← Back to the board Trust hub: Aggregate NBA NHL MLB Soccer Tennis Cricket UFC Model CLV Calibration

Trust · Tennis

Brier and calibration buckets for the tennis pipeline. Pinnacle pre-match no-vig P(player_a) is the anchor; outcome from tennis_in_play_match_meta (only outcome_type='completed' included — retirements/walkovers excluded from headline per finalizer convention).

Rolling Brier: 0.1921  |  Uniform baseline: 0.2500
N settled = 1415

Live coverage — same-day withdrawal gate

Tour-asymmetric by design. Disclosed honestly so users can size accordingly.

Tennis live-coverage withdrawal-gate status by tour.
TourSourceStatus Last cycleWithdrawals last 7d
WTA api.wtatennis.com (official Pulselive) healthy 4 tournaments, 364 rows 0
ATP api.atptour.com blocked Cloudflare managed challenge — discovered 2026-05-31 (Playwright network capture)
ATP coverage gap. ATP picks have NO live same-day withdrawal coverage. Chronic injury history (retirement_risk.py) still applies. Apply tighter sizing on ATP favorites during Slam first weeks.

Policy: CLAUDE.md jurisdiction rule + 2026-05-17 FBref/ufcstats memory — no bypass. Resolution path: Paid data source (Sportradar tennis ~$30-50k/yr) or direct WTA-style relationship.

Acceptance gate

✓ N=1415 ≥ N=40 per-bucket target for the bucketed Brier check.

Calibration buckets (5pp bands)

Tennis calibration by predicted-probability bucket.
BucketN Predicted meanActual rateGap (pp) Tripwire
0.05 – 0.10 32 0.076 0.125 +4.9pp ok
0.10 – 0.15 51 0.128 0.039 -8.8pp watch
0.15 – 0.20 55 0.175 0.164 -1.1pp ok
0.20 – 0.25 64 0.226 0.219 -0.7pp ok
0.25 – 0.30 80 0.272 0.250 -2.2pp ok
0.30 – 0.35 97 0.325 0.299 -2.6pp ok
0.35 – 0.40 90 0.376 0.367 -0.9pp ok
0.40 – 0.45 116 0.425 0.371 -5.4pp watch
0.45 – 0.50 88 0.475 0.409 -6.6pp watch
0.50 – 0.55 101 0.527 0.505 -2.3pp ok
0.55 – 0.60 115 0.574 0.487 -8.7pp watch
0.60 – 0.65 106 0.623 0.632 +0.9pp ok
0.65 – 0.70 84 0.674 0.643 -3.1pp ok
0.70 – 0.75 87 0.725 0.713 -1.2pp ok
0.75 – 0.80 81 0.775 0.778 +0.3pp ok
0.80 – 0.85 63 0.827 0.905 +7.8pp watch
0.85 – 0.90 57 0.869 0.895 +2.5pp ok
0.90 – 0.95 34 0.928 1.000 +7.2pp watch
0.95 – 1.00 10 0.969 1.000 +3.1pp ok

Buckets shown only when N ≥ 5. Tripwire (tennis-expert 2026-05-24 lock): green <5pp | yellow 5-10pp or thin-N | red ≥10pp at N≥30 (action: hand to win-rate-optimizer).

ATP / WTA tour-parity audit

Tennis ATP/WTA tour-parity coverage audit.
Tour N total N settled N with anchor Settlement % Anchor %
ATP 884 667 876 75.5% 99.1%
WTA 954 755 947 79.1% 99.3%
Parity gap (ATP − WTA) -70 -3.7pp -0.2pp

Per audit competitive-advantage #9: WTA parity is a user-acquisition advantage. ATP-side bias greater than 5pp on settlement or anchor rates flagged in red.

Recent settled matches

Recent settled tennis matches with model probability and result.
MatchTourSurface P(A)WinnerHit?
Alexander Zverev vs Botic van de Zandschulp atp hard 0.863 a hit
Qinwen Zheng vs Elena Rybakina wta hard 0.265 b hit
Ben Shelton vs Carlos Alcaraz atp hard 0.231 a miss
Jessica Pegula vs Emma Navarro wta hard 0.777 a hit
Frances Tiafoe vs Alex Michelsen atp hard 0.576 a hit
Aryna Sabalenka vs Linda Noskova wta hard 0.703 a hit
Alexander Zverev vs Luciano Darderi atp hard 0.875 a hit
Iva Jović vs Coco Gauff wta hard 0.268 b hit
Naomi Osaka vs Elena Rybakina wta hard 0.438 b hit
Francisco Cerundolo vs Alexander Blockx atp hard 0.403 b hit
Iga Swiatek vs Qinwen Zheng wta hard 0.804 b miss
Mirra Andreeva vs Anastasia Potapova wta hard 0.823 a hit
Anna Kalinskaya vs Emma Navarro wta hard 0.516 b miss
Ben Shelton vs Stefanos Tsitsipas atp hard 0.723 a hit
Jessica Pegula vs Sorana Cirstea wta hard 0.709 a hit
Aryna Sabalenka vs Taylor Townsend wta hard 0.814 a hit
Alexander Zverev vs Alejandro Tabilo atp hard 0.819 a hit
Jakub Mensik vs Learner Tien atp hard 0.436 b hit
Naomi Osaka vs Elise Mertens wta hard 0.731 a hit
Yuliia Starodubtseva vs Elena Rybakina wta hard 0.131 b hit

Open upcoming matches with anchor

No upcoming matches with anchors yet.

Variance disclosure: tennis is best-of-3 single-match noisy. Brier reads meaningful at N≥200 across surface×tour-level segments per acceptance framework. Pre-RG numbers are preliminary.