How it works · The method and the full record
Every game, scored.
Every call, graded.
PrizmBet breaks each game into the factors that decide it, turns them into one confidence number, puts the market’s number next to ours, and grades the call after the final. The misses stay on the record.
From data to a graded call
PLAY, LEAN, SKIP — and how each tier has done
The confidence number puts every game in one of three tiers. If the number means anything, higher tiers should win more often. Since launch, they do:
67.8% · 233 games
Since launch = Sept 1 – Oct 26, 2026, all sports. Pre-launch = Mar 1 – Aug 31, 2026, when the model was graded in testing before the app reached the App Store. SKIP means the factors don’t clearly favor either side; most soccer matches land there, and soccer draws count as misses.
NFL 2026 regular season
| Week | Record | Won | PLAY 65%+ |
|---|---|---|---|
| Week 1 | 11–5 | 68.8% | 4–3 |
| Week 2 | 8–8 | 50.0% | 6–3 |
| Week 3 | 9–7 | 56.3% | 4–2 |
| Week 4 | 11–5 | 68.8% | 8–2 |
| Week 5 | 9–5 | 64.3% | 6–2 |
| Week 6 | 8–6 | 57.1% | 5–3 |
| Week 7 | 8–5 | 61.5% | 5–4 |
| Season | 64–41 | 61.0% | 38–19 |
Through Week 7, as of October 26, 2026. Monday night’s game is added once it’s final. One tie (Week 5) is a push and not counted. Updated every Monday.
Since launch · Sept 1 – Oct 26, 2026
| Sport | Record | Won |
|---|---|---|
| NFL2026 regular season | 64–41 | 61.0% |
| College footballsince Oct 20, 54 games · early | 34–20 | 63.0% |
| NBA2026–27, 44 games · early | 29–15 | 65.9% |
| MLBSept 1 – Oct 24 | 219–178 | 55.2% |
| MLSSept – Oct | 61–42–34W–D–L | 44.5%64.2% with a winner |
| La Liga56 games · early | 30–16–10W–D–L | 53.6%75.0% with a winner |
| Premier League43 games · early | 23–11–9W–D–L | 53.5%71.9% with a winner |
| Champions League18 games · early | 12–3–3W–D–L | 66.7%80.0% with a winner |
| All since launch854 games | 472–382 | 55.3% |
Pre-launch testing · Mar 1 – Aug 31, 2026
| Sport | Record | Won |
|---|---|---|
| NBA2025–26 season, Mar – Jun | 146–49 | 74.9% |
| MLBApr – Aug 31 | 822–671 | 55.1% |
| World Cup 2026Jun 11 – Jul 19 | 65–22–17W–D–L | 62.5%78.8% with a winner |
| MLSJul 22 – Aug 30, 97 games · early | 47–28–22W–D–L | 48.5%68.1% with a winner |
| NFL preseason47 games + 2 ties · early | 25–22 | 53.2% |
| La Liga19 games · early | 10–6–3W–D–L | 52.6%76.9% with a winner |
| Premier League10 games · early | 4–1–5W–D–L | 40.0%44.4% with a winner |
| All pre-launch1,965 games | 1,119–846 | 56.9% |
Soccer has three outcomes, so we show the draws. Our call is on which team wins, and a draw counts as a miss. In club soccer roughly one match in four ends level. Soccer rows show W–D–L (won, drawn, lost), and the small number under “Won” is the record on matches that had a winner. At the World Cup the model called a draw twice and was right both times; those count as wins.
Every record on this page is the call on who wins the game; spread and total calls aren’t included. “Early” means fewer than 100 graded games, too few to read much into yet. Pre-launch counts only picks recorded before the start, except the World Cup, where every graded match in the backend log is included. All games since March 1: 1,591–1,228 (56.4%), ties not counted.
Month by month since launch
October beat September on every measure: the overall win rate rose from 53.8% to 57.7%, PLAY calls from 62.5% to 67.9%, and the gap to the market favorite closed from 4.5 points to 1.6.
| Month | Games | Won | PLAY won | vs favorite |
|---|---|---|---|---|
| September | 530 | 53.8% | 62.5% | −4.5 |
| Octoberthrough Oct 26 | 324 | 57.7% | 67.9% | −1.6 |
“vs favorite” is our win rate minus the market favorite’s on the same games, in percentage points. Soccer draws count as misses.
Why it moves. Part of October’s climb is the mix: fall brings NFL, NBA and college football, where the model reads best, while MLB, our toughest sport, wound down. Two months is a short run, so we’ll add every month here as it finishes, including the ones that dip.
How the model learns. Every graded result goes back into PrizmBet’s own database. Each week, scorecards check which factors are pulling their weight and whether every confidence tier is winning as often as it says it should. Proposed changes run in shadow mode first, scored side by side with the live model, and only go live once they prove out. The record on this page updates every week.
Against the market: since launch, on the 734 games where we stored the market line, our pick won 54.1% and the team the market priced as favorite won 57.5%. The pick alone doesn’t beat taking the favorite. The confidence tiers are where the number earns its keep, and the factor cards are where you see why.
Horse racing
Every runner on the card is scored on class, form, surface, distance, post and connections, then explained in plain English. More than 6,000 races graded at 19 US tracks since April 2026, our top-ranked horse checked against the result in every one.
Race cards, decoded →
How our top-ranked horse finished at each confidence level, and the tracks we cover.
Breeders’ Cup 2026 · Keeneland, Oct 30–31 →
All 14 championship races and how we read every runner.
What PrizmBet does and doesn’t do
Questions
Is PrizmBet accurate?
Since launch (Sept 1 – Oct 26, 2026) our pick has won 55.3% of 854 graded games, with soccer draws counted as misses. The confidence number is what matters: PLAY calls (65%+) won 65.5%, LEAN 56.7%, SKIP 45.1%. In the 2026 NFL regular season through Week 7 we’re 64–41 (61.0%). Including pre-launch testing from March 1, it’s 1,591–1,228 (56.4%).
Is PrizmBet getting better?
Since launch, month over month, yes: the overall win rate went from 53.8% in September to 57.7% in October, PLAY calls from 62.5% to 67.9%, and the gap to the market favorite closed from 4.5 points to 1.6. Part of that is the sport mix, since fall brings NFL, NBA and college football. Every graded result feeds weekly scorecards, and the record here updates every week, dips included.
How is a call graded?
The call is locked before the game starts and graded automatically against the final score. A miss stays on the record. NFL ties are pushes and left out of the count. In soccer, a draw counts as a miss for a pick of either team, and we also show the record on matches that had a winner.
What do PLAY, LEAN and SKIP mean?
PLAY is 65% confidence or higher, LEAN is 55–64%, SKIP is under 55%. SKIP means the factors don’t clearly favor either side.
Is PrizmBet better than just taking the favorite?
Not across the board. Since launch, on the 734 games where we stored the market line, our pick won 54.1% and the favorite won 57.5%. Where PrizmBet earns its keep is the confidence number and the reasons: PLAY calls (65%+) have won about two in three, and every game shows the factors behind the number, including the ones pointing the other way.
Does PrizmBet tell you who to bet on?
No. PrizmBet is a sports analytics app. It doesn’t accept wagers, sell picks or give betting advice. It shows the data and leaves the decision to you. 21+.
Does PrizmBet publish its formula?
No. Every factor is named on every game, but the weights and the model are proprietary. The record is public so you can judge the output yourself.
Sportsbook is where you bet. PrizmBet is how you decide.
Record figures are from PrizmBet’s graded prediction log as of October 26, 2026, and update as games finish. Past results don’t predict future results. PrizmBet is a data-driven decision-support tool from Prizm Bet, Inc. Not a sportsbook. It does not accept wagers and does not sell picks. Data only, not betting advice. 21+. Problem gambling help: call or text 1-800-MY-RESET. PrizmBet is not affiliated with, endorsed by or sponsored by any league or team.
prizm.bet · Racing · Privacy · Terms · Support · X @PrizmBet