Accuracy — verdicts, scored against reality

No security API on any chain publishes its own hit rate. This one does. Every verdict served is enrolled as a tracked observation; a daily job re-reads the chain and records what actually happened. This page is that record, live and un-edited.

Methodology

• Retrospective cohort — tokens observed at launch by the production sniper pipeline (launch-time safety signals recorded months ago, before this site existed). We check what the chain says about them now. Because the signals predate the outcomes, this is a genuine natural experiment.
• Prospective cohort — every verdict this service serves is enrolled at serve time and re-checked daily from age 7 days. This cohort is young; it matures on its own schedule and cannot be rushed.
• Outcome classes — alive (pool healthy), drained (<10% of baseline liquidity or under 1 XRP), dead_pool (AMM gone entirely), rugged_flags (issuer global-froze the token), unknown (chain unreadable — always excluded from rates, never guessed).

Honest limitations

• The retrospective cohort contains almost exclusively launches that failed launch-time screening (the sniper's gate was strict). Until the prospective cohort matures there is no passed-screen comparison group, so we publish the mortality base rate — not a discrimination claim.
• Survival ≠ legitimacy. A pool can stay alive while a token still fails socially or economically. These are liquidity-level outcomes only.
• Small cohorts move in big steps. Percentages here come with their raw counts, always.