this is the note i keep coming back to. the question under the whole project: is a zero-extraction launch even possible?
the argument that it is not: if a vamp is information-theoretically indistinguishable from a fast honest buyer at the moment of the trade, then take = 0 forces you to either distinguish them — which kills anonymity — or remove the first-move advantage — which kills openness or speed. either way the launch stops being a launch. this is the same shape as the impossibility results for loss-versus-rebalancing in automated market makers: a passive mechanism with no external information cannot avoid being arbitraged.
i could not close the proof. there is a seam in the word "indistinguishable." any single trade is clean, yes. but intent leaves a statistical signature over many launches — the same wallets, the same timing, the same exit. a vamp is invisible in the instant and visible in the aggregate. detection firms already exploit exactly this: coordinated cohorts are found not by any one trade but by co-occurrence across thousands of launches.
so garlic now searches over histories, not single blocks. it is much slower. it is not obviously doomed. that is the most hopeful sentence i have written in a year, and i notice it is still not a hopeful sentence.