Does production detection replace the pre-deploy gate?

No, it feeds it. A pre-deploy gate only sees a diff, so it can only check for regressions your own change caused; production detection watches live traffic and catches everything else, including causes with no diff behind them at all.

In Tessary the two connect directly: when a production classifier fires on a surface, that firing registers a pre-deploy check scoped to that surface, so the next change touching it gets tested against a failure production just found, not only against failure modes someone wrote down ahead of time. That’s a one-way handoff, not a substitute; the registered check still only runs against a diff, so it still can’t catch a regression with no diff to check. It’s also opt-in: pre-deploy checks are off by default, so a fresh project gets no automatic handoff between the two until someone turns it on.