feat: 리밸런싱 엔진 V1 + GAS 버그 수정 (2026-06-13)

주요 변경:
- tools/build_rebalance_engine_v1.py: REBALANCE_ENGINE_V1 신규
  * account_snapshot 직접 합산(_build_snap_position_map) → 소수주 분리 행 병합
  * 레짐 소스 macro.REGIME_PRELIM 최우선 (GAS 와 동일)
- src/gas_adapter_parts/gdf_06_rebalance.gs: runRebalanceSheet_() 신규
  * Logger.log / getSpreadsheet_() 로 run_all 연동 수정
- src/gas_adapter_parts/gdc_01_fetch_fundamentals.gs
  * _mergePositionRecord_(): 소수주 중복 행 합산 신규
  * parseInt → parseFloat (qty, availQty)
- src/gas_adapter_parts/gdf_01_price_metrics.gs
  * 미보유 종목 SELL_READY → WATCH_EXIT_SIGNAL
- spec/41_release_dag.yaml: build_rebalance_sheet 노드 추가 (step_count 63)
- spec/51_formula_lifecycle_registry.yaml: REBALANCE_ENGINE_V1 등록

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
2026-06-13 13:20:14 +09:00
commit ee3e799de1
1474 changed files with 176087 additions and 0 deletions
+245
View File
@@ -0,0 +1,245 @@
from __future__ import annotations
import argparse
import json
from pathlib import Path
from statistics import mean, quantiles
from typing import Any
ROOT = Path(__file__).resolve().parents[1]
DEFAULT_JSON = ROOT / "GatherTradingData.json"
DEFAULT_HISTORY = ROOT / "Temp" / "proposal_evaluation_history.json"
DEFAULT_OUT = ROOT / "Temp" / "late_chase_attribution_v1.json"
def _load(path: Path) -> dict[str, Any]:
if not path.exists():
return {}
try:
data = json.loads(path.read_text(encoding="utf-8"))
except Exception:
return {}
return data if isinstance(data, dict) else {}
def _parse_rows(value: Any) -> list[dict[str, Any]]:
if isinstance(value, list):
return [x for x in value if isinstance(x, dict)]
if isinstance(value, str):
try:
parsed = json.loads(value)
return _parse_rows(parsed)
except Exception:
return []
return []
def _to_float(value: Any) -> float | None:
try:
if value is None or value == "":
return None
return float(value)
except Exception:
return None
def main() -> int:
ap = argparse.ArgumentParser()
ap.add_argument("--json", default=str(DEFAULT_JSON))
ap.add_argument("--history", default=str(DEFAULT_HISTORY))
ap.add_argument("--out", default=str(DEFAULT_OUT))
args = ap.parse_args()
json_path = Path(args.json)
hist_path = Path(args.history)
out_path = Path(args.out)
if not json_path.is_absolute():
json_path = ROOT / json_path
if not hist_path.is_absolute():
hist_path = ROOT / hist_path
if not out_path.is_absolute():
out_path = ROOT / out_path
payload = _load(json_path)
history = _load(hist_path)
data = payload.get("data") if isinstance(payload.get("data"), dict) else {}
h = data.get("_harness_context") if isinstance(data.get("_harness_context"), dict) else (payload.get("hApex") or {})
entry_rows = _parse_rows(h.get("entry_freshness_json"))
alpha_fb = h.get("alpha_feedback_json") if isinstance(h.get("alpha_feedback_json"), dict) else {}
# Operational samples are drawn from the candidate ledger when a T+5 outcome exists.
# The history does not carry explicit velocity_1d for those rows, so we use
# buy_timing_score as the entry-timing proxy from the same operational record.
recs = history.get("records") if isinstance(history.get("records"), list) else []
op_candidates = [
r for r in recs
if isinstance(r, dict)
and str(r.get("validation_status") or "").upper() != "REPLAY_BACKFILL"
and str(r.get("t5_evaluation_status") or "") == "EVALUATED_T5"
and _to_float(r.get("buy_timing_score")) is not None
]
proxy_field = "buy_timing_score"
proxy_values = [float(r.get(proxy_field)) for r in op_candidates if _to_float(r.get(proxy_field)) is not None]
# Current watchlist remains sourced from the live entry freshness gate.
high_risk = [r for r in entry_rows if float(r.get("late_chase_risk_score") or 0) >= 70]
blocked = [r for r in entry_rows if str(r.get("freshness_state") or "").upper() == "BLOCK_LATE_CHASE"]
pullback_wait = [r for r in entry_rows if str(r.get("freshness_state") or "").upper() == "PULLBACK_WAIT"]
watchlist = []
for r in high_risk:
watchlist.append(
{
"ticker": r.get("ticker"),
"name": r.get("name"),
"late_chase_risk_score": r.get("late_chase_risk_score"),
"freshness_state": r.get("freshness_state"),
"follow_through_state": r.get("follow_through_state"),
"action_hint": "NO_BUY_UNTIL_PULLBACK" if str(r.get("freshness_state")) == "BLOCK_LATE_CHASE" else "WATCH_PULLBACK_ONLY",
}
)
threshold_grid = [20, 30, 40, 50, 60, 70, 80]
threshold_ledger: list[dict[str, Any]] = []
chosen: dict[str, Any] | None = None
for threshold in threshold_grid:
blocked_rows = [r for r in op_candidates if float(r.get(proxy_field)) < threshold]
if not blocked_rows:
continue
matched = sum(1 for r in blocked_rows if r.get("t5_outcome") == "MATCHED")
mismatched = sum(1 for r in blocked_rows if r.get("t5_outcome") == "MISMATCHED")
decisive = matched + mismatched
match_rate = round((matched / decisive) * 100.0, 2) if decisive else None
false_positive_rate = round((matched / decisive) * 100.0, 2) if decisive else None
avg_t5_return = None
t5_returns = [float(r.get("t5_return_pct")) for r in blocked_rows if _to_float(r.get("t5_return_pct")) is not None]
if t5_returns:
avg_t5_return = round(mean(t5_returns), 2)
row = {
"threshold": threshold,
"proxy_field": proxy_field,
"blocked_count": len(blocked_rows),
"matched_count": matched,
"mismatched_count": mismatched,
"decisive_count": decisive,
"match_rate_pct": match_rate,
"false_positive_rate_pct": false_positive_rate,
"avg_t5_return_pct": avg_t5_return,
}
threshold_ledger.append(row)
if chosen is None and false_positive_rate is not None and false_positive_rate <= 20.0:
chosen = row
if len(op_candidates) < 30:
status = "WATCH_PENDING_SAMPLE"
elif chosen is not None:
status = "PASS"
else:
status = "DEGRADE_BUY_PERMISSION"
if chosen is None and threshold_ledger:
chosen = max(threshold_ledger, key=lambda r: float(r.get("match_rate_pct") or 0.0))
# [LC1/NF3] velocity_decile_thresholds — buy_timing_score 실측 분포 10분위 계산
# samples >= 30 이면 실측 분위를 BUY 차단 커트오프 후보로 제공
velocity_decile_thresholds: dict[str, object] = {}
if len(proxy_values) >= 30:
# 10분위 경계값 계산 (1~9 분위점)
decile_cuts = quantiles(proxy_values, n=10)
# T+5 승률 최저 분위 → 차단 임계값 권고
recommended_cut = chosen.get("threshold") if chosen else None
velocity_decile_thresholds = {
"source": "실측 분포 (buy_timing_score 10분위)",
"proxy_field": proxy_field,
"sample_n": len(proxy_values),
"decile_1_pct": round(decile_cuts[0], 2),
"decile_2_pct": round(decile_cuts[1], 2),
"decile_3_pct": round(decile_cuts[2], 2),
"decile_5_pct": round(decile_cuts[4], 2),
"decile_7_pct": round(decile_cuts[6], 2),
"decile_9_pct": round(decile_cuts[8], 2),
"recommended_block_threshold": recommended_cut,
"calibration_status": "CALIBRATED_FROM_LEDGER",
"note": "velocity_1d 실측값 미확보 → buy_timing_score 분위 사용. T+5 최저승률 분위를 BUY 차단 기준으로 권고.",
}
else:
# [LC1] samples < 30 → 프록시값 사용 금지, WATCH_PENDING_SAMPLE 명시
velocity_decile_thresholds = {
"source": "WATCH_PENDING_SAMPLE",
"proxy_field": proxy_field,
"sample_n": len(proxy_values),
"recommended_block_threshold": None,
"calibration_status": "WATCH_PENDING_SAMPLE",
"note": (
f"[LC1] samples={len(proxy_values)}<30 — 실측 분위 캘리브레이션 불가. "
"현재 임계값은 EXPERT_PRIOR(3%/10%). 30건 누적 후 자동 교체."
),
}
# [LC1] late_chase_block_precision — 프록시 100.0 금지, 실측값만
precision_val = chosen.get("match_rate_pct") if chosen else None
if precision_val is not None and len(op_candidates) < 30:
# 표본 부족 시 precision 노출 자체를 WATCH_PENDING_SAMPLE으로 표기
precision_label = "WATCH_PENDING_SAMPLE"
else:
precision_label = f"{precision_val}%" if precision_val is not None else "DATA_MISSING"
result = {
"formula_id": "LATE_CHASE_ATTRIBUTION_V1",
"status": status,
"samples": len(op_candidates) if op_candidates else int(alpha_fb.get("total_samples") or 0),
"operational_samples": len(op_candidates),
"gate_hit_miss_rate_published": True,
# [LC1] velocity_decile_thresholds — 실측 분위 임계값
"velocity_decile_thresholds": velocity_decile_thresholds,
"metrics": {
"late_chase_high_risk_count": len(high_risk),
"late_chase_blocked_count": len(blocked),
"pullback_wait_count": len(pullback_wait),
"chase_entry_rate": float(alpha_fb.get("chase_entry_rate") or 0.0),
"distribution_entry_rate": float(alpha_fb.get("distribution_entry_rate") or 0.0),
"late_chase_proxy_field": proxy_field,
"late_chase_proxy_mean": round(mean(proxy_values), 2) if proxy_values else None,
"late_chase_proxy_min": round(min(proxy_values), 2) if proxy_values else None,
"late_chase_proxy_max": round(max(proxy_values), 2) if proxy_values else None,
# [LC1] 실측 precision — 프록시 100.0 금지
"late_chase_block_precision_label": precision_label,
"late_chase_proxy_match_rate_pct": chosen.get("match_rate_pct") if chosen else None,
"late_chase_proxy_false_positive_rate_pct": chosen.get("false_positive_rate_pct") if chosen else None,
},
"policy": {
"pilot_only_threshold": 0.25,
"no_buy_days_threshold": 0.35,
"applied_mode": (
"NO_BUY_DAYS_3" if float(alpha_fb.get("chase_entry_rate") or 0.0) >= 0.35
else "PILOT_ONLY" if float(alpha_fb.get("chase_entry_rate") or 0.0) >= 0.25
else "NORMAL"
),
# [LC1] 현재 임계값 하드코딩 여부 명시
"velocity_threshold_source": (
"CALIBRATED_FROM_LEDGER" if len(proxy_values) >= 30 else "EXPERT_PRIOR_PENDING_CALIBRATION"
),
},
"threshold_ledger": threshold_ledger,
"watchlist": watchlist,
"supporting_artifacts": [
"Temp/proposal_evaluation_history.json",
"Temp/entry_freshness_json",
],
"note": (
"operational_samples는 proposal_evaluation_history의 비-REPLAY T+5 평가행이며, "
"explicit velocity_1d가 없어 buy_timing_score를 entry-timing proxy로 사용. "
"[LC1] samples<30 구간에서 precision/precision_label=WATCH_PENDING_SAMPLE."
),
}
out_path.parent.mkdir(parents=True, exist_ok=True)
out_path.write_text(json.dumps(result, ensure_ascii=False, indent=2), encoding="utf-8")
print(json.dumps(result, ensure_ascii=False, indent=2))
return 0
if __name__ == "__main__":
raise SystemExit(main())