Model accuracy & methodology
Live model evidence · Official Picks only
74% accuracy · n = 23 settled
26 published · 23 settled · 3 awaiting · 0 void
More results are needed before performance differences are considered reliable.
First published prediction · 8 Oct 2026
Filtered record, not all-time: earlier entries without captured pre-match odds were excluded on 11 October 2026. How results are counted.
MODEL EVALUATION
Model health · Collecting evidence
Model changes: No structured review recommended yet.
Analyse → Propose → Validate → Backtest → Version → Deploy deliberately. This page never changes model weights.
MAIN DRAW VS QUALIFYING
Subgroup differences are descriptive, not evidence that one category performs better.
PERFORMANCE BY CONFIDENCE
Subgroup differences are descriptive, not evidence that one category performs better.
SURFACE PERFORMANCE
All currently settled predictions are on hard courts.
Subgroup differences are descriptive, not evidence that one category performs better.
TOURNAMENT TIER
Subgroup differences are descriptive, not evidence that one category performs better.
ROLLING PERFORMANCE
Settled-prediction windows, not sparse daily percentages.
All Matches includes published main-draw and qualifying predictions; unverified historical draws remain separate. Withheld assessments are excluded from accuracy and ROI. Qualifying data checks do not establish calibrated confidence. Original published snapshots retain the policy and score recorded at publication.
How the model and record work
TennisDollar generates statistical predictions using separate main-tour models for ATP and WTA. These are automated model outputs, not a claim that an analyst has personally reviewed every match.
Data sources and coverage
Fixtures, historical match results, available serve/return statistics and result updates come from API Tennis. Bookmaker odds are third-party prices; each available odds badge identifies its source and capture time. Prices are separate from model confidence.
The analysis uses available eligible main-tour singles history from the preceding two years. Qualifying, lower-tour and other ineligible matches are excluded. Missing statistics, small surface samples, player inactivity and uncertain inputs can limit an assessment; the model does not invent missing observations.
What supports a prediction?
The model compares recent form, surface results and serve/return evidence. Recent results receive more weight, while sparse surface or point-stat samples reduce their influence. The saved analysis shows the values and model version associated with the original prediction.
For example, a player can have stronger overall results but weaker results on the match surface. Those signals can disagree. A single favourable statistic does not prove that player will win, and a confidence score of 7/10 is not a measured 70% chance.
Which results are included?
The live record counts published Official Picks rated 6.0 or above, captured before play. Leans, retrospective backtests, unpublished outputs and unresolved matches are excluded from win-rate calculations. Voids count as neither wins nor losses. Accuracy = wins ÷ (wins + losses).
On 11 October 2026, earlier records without captured pre-match odds were excluded from the displayed history. The cutoff for that one-time cleanup was 10 October 2026 at 23:48:09 UTC. Later publications are not removed by that cleanup. This is a filtered record, not complete all-time performance. Missing odds alone do not invalidate a match outcome; removing those entries changes the sample and may change its accuracy.
The homepage combines eligible ATP and WTA results; this performance page applies your selected tour, period and model version. Small samples can change sharply. The publication and settlement record is maintained by TennisDollar, not independently audited.
Publication, settlement and corrections
Original predictions, confidence and captured odds remain frozen. Results are matched to verified provider match outcomes under the existing settlement rules. Later results are not used to rewrite an earlier prediction.
If you find an error, send the prediction URL and the detail you believe is wrong through Contact & support. We do not present this record as evidence of guaranteed future accuracy or profit.
About TennisDollar and its publication standards · Inspect past predictions and results · Separate historical testing
Play-style live performance · all draws
Play-style leaderboard
Which styles win their matches? ATP results for the selected period.
How style results are counted
Each player has at most one qualifying style, based on pre-match records. All-rounder and Pressure Performer take priority when their combined requirements are met, followed by Serve Dominant, Elite Returner, Serve Bot, Break-point Defender and Clinical Breaker. Verified Southpaw is used only when no statistical style qualifies. Otherwise, no badge is shown.
Only verified completed matches count, including leans and “Too close to call” matches. Retirements, walkovers and backtests are excluded. Each match contributes to one pairing at most. Same-style pairings and players without a qualifying badge are excluded. Earlier multi-badge records use this same priority to select one of their saved pre-match labels; the original records are preserved. Records without saved labels are excluded.
The leading style appears first; tied records are alphabetical. These are observed results, not proof that a style causes wins. Players, opponents and surfaces also differ. Ten matches is only a display threshold, not statistical validation. Badges and leaderboard results do not change model confidence.