Historical bars seed

Pre-warms the historical_bars table so the first backtest is fast.

Run once after FLAG_HISTORICAL_BARS_ON_DEMAND is enabled on production.

When to use

After enabling FLAG_HISTORICAL_BARS_ON_DEMAND=1 on raxx-api-prod.
After provisioning a new production Postgres instance (table is empty).
Any time the historical_bars table is truncated or restored from backup.

Invocation

heroku run --app raxx-api-prod \
    python scripts/seed_historical_bars.py \
    --symbols SPY,QQQ,IWM,AAPL,MSFT \
    --lookback-days 730

Common variants:

# Dry run — prints what would be fetched, no DB writes
heroku run --app raxx-api-prod \
    python scripts/seed_historical_bars.py \
    --symbols SPY,QQQ --lookback-days 365 --dry-run

# 5-year seed for the full default symbol set
heroku run --app raxx-api-prod \
    python scripts/seed_historical_bars.py \
    --symbols SPY,QQQ,IWM,AAPL,MSFT,TSLA,NVDA \
    --lookback-days 1825

Expected output

Seed parameters:
  symbols:       SPY, QQQ, IWM, AAPL, MSFT
  date range:    2024-05-28 .. 2026-05-28
  lookback_days: 730
  mode:          paper
  dry_run:       False

  Seeding SPY (2024-05-28 .. 2026-05-28) ... 502 bars cached.
  Seeding QQQ (2024-05-28 .. 2026-05-28) ... 502 bars cached.
  ...

Done. 2510 real bars ensured in historical_bars table.

Already-cached bars are skipped (gap detection). Re-running is idempotent.

Notes

Requires ALPACA_PAPER_API_KEY / ALPACA_PAPER_API_SECRET (or live equivalents) set on the dyno.
The script exits with an error if credentials are missing rather than writing synthetic bars to the persistent DB.
On first backtest for any symbol not in the seed list, the runner fetches bars on demand and caches them automatically.