Implementation hand-off

A winner that never ships is worth nothing. The implementation queue is the list of experiments that produced something worth building, with the variant, the change and the result attached.

What makes it useful is what it refuses to list. An earlier version treated every finished experiment with no rollout as work waiting to be done, which meant losing variants and flat tests sat in the queue as if somebody had decided to build them. Entries now have to clear the same gates the rest of the console applies: a direction, enough traffic per arm, and a lift interval that stays on one side of zero.

It reads both goals rather than one. Conversions are consulted first, because that is the primary objective and its maths carries no approximation, and revenue second, so an experiment that won on revenue alone still reaches the queue.

Each entry tracks its own state through to hard-coded, and a separate signal flags winners that were deployed and are still being served by the platform months later. That is the case worth catching: it works, but the change belongs in your codebase.