Validated algos — the ones that show backtest results — passed a pre-registered gauntlet on up to 10 years of historical data: out-of-sample windows, walk-forward tests, regime slices, and a one-shot holdout. A small set of experimental algos is listed with a clear NOT BACKTESTED label and no performance history. Once tracked, an algo's forward record is written daily against real prices from the date tracking began (the "since" date on its listing, engine v2) and is never revised. The full process is public on the
methodology page. Past performance does not guarantee future results.