A run executes here, against data the platform already holds — nothing you submit leaves. Forward testing is the step after: it fixes a strategy to a date, so it can be scored on bars that did not exist when you committed to it.
The gap it closes
Every check on the scorecard evaluates the single run it was handed. None of them can see a parameter search you ran beforehand — and because market data is public, that search can happen entirely off this platform. The platform measured it: a search over roughly 3,000 parameter combinations against pure noise, submitting only the best-looking one, clears all eight checks.
Scoring a strategy on data that arrived after you committed is the one defense that search cannot reach, because there is nothing to optimize in advance.
Registering
On a strategy’s Forward tab, registering fixes the current spec — by hash — to the moment you press the button. From then on, only trades signalled after that momentcount toward the strategy’s forward record. Accrual runs as a scheduled sweep on the server, so the numbers update on their own, not when you open the page.
No verdict, on purpose
The Forward tab shows days live, forward bars, and forward trades — and no pass/fail. There is not yet a threshold for how much forward evidence is enough, and picking one before there is real accrued data to look at would be a guess dressed as a standard. It records what has happened, not what it means.
Withdrawal is final
- Withdrawing stops accrual. The registration stays on record with whatever it had — a record you could delete would let you keep only the runs that went well.
- A withdrawn spec cannot be registered again. Resuming it would let a strategy skip the window it was withdrawn during, which is the cherry-picking this mechanism exists to prevent.
- To commit the idea again, refine it. A changed spec is a different hypothesis: new hash, new clock, starting from when you commit it.