Forward testing falsifies; it does not confirm
FORWARD -> APPROVED needs a stopping rule, or it becomes optional stopping: check continuously, approve the first time the result looks acceptable, and approvals land on lucky streaks. The obvious remedy — run until the power plan's n — turns out to cost one to two years per Strategy. At 0.5 trades per day against a 0.25R effect with roughly 1R trade-level SD, an independent confirmation needs about 125 trades at alpha 0.05 and about 265 once the search correction is applied.
The forward stage is therefore a pre-declared falsification window with a rejection boundary, evaluated once at the end. Passing means no implementation, cost or obtainability defect was detected. It is explicitly not confirmation of edge, and the Register records it in those words.
Why this is the honest framing
Both defects the donor programme caught at this stage were implementation failures, not edge questions: Z-ENTRY-IMPLEMENTABLE found the fade's +0.1R did not exist because the engine booked unobtainable entries, and AMENDMENT-4a found the entry was a limit rather than a stop, at the cost of two of the campaign's first three setups. That is what forward testing is good at, and claiming more for it would be the same overreach the reference architecture makes when it draws the stage as a loop with no stated duration.
Considered options
- Powered confirmation. Statistically the strongest option and it makes forward capacity the binding constraint on the entire programme — perhaps two strategies ever reach live.
- Group-sequential design with an alpha-spending boundary. Valid early stopping, so good strategies reach live sooner and bad ones die faster. Deferred rather than rejected: more machinery, and the boundaries must be declared before the first look. Worth revisiting if forward capacity becomes the bottleneck.
Consequences
A Strategy goes live with its edge never independently confirmed. That must be stated plainly wherever approval is recorded, not glossed — the whole value of this stage is that nobody mistakes "nothing broke" for "it works".