init: company-haness 설계

.claude/hooks/bench_cascade/aggregate.py

@@ -0,0 +1,48 @@
+"""blinded paired pairwise 집계 수학. 실질 arm 관점: pair 의 첫 arm = "first", 둘째 = "second".
+forward orientation 은 X=첫 arm, reversed 는 X=둘째 arm(뒤집힘). 단순평균 금지 — 순위 기반."""
+from collections import Counter
+
+
+def normalize(orientation, winner):
+    """judge 의 X/Y 승자를 실질 arm 관점(first/second)으로 정규화. tie 는 그대로 tie."""
+    if winner == "tie":
+        return "tie"
+    if orientation == "forward":
+        return "first" if winner == "X" else "second"
+    return "second" if winner == "X" else "first"  # reversed: X=둘째 arm
+
+
+def stable(fwd_real, rev_real):
+    """두 orientation 이 같은 실질 승자를 판정하는가."""
+    return fwd_real == rev_real
+
+
+def preference_score(wins, ties, valid_stable):
+    """(wins + 0.5*ties) / valid_stable. arm 의 우호도."""
+    if valid_stable <= 0:
+        return 0.0
+    return (wins + 0.5 * ties) / valid_stable
+
+
+def panel_agreement(stable_verdicts):
+    """최빈 판정이 차지하는 비중. 0.0 if empty."""
+    if not stable_verdicts:
+        return 0.0
+    top = Counter(stable_verdicts).most_common(1)[0][1]
+    return top / len(stable_verdicts)
+
+
+def flip_consistency(paired):
+    """paired: [{fwd, rev}...] 각 실질 arm 관점. fwd==rev 면 flip 일관성 유지."""
+    if not paired:
+        return 0.0
+    ok = sum(1 for p in paired if p["fwd"] == p["rev"])
+    return ok / len(paired)
+
+
+def panel_verdict(stable_verdicts):
+    """panel 의 최종 판정. unstable = 투표 일관성 부족."""
+    if len(stable_verdicts) < 2:
+        return "unstable"
+    verdict, cnt = Counter(stable_verdicts).most_common(1)[0]
+    return verdict if cnt >= 2 else "unstable"