Backtest

Run a DSL pack against a candle series and emit a receipt. No orders are sent. No network is touched.

CLI

krellbot backtest <pack.json> --venue kraken|coinbase [--data csv] [--json] \
  [--slippage-mult N] [--fee-bps N] [--slippage-bps N] \
  [--from YYYY-MM-DD] [--to YYYY-MM-DD] [--allow-gaps]

--data is a CSV with the cache format: header ts_ms,open,high,low,close,volume, then one row per bar. The from / to flags trim the series by date in UTC.

Defaults when not overridden: Kraken fee_bps=40, Coinbase fee_bps=120, slippage_bps=5, slippage_mult=1.0.

Receipt

--json prints a single JSON object on stdout. Keys exactly: engine_version, pack_sha256, data_manifest_sha256, venue, pair, tf, from, to, fee_bps, slippage_bps, slippage_mult, metrics, equity_curve. The metrics block carries total_return_pct, cagr_pct, max_drawdown_pct, return_to_dd, per_year, trade_count, exposure_pct, buy_and_hold. The equity_curve is at most 64 numbers; its last point equals metrics.total_return_pct within 0.01.

Fill rules

Signal on the close of bar t fills at the open of bar t+1.
Stop, while long, exits at min(stop, open[t+1]) and applies sell slippage. The signal fill of the same bar is skipped.
A flat-price, zero-cost round trip changes equity by exactly 0.
cash stays >= 0 on every bar; equity == cash + qty * close.

Cache

krellbot data import kraken-ohlcvt <zip> --pair PAIR --timeframe TF streams the zip and writes one CSV plus a JSON manifest under $KRELLBOT_HOME/cache/. The manifest's sha256 is the hex digest of the CSV bytes; the backtest records that hash in the receipt.

Gap policy

Series missing more than 1% of expected bars is refused unless --allow-gaps is passed. The error names the pair and the missing count.

← docs