Audit: the deploy gate — ROBUST or MIRAGE

The audit freezes one parameter set (the search's training winner) and grades it on two separate out-of-sample periods it never touched. The verdict takes the worst of the two windows — no cherry-picking. Verdicts are blunt:

• ROBUST — made money in BOTH unseen periods with consistent months. Earned the right to be deployed.
• MIRAGE — the profit was an illusion of the optimizer; it evaporated out of sample.

Most strategies fail here. A failure is the tool doing its job — catching a loser before it costs real money.

Check the road, not just the destination

The equity curve shows the audited candidate over the full range; the underwater chart shows every decline from a peak — how deep the pain got and how long it lasted.

The printable audit report

From the run page, Open the full report → Save as PDF (or Print). Both offer to append the strategy's full parameter settings on their own page. What was tested, on what data, with what settings, and the verdict — a paper trail for a trading partner, a prop firm, or future you.

Reading a candidate's trade analysis

The trade-analysis page of an audit candidate is the full backtest of that setting over the whole range: the training period and both test periods pooled. Pooled numbers can look fine while one test period failed, because the training period and the good test carry it. So the page shows the audit's periods side by side under the verdict, each test graded by the audit's own gates, labels the pooled tiles as "whole range", and shades the equity chart by period. The verdict is the worse of the two test periods; read those columns, not the pooled tiles.

A thin verdict

"thin_test" means one of the test periods had fewer trades than the gate, so the audit did not judge the setting. It is not a pass. The page now says what the short test did with the trades it had, and whether that shape would pass with more of them, and it works out the range that would give each test window enough trades at the setting's own pace, or the Min test trades value that would give a verdict on the data you have. Perturb and Certify are not offered from a thin verdict: a perturbation would be as thin, and certification needs a ROBUST perturbation. What is offered is "Park in incubation (unjudged)": one Robustness look at the setting, after which it is parked on the Deployment page's "parked without a verdict" shelf, apart from settings that came through the chain. Its checks on new data become the test the audit lacked.