Best Practices

Three habits that separate a deployable strategy from a fitted curve.

Pulled from the patterns we keep seeing on live deployments that survived the first month vs. ones that didn't. Not theory — the recurring failure modes the agent has flagged on real backtests.

Don't optimize against the test set

If you tune parameters and validate on the same window, your backtest is a souvenir, not a forecast.

Most strategies that look brilliant on a single backtest die in the first week of live trading. Curve-fitting to a specific window is the single most common reason — the parameters discovered describe the past, not a tradable edge.

A simple split

Tune your parameters on the older 60% of available history. Hold out the most recent 30% to validate without touching it. Leave the last 10% as a forward-walk check after you deploy.

Red flag

If your validation window collapses and the tune window looks great, the strategy is overfit. No amount of "one more parameter tweak" recovers from this — start over with a different idea.

Best Practices

Deposit

Best Practices

Deposit

Best Practices

Three habits that separate a deployable strategy from a fitted curve.

Pulled from the patterns we keep seeing on live deployments that survived the first month vs. ones that didn't. Not theory — the recurring failure modes the agent has flagged on real backtests.

Don't optimize against the test set

If you tune parameters and validate on the same window, your backtest is a souvenir, not a forecast.

A simple split

Tune your parameters on the older 60% of available history. Hold out the most recent 30% to validate without touching it. Leave the last 10% as a forward-walk check after you deploy.

Red flag

If your validation window collapses and the tune window looks great, the strategy is overfit. No amount of "one more parameter tweak" recovers from this — start over with a different idea.