Backtesting
What a Lab backtest is, how to run one, the queue and per-plan caps, how to read the report — the verdict, the fair baseline, the fill model and its measured optimism — and the final test.
What this is for. A backtest replays a strategy over past data at prices a follower could have paid, and sets the result beside a fair baseline, so you can see whether the idea did better than doing the obvious thing. It answers "what would this have done", not "what will it do".
Simulated performance
Simulated. Replayed on past trades. Not a forecast or advice. Backtests replay the past at prices a follower could have paid; paper trading scores live signals on real order books without placing orders. Neither is a forecast or advice, and both carry a measured optimism.
These results are based on simulated or hypothetical performance results that have certain inherent limitations. Unlike the results shown in an actual performance record, these results do not represent actual trading. Also, because these trades have not actually been executed, these results may have under- or over-compensated for the impact, if any, of certain market factors, such as lack of liquidity. Simulated or hypothetical trading programs in general are also subject to the fact that they are designed with the benefit of hindsight. No representation is being made that any account will or is likely to achieve profits or losses similar to these being shown.
Run one
Open the strategy (its row under Strategies in the rail).
Under Test it on past trades, pick 30 days, 60 days or 90 days (the default), or Dates… with a From and To. A window can be at most 366 days.
Press Backtest — or Save and backtest if you have unsaved edits. The assistant can also start one for you in a chat.
The queue
Backtests run on a dedicated runner, with a priority per plan (standard, priority or highest — see the plan table) and fairness between users. While yours waits or runs, the page says so in words:
- N ahead of you · ~4 min while it queues, Next in line when it is next;
- Running · ~2 min left while it runs (Running · taking longer than usual past the estimate);
- A backtest usually takes a few seconds once a runner has it; The backtest failed. if it does.
Two per-plan caps apply (plan table):
| Cap | When you reach it |
|---|---|
| Backtests a day (UTC) | "Your plan runs N backtests a day; the count resets at midnight UTC." — API 429 lab_daily_limit, Retry-After = seconds to midnight UTC |
| Backtests queued or running at once | "N backtests are already running; wait for one to finish." — API 429 lab_busy |
By API, GET /v1/lab/backtests/{id} carries queue_position (runs ahead of it, 0 = next) and eta_s
while a run waits. The queue wait of your recent backtests is on the Usage tab of
Usage & billing.
Read the report
A copy strategy's report reads top to bottom:
- The verdict, in words — Weak evidence, Promising, not proven or Strong evidence, with Why …? beside it. Strong needs at least 100 copies and a t-statistic of at least 2; Promising at least 30 copies and t ≥ 1. A sparkline compares This strategy with copying everyone.
- The big numbers — Per $10 copied (the result per $10 put into copies, so strategies of any size compare), Copies, Bankroll at the end, then the money in play.
- The simulated-results notice, with the run's own Fill model sentence and its measured optimism — how much better than real fills this model has measured to be.
- How this was computed — the fair baselines (Copy everyone in the same markets under the same rules, Without the best wallet, Without the best 3 days), a per-period table, the bankroll chart, room and costs, the funnel (Signals → Within the rules → Copied), Per wallet, and What a backtest cannot know.
From the report, Paper-trade this starts a paper run; Change the rules (on a weak verdict) or Open the strategy goes back to the settings.
How to read it honestly
A result that disappears without the best wallet or without the best 3 days rests on one wallet or one lucky stretch. A result that barely beats Copy everyone says the markets did the work, not the selection. When many variants were tried, the best one looks good by chance — the assistant's searches report that explicitly (searches).
Own-rule backtests and the final test
A strategy on your own rules is replayed on the recorded order books. Its report shows Per $10 staked, Entries and Bankroll at the end; compares against The baseline: same price, random moment and The markets you chose; and checks robustness with a t-statistic clustered by event, Without the best market, Without the best 3 days and the number of rule sets tried.
The report covers only the first three quarters of the window; the last quarter stays sealed. Run the final test? reads the strategy on that unseen quarter — once per strategy, so tune first and seal your choice before you look. Afterwards the button reads Open the final test.
Related
- Strategies — every setting a backtest runs with
- Paper trading — the next step after a backtest
- Research and data — what history a backtest can use
- Plans & billing — backtests a day, at once, and queue priority per plan
Strategies
What a Lab strategy is, how to create one, every setting of the strategy editor, the Draft · Tested · Paper · Live stages, and saving, archiving and reopening from the Strategies rail.
Paper trading
How a Lab paper run scores a strategy on live order books without placing orders — starting and stopping, what a copy is, the 50-copy gate, resting orders, notifications and per-plan limits.