Trust hub: Aggregate NBA NHL MLB Soccer Tennis Cricket UFC Model CLV Calibration

NBA Trust Dashboard

Current NBA pricing status: display-only

Display-only is the safe default. EV-live activates only after the calibration gates pass. Operator flips NBA_EV_LIVE=1 manually after audit clears.

Playoff Cohort (window opens 2026-04-12)
Picks: 209 · Brier: 0.2283 vs market 0.2246 (lift +0.0037; model worse )

Sleeper injury watcher — HEALTHY
Poll cadence: 30s · Last row count: 65 · Last fetch: 0s ago

Blend v2 cohort split — deploy date 2026-05-25
v1 (Sq 0.45 / Sh 0.30 / PM 0.25): N=528 · Brier 0.2309 vs mkt 0.1749 (lift +0.0561) v2 (Sq 0.80 / Sh 0.20): N=17 · Brier 0.2805 vs mkt 0.2814 (lift -0.0008)

Overall Calibration (all settled NBA picks)

N: 545 · Model Brier: 0.2325 · Market Brier: 0.1782 · Brier lift: +0.0543 (model worse)

Calibration by probability bucket (5pp width)

NBA calibration by predicted-probability bucket.
Bucket	N	Predicted mean	Actual rate	Gap (pp)	Tripwire
0.50–0.55	71	0.525	0.648	+12.32	red
0.55–0.60	54	0.574	0.537	-3.72	green
0.60–0.65	69	0.626	0.522	-10.46	red
0.65–0.70	183	0.676	0.612	-6.41	yellow
0.70–0.75	157	0.720	0.713	-0.64	green

Tripwire thresholds (tennis-pattern 2026-05-24 lock):

green — |gap| < 5.0pp (in calibration)
yellow — 5.0pp ≤ |gap| < 10.0pp (watch)
red — |gap| ≥ 10.0pp AND N ≥ 30 (action required)
yellow_thin_n — large gap but N < 30 (more samples needed)

Ship-gate criteria for NBA_EV_LIVE promotion (all must hold):

N ≥ 200 settled NBA picks
Model Brier ≤ market Brier + 0.005 (no negative Brier lift beyond noise)
Zero red buckets across the 0.50–0.95 probability range

Calibration data is sourced from predictions JOIN game_outcomes on (game_id, sport='nba'). Market Brier is computed from square_prob (no-vig market anchor) on the same picks. Model Brier uses consensus_prob.