vignette/scripts/backfill-claude-token-usage.py
Yun Chan 16e791e044 G0~G8 성과·동맹 측정 OS 작업 일괄 고정
8월 7일까지 워킹트리에만 남아 있던 미커밋 작업을 커밋한다. 여러 사본
폴더(worktree·clone)에 흩어져 있던 중간 스냅샷을 정리하기 전에 원본을
git 이력으로 고정하는 것이 목적이다.

- contracts/routes/services: measurement, outcome_trajectory, rupture_repair,
  deliberate_practice, calibration_transfer, supervision_research,
  multimodal_alliance, continuous_improvement 계열 신규 모듈과 테스트
- infra/db/init: 07~16 마이그레이션(측정 기반~calibration transfer 실행)
- apps/web: 세션 리뷰 카드·관리 화면·E2E 스펙 추가
- docs/ops: G0~G8 라이브 통합·배포·롤백 증거 문서와 evidence JSON/PNG
- scripts: smoke·ledger·릴리스 에이전트·NAS 프리뷰 운영 스크립트

engine.public 로그 .bak과 apps/web/test-results 산출물은 커밋에서 제외했다.
2026-08-08 01:30:53 +09:00

136 lines
4.9 KiB
Python

#!/usr/bin/env python
"""과거 Claude CLI 턴을 로컬 Claude JSONL의 실제 usage로 보수적으로 백필한다.
기본은 dry-run이다. 본문은 출력하지 않으며, DB text hash + 생성 시각 창에 후보가
정확히 하나인 기존 0/0 Claude 턴만 --apply 대상으로 삼는다.
"""
from __future__ import annotations
import argparse
import asyncio
import json
import os
from pathlib import Path
import sys
def _repo_root() -> Path:
return Path(__file__).resolve().parents[1]
def parse_args() -> argparse.Namespace:
default_root = (
Path(os.environ.get("USERPROFILE", ""))
/ ".claude"
/ "projects"
/ "D--workspace-vignette-apps-api"
)
parser = argparse.ArgumentParser(
description="Backfill exact Claude token usage from local Claude JSONL metadata.",
)
parser.add_argument("--claude-project-dir", type=Path, default=default_root)
parser.add_argument("--before-seconds", type=float, default=30.0)
parser.add_argument("--after-seconds", type=float, default=180.0)
parser.add_argument("--apply", action="store_true")
parser.add_argument(
"--expected-matches",
type=int,
help="Required with --apply; aborts if the current exact-match count differs.",
)
return parser.parse_args()
async def run(args: argparse.Namespace) -> dict[str, object]:
api_root = _repo_root() / "apps" / "api"
sys.path.insert(0, str(api_root))
from app import db
from app.services.claude_usage_backfill import (
load_claude_usage_candidates,
match_claude_usage,
)
source_root = args.claude_project_dir.resolve()
if not source_root.is_dir():
raise RuntimeError(f"Claude project directory not found: {source_root}")
candidates = load_claude_usage_candidates(source_root)
await db.init_pool()
try:
async with db.get_pool().acquire() as conn:
rows = await conn.fetch(
"""
SELECT id::text, created_at, text_masked
FROM app.turns
WHERE speaker = 'client'
AND llm_provider = 'claude_cli'
AND COALESCE(tokens_in, 0) <= 0
AND COALESCE(tokens_out, 0) <= 0
ORDER BY created_at, id
"""
)
report = match_claude_usage(
rows,
candidates,
before_seconds=args.before_seconds,
after_seconds=args.after_seconds,
)
matches = list(report.matches)
if args.apply:
if args.expected_matches is None:
raise RuntimeError("--expected-matches is required with --apply")
if len(matches) != args.expected_matches:
raise RuntimeError(
"exact match count changed: "
f"expected={args.expected_matches} actual={len(matches)}"
)
async with conn.transaction():
for match in matches:
status = await conn.execute(
"""
UPDATE app.turns
SET tokens_in = $2,
tokens_out = $3
WHERE id = $1::uuid
AND speaker = 'client'
AND llm_provider = 'claude_cli'
AND COALESCE(tokens_in, 0) <= 0
AND COALESCE(tokens_out, 0) <= 0
""",
match.turn_id,
match.tokens_in,
match.tokens_out,
)
if status != "UPDATE 1":
raise RuntimeError(
f"guarded update failed for one matched turn: {status}"
)
finally:
await db.close_pool()
deltas = [abs(match.delta_seconds) for match in matches]
return {
"schema": "vignette.claude_token_backfill.v1",
"mode": "apply" if args.apply else "dry_run",
"source_files": len(list(source_root.glob("*.jsonl"))),
"usage_candidates": len(candidates),
"eligible_zero_token_turns": len(rows),
"exact_matches": len(matches),
"unmatched_turns": report.unmatched_turns,
"ambiguous_turns": report.ambiguous_turns,
"tokens_in": sum(match.tokens_in for match in matches),
"tokens_out": sum(match.tokens_out for match in matches),
"max_abs_delta_seconds": round(max(deltas), 3) if deltas else None,
"applied": bool(args.apply),
}
def main() -> int:
args = parse_args()
result = asyncio.run(run(args))
print(json.dumps(result, ensure_ascii=False, indent=2, sort_keys=True))
return 0
if __name__ == "__main__":
raise SystemExit(main())