Parlay Lab
Top 3-leg tickets ranked by correlation-adjusted EV. Fair probabilities are devigged from cross-book consensus; each leg is line-shopped to the best price.
Demo slate — realistic sample odds, not live lines. Live DK/FD/PrizePicks odds connect at season launch. Tickets mixing legs from one game are shown at naive multiplied prices; books reprice those as SGPs, so cross-game tickets are the actionable ones.
Line-distance gate active. At 3 legs we drop any projected leg priced beyond 1.02×its projection — that's where our variance model's tail error compounds past a 4% edge. Mode: pre-fit.
Childs %combines Shawn's projected means with SportsHack's variance. Families are Gamma (yardage) / Negative-Binomial (TDs), parity-locked to the Python reference; receptions use the compound targets×catch-rate model where a 2024 catch rate is known (NegBin fallback otherwise). CVs are still heuristic and target dispersion is a placeholder — directional, not calibrated. Gamma is calibration-validated out-of-sample (ECE 0.02, 63% better than Normal); the paid tier still waits on the full fit.
No 3-leg tickets clear the gate on this slate.
The line-distance gate (pre-fit) only admits projected legs within 1.02× their projection at 3legs. That's strict by design: the measured verdict is that beyond ~1.03× mean, the tail product isn't defensible at any leg count until CV is fit. This is the gate telling the truth, not a bug — try 2 legs, or wait for the fit.