Reading the /backtest charts

The big table at the top breaks the backtest into cells of (coin × year). Each cell shows the gross P&L for that coin in that year, with long / short split where relevant.

Greener cells are positive years; red cells negative. The strategy's stability hinges on having few deeply red cells — a strategy that gets +200% one year and -80% the next isn't really making money, it's just gambling.

The line chart below the grid shows equity through time. The red shaded regions are drawdown windows — periods when equity was below its all-time high.

What matters is the max drawdown depth (worst point) and the recovery time (how long until the new high). v6.4 and v7 both stay under 50% max drawdown in backtests; recovery times range from weeks to months.

The backtest covers 2020 through April 2026. Strategy parameters were tuned on a subset (in-sample); the remainder is the holdout (OOS). The OOS portion is what really matters — anyone can fit a strategy to in-sample data, the question is whether it generalises.

On both v6.4 and v7, OOS performance is degraded vs in-sample (expected) but still profitable on year-weighted score. The /methodology page explains the splits in detail.

The backtest uses dYdX maker fees on entries/TPs and taker fees on SLs — these are baked in, so the numbers are post-fee. It does NOT model slippage from your own orders impacting the orderbook (the bot trades in small notional sizes where this is negligible).

Funding rates are modeled approximately. The backtest does not include any 'live' costs like server hosting or Botely subscription fees.