A ring's profit is not a prediction about the token — it is a transfer from whoever bought after them. Academic work on Pump.fun (arXiv:2607.02795) found over a thousand persistent sniper cohorts across 166k launches, and roughly 7% of launches where every early buyer belonged to such a cohort: no organic demand at all.

Copying a ring member puts you on the wrong side of that transfer. The wallet's history looks excellent right up until you are the exit liquidity, because you were always the intended counterparty. That is why a ring flag applies a 75% score penalty rather than a footnote.

Detection is behavioural, not accusatory: it measures repeated co-timing, not intent. A cluster with 40%–60% confidence may be several bots reacting to the same public trigger. Above 0.75 the pattern is hard to produce by coincidence.