kjh2064
af1236202d
- F14: late_chase_risk_score 검증 * GAS가 유일한 생산처 (Python canonical 없음) * migration_action: KEEP_IN_GAS로 정정, status: DONE - F02/F03/F04/F06: priceBasis 로직 포팅 * formulas/price_basis_v1.py: select_price_basis_tier2/tier1 구현 * tests/parity/test_price_basis_parity_v1.py: 8 parity 테스트 (모두 PASS) * GAS Number.isFinite() 의미론 정확히 재현 (math.isfinite 사용) * 모든 테스트 112/112 PASS 남은 작업 (4개): - F05: decision_logic (action assignment) - F07: score_logic (threshold addition) - F10: routing decision - F15: late_chase_gate Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
84 lines
3.0 KiB
Python
84 lines
3.0 KiB
Python
#!/usr/bin/env python3
|
|
from __future__ import annotations
|
|
|
|
import json
|
|
import sys
|
|
from pathlib import Path
|
|
|
|
ROOT = Path(__file__).resolve().parent.parent
|
|
|
|
def main() -> int:
|
|
accuracy_path = ROOT / "Temp" / "prediction_accuracy_harness_v2.json"
|
|
honest_path = ROOT / "Temp" / "honest_performance_guard_v1.json"
|
|
|
|
if not accuracy_path.exists():
|
|
print(f"accuracy harness file missing: {accuracy_path}")
|
|
return 1
|
|
if not honest_path.exists():
|
|
print(f"honest guard file missing: {honest_path}")
|
|
return 1
|
|
|
|
try:
|
|
acc_data = json.loads(accuracy_path.read_text(encoding="utf-8"))
|
|
honest_data = json.loads(honest_path.read_text(encoding="utf-8"))
|
|
except Exception as e:
|
|
print(f"Failed to parse json: {e}")
|
|
return 1
|
|
|
|
errors = []
|
|
|
|
# 1. factor_outcome_join_rate_pct >= 95
|
|
audit = acc_data.get("data_origin_audit", {})
|
|
op_count = audit.get("operational_sample_count", 0)
|
|
untagged = audit.get("untagged_row_count", 0)
|
|
|
|
if op_count > 0:
|
|
factor_outcome_join_rate_pct = 100.0 * (1.0 - (untagged / op_count))
|
|
else:
|
|
factor_outcome_join_rate_pct = 100.0
|
|
|
|
if factor_outcome_join_rate_pct < 95.0:
|
|
errors.append(f"factor_outcome_join_rate_pct is {factor_outcome_join_rate_pct:.2f}% (Expected >= 95%)")
|
|
|
|
# 2. live_sample_under_30_unlock_count == 0
|
|
live_sample_under_30_unlock_count = 0
|
|
calibration_state = acc_data.get("calibration_state", "")
|
|
t5_sample = acc_data.get("t5_sample", 0)
|
|
|
|
if t5_sample < 30 and calibration_state not in ("INSUFFICIENT_SAMPLES", "UNKNOWN", ""):
|
|
if calibration_state == "CALIBRATED":
|
|
live_sample_under_30_unlock_count += 1
|
|
errors.append(f"t5_sample={t5_sample} < 30 but calibration_state is unlocked ({calibration_state})")
|
|
|
|
# 3. replay_live_mixed_metric_count == 0
|
|
replay_live_mixed_metric_count = 0
|
|
replay_in_live = audit.get("replay_in_live_stats", 0)
|
|
if replay_in_live > 0:
|
|
replay_live_mixed_metric_count += 1
|
|
errors.append(f"Replay samples mixed in live stats: {replay_in_live}")
|
|
|
|
gate_passed = (factor_outcome_join_rate_pct >= 95.0) and \
|
|
(live_sample_under_30_unlock_count == 0) and \
|
|
(replay_live_mixed_metric_count == 0)
|
|
|
|
result = {
|
|
"formula_id": "HONEST_PERFORMANCE_GUARD_VALIDATOR_V1",
|
|
"factor_outcome_join_rate_pct": factor_outcome_join_rate_pct,
|
|
"live_sample_under_30_unlock_count": live_sample_under_30_unlock_count,
|
|
"replay_live_mixed_metric_count": replay_live_mixed_metric_count,
|
|
"errors": errors,
|
|
"gate": "PASS" if gate_passed else "FAIL"
|
|
}
|
|
|
|
# Write output to Temp
|
|
out_dir = ROOT / "Temp"
|
|
out_dir.mkdir(parents=True, exist_ok=True)
|
|
out_path = out_dir / "honest_performance_guard_validation_v1.json"
|
|
out_path.write_text(json.dumps(result, ensure_ascii=False, indent=2), encoding="utf-8")
|
|
|
|
print(json.dumps(result, ensure_ascii=True, indent=2))
|
|
return 0 if gate_passed else 1
|
|
|
|
if __name__ == "__main__":
|
|
sys.exit(main())
|