Problem: CLV workflows need a durable, queryable history of book prices — not just the latest API pull.

Solution: close-archive writes normalized odds snapshots to SQLite (one row per quote). Append JSONL from cron or your line-shop-cli adapter. beat-close compares, per book, the latest quote at open vs close cutoffs and marks whether your open price was no worse than the close (American-odds ordering: −105 beats −110, +150 beats +140).

{"ts_utc":"2026-03-01T19:30:00Z","book":"draftkings","market_type":"spreads","event_key":"nba:20260301:LAL@BOS","participant_key":"LAL","american_odds":-108,"line":7.5}

Optional: implied_prob, any extra fields are preserved in raw_json.

pip install -e .
close-archive init odds.sqlite
close-archive append odds.sqlite --file snapshots.jsonl
# or:  your_generator | close-archive append odds.sqlite

close-archive export-csv odds.sqlite > dump.csv

close-archive beat-close odds.sqlite \
  --event "nba:20260301:LAL@BOS" \
  --participant LAL \
  --open-before "2026-03-01T18:00:00Z" \
  --close-before "2026-03-01T23:00:00Z"

Use ISO8601 UTC strings for timestamps (lexicographic compare matches time order).

• Emit JSONL from line-shop-cli with a thin mapper.
• Feed eval summaries into nba-clv-dashboard / backtest-report-gen.

• No live book API — ingest only.
• Parquet export — optional extra (pip install '.[parquet]') not implemented in v0.1; use export-csv or DuckDB on SQLite.

MIT