How accurate is Clutch?
Rebuilt every morning from the Ledger: the season's graded calls, how often each number came true, and the same games for ESPN BPI, the market and picking the home team every night.
This season, call by call
| NUMBER | ACCURACY SCORE | GAMES |
|---|---|---|
| Clutch's call | 0.189 | 36 |
| ESPN BPI | 0.179 | 36 |
| The market · final price | 0.191 | 36 |
The accuracy score (the Brier score) grades the probability itself, not just the side: lower is better, 0.250 is a coin flip.
31 more calls are left out because they can't be proven to predate tip; the count is printed so the record can't quietly shrink. Graded games from Oct 20, 2026 to Oct 24, 2026. Preseason is never graded.
When Clutch says 70%, does it happen 70%?
A 70% call should come true about 70 times in 100 — if Clutch's 70s win only half the time, the number is lying, however many games it gets right. So the season's calls are put in bands by the number Clutch gave the side it favoured, and each band is checked against what happened.
No band holds 20 calls yet, so each one shows how many came true rather than a rate.
| CLUTCH'S BAND | CALLS | SAID | HAPPENED | LIKELY RANGE |
|---|---|---|---|---|
| 50–60% | 18 | 54% | 10 of 18 | — |
| 60–70% | 7 | 64% | 3 of 7 | — |
| 70–80% | 6 | 77% | 6 of 6 | — |
| 80–100% | 5 | 83% | 5 of 5 | — |
SAID is the average of Clutch's numbers in the band; HAPPENED is how often that side won. A band shows its rate and likely range from 20 calls.
The two-season test
Replayed over two seasons (2024-25 and 2025-26, 2,631 games), each game predicted only from the games before it, Clutch's probabilities were more accurate than ESPN BPI's in both seasons (clearly in 2025-26; in 2024-25 the gap was within the noise) by the accuracy score, which grades the probability itself. Against the market's final price they were level: as good as the market, not better. On winners alone, the three were within 1.3 points of each other each season. Nobody publishing an NBA model reliably beats the market, and Clutch doesn't claim to.
| NUMBER | 2025-26 RIGHT | SCORE | 2024-25 RIGHT | SCORE |
|---|---|---|---|---|
| Clutch's call | 68.5% | 0.198 | 70.0% | 0.202 |
| ESPN BPI | 68.6% | 0.202 | 69.0% | 0.206 |
| The market · final price | 68.6% | 0.197 | 68.7% | 0.201 |
| Always the home team | 55.5% | 0.247 | 54.6% | 0.248 |
2025-26: 1,316 games · 2024-25: 1,315 games. RIGHT is the share of games where the side given over 50% won; SCORE is the accuracy score (lower is better). Every game was predicted by a model fit only on earlier games, then graded on the same games as ESPN BPI and the market's final price before tip. This test chose the model; it is not its record. The full test and the model's limits →
When the test said 70%, did it happen 70%?
Across the test's 2,631 games, every band came true within 2.4 points of what Clutch said, each inside its likely range.
| CLUTCH'S BAND | GAMES | SAID | HAPPENED | LIKELY RANGE |
|---|---|---|---|---|
| 50–60% | 677 | 55.3% | 55.5% | 52–59% |
| 60–70% | 718 | 65.1% | 64.8% | 61–68% |
| 70–80% | 615 | 74.6% | 72.2% | 69–76% |
| 80–100% | 621 | 86.0% | 86.3% | 83–89% |
The same bands as the live table above, counted the same way on the side Clutch favoured, over both test seasons together. A replay, so it says how the method holds up; the live bands say how this season's sealed numbers do.
Last season's live record
In 2025-26, made by the model Clutch has since retired, the Ledger graded 62 calls: 37–25 (59.7%), against 37–25 for picking the home team on the same games. It stays on the Ledger exactly as graded.
Questions
How accurate is Clutch?
This season Clutch is 24–12 on 36 calls graded after the final, 66.7% right. Picking the home team every night went 18–18 on the same games.
Is Clutch more accurate than ESPN BPI?
In the test it had to pass before it shipped: Replayed over two seasons (2024-25 and 2025-26, 2,631 games), each game predicted only from the games before it, Clutch's probabilities were more accurate than ESPN BPI's in both seasons (clearly in 2025-26; in 2024-25 the gap was within the noise) by the accuracy score, which grades the probability itself. Against the market's final price they were level: as good as the market, not better. On winners alone, the three were within 1.3 points of each other each season. Live, this season's games are graded beside ESPN BPI on the Ledger.
Are Clutch's probabilities calibrated?
In the two-season test (2,631 games), each band of Clutch's numbers came true within 2.4 points of what it said: its 50–60% calls averaged 55% and won 56%; 60–70% calls averaged 65% and won 65%; 70–80% calls averaged 75% and won 72%; 80–100% calls averaged 86% and won 86%. This season's live bands are rebuilt every morning above.
How is a call graded?
Clutch's number for every game is sealed in a public file on GitHub before tip-off and graded after the final against that same number. The side it gives over 50% is its call. Right and wrong calls stay on the record, in date order.
Where these numbers come from
Every number Clutch publishes is sealed with a SHA-256 fingerprint in a public repository before tip, and the Ledger grades that sealed number after the final. This page is rebuilt from the Ledger every morning; nothing on it is typed by hand. How to check a seal yourself →
Clutch's numbers on this page are free to reuse under CC BY 4.0: credit Clutch and link to clutchledger.com. Scores are public record; ESPN BPI and market figures belong to their sources.