User picks vs the model
Every user with at least 10 settled picks in the last 12 months, ranked by ROI. Identities are anonymised — handles are one-way hashes of the user id, so no PII leaks. The Edgewise Model row is the auto-logged baseline anyone can audit at /audit/public.
Methodology. Each pick is staked 1u flat. A win returns (price − 1)u, a loss returns −1u, a void returns 0u. ROI = total profit / total settled stake. Brier = mean squared error of model probability vs binary outcome (lower is better). CLV proxy = average gap between model probability and the median-book implied probability at log time.
Eligibility. Minimum 10 settled (non-void) picks in the last 12 months. Users with fewer picks are hidden until they qualify.
Privacy. Handles are user_ + the first 8 hex chars of a SHA-256 hash of the user id. No display names, emails, or other identifiers are exposed.