The scores get tested in public

Pre-registered: rolling-origin validation

The question under test: do wallets our method scored highly at time T stay accurate on events that resolve after T? The design refits the full scoring pipeline as of several historical cutoff dates, freezes each cutoff’s rankings, and measures forward accuracy only on events those rankings could not have seen. The design was committed on 2026-07-10, before any cutoff ran.

• Publication rule: results publish here whichever way they come out. That decision was made before the first result, and it is binding.
• Robustness rider kept:a re-run keyed to each market’s on-chain resolution timestamp, guarding against lookahead in event dating.
• Rider closed 2026-07-11, before validation began:a measured check of whether excluded market-maker and bot wallets tilt the estimation population found the effect immaterial (at most 0.01¢ on any price-baseline bucket and 0.03¢ on any category prior). The planned estimation-population rider is therefore dropped, and that decision is recorded here in advance.
• Schedule: first cutoffs run the week of 2026-07-14; the full grid and a dated report follow here.

Already public

• The methodology: price-relative scoring, empirical-Bayes shrinkage, independence rules, and a limitations section that states what the scores cannot yet claim.
• Per-wallet calibration diagrams on every wallet page, computed from the same scored trades the rankings use.
• The receipts ledger: every featured alert, recorded at detection and graded on resolution.