changelog

Holder scoring rebuilt after an outcome audit

An outcome audit found the old holder score was inverted versus real rug outcomes — the tokens it rated “clean” went on to rug more often than the ones it flagged. Engine v2 replaces top-10-concentration scoring with fresh-wallet share and bundler wallet-tag reads, the inputs that actually tracked risk on our corpus. Missing wallet data is now reported as UNKNOWN rather than treated as “clean,” so a token no longer earns a passing holder read just because its wallets couldn’t be read. The rebuilt scorer was validated on a held-out set before shipping. Alongside the rebuild: a new descriptive TRAJECTORY axis shows a token’s live condition (observed drawdown, not prediction) next to the static structural score, and verdicts were relabeled to say what they measure — structural hygiene, not price prediction.