42 @@ def load_registry(p: Path) -> dict[str, dict]:
     return {t["id"]: t for t in data.get("thresholds", []) if "id" in t}
 
 
+def _priority_from_registry_entry(entry: dict, source_tag: str, urgency_bias: int) -> dict:
+    sample_n = int(entry.get("sample_n", 0) or 0)
+    source = str(entry.get("source", "EXPERT_PRIOR"))
+    threshold_class = str(entry.get("threshold_class", "standard"))
+    urgency = urgency_bias
+    if source == "EXPERT_PRIOR":
+        urgency += 10
+    if source == "PROVISIONAL":
+        urgency += 20
+    if threshold_class == "live_critical":
+        urgency += 15
+    if sample_n == 0:
+        urgency += 5
+    if sample_n > 0:
+        urgency += max(0, 30 - sample_n)
+    return {
+        "calibration_id": entry.get("id", ""),
+        "current_value": entry.get("value"),
+        "owner_formula": entry.get("owner_formula", ""),
+        "source": source,
+        "sample_n": sample_n,
+        "linked_factor": source_tag,
+        "alpha_action": "registry_review",
+        "urgency_score": urgency,
+        "calibration_path": (
+            (
+                "표본 30건 이상 확보 후 PROVISIONAL 승격 → "
+                if sample_n >= 30
+                else f"표본 {30 - sample_n}건 추가 수집 후 PROVISIONAL 승격 → "
+            )
+            + "실측 T+5 승률 기반 최적값 backtest → CALIBRATED 확정"
+        ),
+        "rationale": f"source={source}, class={threshold_class}, sample_n={sample_n}",
+    }
+
+
 def main() -> int:
     afl_data  = load_json(AFL)
     reg_index = load_registry(REG)
@@ -112,48 +183,32 @@ def main() -> int:
     priority_list: list[dict] = []
 
     for adj in adjustments:
-        factor   = adj.get("factor", "")
-        action   = adj.get("action", "")
-        rationale = adj.get("rationale", "")
-        reg_ids  = FACTOR_TO_REGISTRY.get(factor, [])
+        factor = str(adj.get("factor", ""))
+        action = str(adj.get("action", ""))
+        rationale = str(adj.get("rationale", ""))
+        reg_ids = FACTOR_TO_REGISTRY.get(factor, [])
 
         for rid in reg_ids:
             reg_entry = reg_index.get(rid)
             if not reg_entry:
                 continue
-            source   = reg_entry.get("source", "EXPERT_PRIOR")
-            sample_n = int(reg_entry.get("sample_n", 0) or 0)
-            value    = reg_entry.get("value")
-            formula  = reg_entry.get("owner_formula", "")
+            item = _priority_from_registry_entry(reg_entry, factor, miss5_count if factor == "passive_signal_quality" else 0)
+            item["alpha_action"] = action or "feedback_review"
+            if rationale:
+                item["rationale"] = rationale[:200]
+            priority_list.append(item)
 
-            # 보정 우선도 점수: miss5_count 기여 + 미보정 가중
-            urgency = 0
-            if factor == "passive_signal_quality":
-                urgency += miss5_count  # miss가 많을수록 높은 urgency
-            if source == "EXPERT_PRIOR":
-                urgency += 10
-            if sample_n == 0:
-                urgency += 5
-
-            priority_list.append({
-                "calibration_id":   rid,
-                "current_value":    value,
-                "owner_formula":    formula,
-                "source":           source,
-                "sample_n":         sample_n,
-                "linked_factor":    factor,
-                "alpha_action":     action,
-                "urgency_score":    urgency,
-                "calibration_path": (
-                    (
-                        "표본 30건 이상 확보 후 PROVISIONAL 승격 → "
-                        if sample_n >= 30
-                        else f"표본 {30 - sample_n}건 추가 수집 후 PROVISIONAL 승격 → "
-                    )
-                    + "실측 T+5 승률 기반 최적값 backtest → CALIBRATED 확정"
-                ),
-                "rationale":        rationale[:200] if rationale else "",
-            })
+    if not priority_list:
+        # alpha_feedback_loop가 비어 있어도 registry 자체의 보정 debt를 추적할 수 있게 한다.
+        for reg_id, reg_entry in reg_index.items():
+            source = str(reg_entry.get("source", "EXPERT_PRIOR"))
+            if source not in {"EXPERT_PRIOR", "PROVISIONAL"}:
+                continue
+            tag = f"registry:{source.lower()}"
+            item = _priority_from_registry_entry(reg_entry, tag, 0)
+            if source == "PROVISIONAL":
+                item["urgency_score"] += 5
+            priority_list.append(item)
 
     # 중복 제거 (같은 rid, 높은 urgency 유지)
     seen: dict[str, dict] = {}
@@ -177,7 +232,19 @@ def main() -> int:
     print(f"  Step 2 (30건 후): ALEG_V2_GATE1_BLOCK_PCT 3.0% → 실측 최적값으로 PROVISIONAL 승격")
     print(f"  Step 3 (50건 후): DSD_V1 가중치 logistic regression 최적화")
     print(f"  Step 4 (100건 후): K2_SPLIT_RATIO backtest 비교 → CALIBRATED 확정")
-    print(f"  miss5_count={miss5_count}건 → passive_signal_quality 개선이 T+5 35.86%→50%+ 핵심")
+    registry_health = registry_source_breakdown(reg_index)
+    t5_status = live_t5_status()
+
+    print(f"\n  [캘리브레이션 레지스트리 건강도] (WBS-7.1)")
+    print(f"  total={registry_health['total_thresholds']} {registry_health['counts']}")
+    print(f"  CALIBRATED={registry_health['calibrated_pct']}%  미검증(SPEC_DERIVED+EXPERT_PRIOR)={registry_health['unvalidated_pct']}%")
+
+    if t5_status["status"] == "DATA_GATED":
+        print(f"  miss5_count={miss5_count}건 → T+5 현재 DATA_GATED(sample=0) — passive_signal_quality 개선 영향은 표본 누적 후 측정 가능")
+    elif t5_status["status"] == "ARTIFACT_MISSING":
+        print(f"  miss5_count={miss5_count}건 → T+5 산출물 없음(Temp/prediction_accuracy_harness_v2.json) — 먼저 생성 필요")
+    else:
+        print(f"  miss5_count={miss5_count}건 → T+5={t5_status['t5_match_rate_pct']}% (as_of={t5_status.get('as_of_date')}) → passive_signal_quality 개선 핵심")
 
     result = {
         "status": "CALIBRATION_PRIORITY_OK",
@@ -191,10 +258,14 @@ def main() -> int:
             "step3": "50건 후: DSD_V1 가중치 logistic regression 최적화",
             "step4": "100건 후: K2_SPLIT_RATIO 30/70~60/40 backtest → CALIBRATED",
         },
+        "priority_basis": "alpha_feedback_loop_v2" if adjustments else "registry_warning_fallback",
+        "registry_health": registry_health,
         "target_improvement": {
-            "current_t5_pct": 35.86,
+            "t5_status": t5_status["status"],
+            "current_t5_pct": t5_status["t5_match_rate_pct"],
+            "t5_as_of_date": t5_status.get("as_of_date"),
             "target_t5_pct": 55.0,
-            "key_lever": "passive_signal_quality (miss5_count=51건 개선)",
+            "key_lever": f"passive_signal_quality (miss5_count={miss5_count}건 개선)",
         },
     }