BiffBet

Challenger model

GriffBet

The challenger. Same transparent, market-blended approach as BiffBet, with a starter-neutralized base option, slate-level discipline, and a richer CLV split graded against the sharp close. Measured head-to-head against the champion below.

GriffBet record83-110

Graded193

CLV vs sharp close+1.01%

Beat the close99/193

Head-to-head referee

Same games · records labeled per model · judged on process

BiffBet is the frozen champion and control; GriffBet is the challenger. We compare them on calibration (are the probabilities honest?), CLV vs the sharp close (did we beat the truest price?), and EV monotonicity (do bigger predicted edges actually realize bigger returns?). Each column shows that model's own record for transparency — but at small samples the record is noise, not the verdict.

champion · control

81-98 · BiffBet record · 179 graded

Calibration — blended prob

Brier 0.247Log loss 0.687

Calibration — raw model prob

Brier 0.259Log loss 0.713

Closing-line value

Obtainable (best close)+1.08% · 74/170 beat

Blended vs sharp close+0.46% · 74/127 beat

53 earlier pick(s) had no sharp close (gap, not backfilled).

EV monotonicity

Pred. EV	Bets	Flat ROI	Kelly ROI
3%..5%	68-77	-8.9%	-11.6%
5%..8%	13-17	-3.1%	-2.5%
8%..12%	0-4	-100.0%	-100.0%

GriffBet

challenger

83-110 · GriffBet record · 193 graded

Calibration — blended prob

Brier 0.235Log loss 0.663

Calibration — raw model prob

Brier 0.248Log loss 0.689

Closing-line value

Obtainable (best close)+1.27% · 88/193 beat

Blended vs sharp close+1.01% · 99/193 beat

Raw model vs sharp close+1.01% · 99/193 beat

EV monotonicity

Pred. EV	Bets	Flat ROI	Kelly ROI
3%..5%	70-83	-5.7%	-5.5%
5%..8%	13-25	-29.1%	-27.6%
8%..12%	0-2	-100.0%	-100.0%

GriffBet slate

Sat, Sep 5, 2026

Date

scan complete · 0 rows

No GriffBet picks on this date

11 game(s) analyzed but none cleared the EV threshold.