What does the ML gate actually do to my strategy?

It filters rather than replaces. Your strategy still decides when a setup exists; the gate then judges whether that particular instance looks like the ones that historically worked out, and can veto the entry. So your logic is upstream and unchanged — turning the gate off returns you to exactly your own rules. Two things worth knowing before enabling it: it will reduce your trade count, sometimes substantially, and a filter that removes trades is only an improvement if it removes more bad than good, which is worth verifying on your own strategy rather than assuming.

"Verify rather than assume" applied to the product's own feature is a good sign, honestly.

Worth adding: a filter doesn't just remove trades, it reshapes the distribution of the ones you keep, usually fewer entries, and the ones that survive are drawn from a narrower slice of conditions. So the honest comparison isn't gated equity curve vs ungated equity curve, it's the same signal list with a flag on which ones got vetoed, then look at the two groups side by side. Also keep in mind the smaller sample: if your strategy fires rarely to begin with, the veto set may be too thin to say much with any confidence. Do you have enough signal history to split it that way yet?