Brier and calibration buckets per league. Source: cricket_match_summaries
(winner) joined to cricket_ipl_pick_log (pre_toss_close model probability).
Settled matches only.
Per-tour calibration status
Each cricket tour is gated independently. A tour is calibrated
when either (a) holdout cohort N ≥ 30 graded matches
(directional Brier-vs-market evidence), or (b) forward picks N ≥
150 (live model-vs-outcome evidence).
✓ 1 tour
calibrated · 3 still building.
IPL
✓ calibrated (holdout)
Calibration evidence: holdout N=93, Brier 0.4944 vs DK 0.5011
Holdout cohort: N=93 · Brier 0.4944 vs DK 0.5011
(model better by 0.0067)
No buckets meet N≥5 threshold yet.
HUNDRED
⏳ building
Calibration evidence: forward N=0 of 150 + no holdout cohort yet
No buckets meet N≥5 threshold yet.
MLC
⏳ building
Calibration evidence: forward N=0 of 150 + no holdout cohort yet
No buckets meet N≥5 threshold yet.
VITALITY
⏳ building
Calibration evidence: forward N=0 of 150 + no holdout cohort yet
No buckets meet N≥5 threshold yet.
Excluded markets (match-fixing policy)
Per cricket-betting-expert framework + audit checklist: certain markets
carry historical match-fixing taint. We DO NOT publish picks or calibration
data on the following:
ITF / Challenger — Match-fixing integrity concerns. We do NOT publish picks or calibration on these tiers.
Cricket Tests (≥2026) — Through 2026 minimum: deferred until model designed for draws + DLS modeling + N≥50 sample.