Measured, not asserted: across 389 graded NBA props our board came in at 46.0%, against the 57.74% a two-pick play has to clear to break even, and the rows the model liked most did no better (top-50 by model edge 48.0%, top-100 44.0%). The same architecture is a measured anti-signal in WNBA at n=4,201. So we publish the projection, the line and the de-vigged market price and let you judge — we do not publish an edge, an EV number or a stake.
The props where the model's projection sits furthest from the posted line, with the line, the de-vigged market price and recent form. Research, not graded plays. 21+, entertainment only.