How this is computed. Every claim is stored the moment it is published, with the stage and fee it asserted. When a club officially announces the deal — or the window shuts on it — every source that touched it is scored: was the move they described the one that happened, did they claim a stage the deal never reached, how far off was their fee, and how early were they relative to everyone else on that deal.

Ranked score is the Wilson lower bound rather than raw accuracy, so a source with three lucky calls cannot outrank one with a hundred and eighty at 85%. Sources with fewer than ten resolved claims are shown as unrated. This dataset only exists because it is recorded continuously — it cannot be bought or backfilled.

Fabrizio Romano is wired the same way as the rest of the journalist wall: we do not use the bulk X API, we ingest an RSS bridge URL for the journalist's public posts and keep linking out to the original post. Once a bridge URL is set, run npm run seed:journalistsand the feed becomes a pollable source. Romano's wider publishing footprint is also covered through GiveMeSport.