RISEBALL REPORT
So you don't swing under it.The ratings called 71% of D1 softball games before they were played.
Every Monday from March 1, the model was refit on games already played and asked to pick the next seven days of winners. 13 seasons (2012 to 2025, no 2020), 71,278 games between D1 teams, and nothing it saw came from on or after game day.
Games called71.0%of 71,278 D1 games · RPI 69.2% · record 66.5%
Seasons ahead of RPI12 of 13the miss: 2012 by 0.1 pts
NCAA tournament76.1%1,737 games · RPI 74.3%
2025 tournament99 of 134RPI 94 of 134
Season by season
Ahead of the RPI in 12 of 13 seasons
The RPI is the number the selection committee leans on, so it is the fair benchmark. The record baseline picks whichever team has the better winning percentage.
Riseball ratingsRPIedge = points ahead of the RPI
Show the season table
| Season | Games | Ratings | RPI | Record | Edge |
|---|---|---|---|---|---|
| 2012 | 5,853 | 69.2% | 69.3% | 66.3% | -0.1 |
| 2013 | 5,750 | 71.5% | 69.1% | 66.8% | +2.4 |
| 2014 | 5,795 | 70.8% | 68.6% | 66.2% | +2.2 |
| 2015 | 5,692 | 72.7% | 70.7% | 68.0% | +2.0 |
| 2016 | 5,343 | 71.0% | 70.0% | 66.4% | +1.0 |
| 2017 | 5,437 | 69.6% | 68.2% | 65.4% | +1.4 |
| 2018 | 5,405 | 71.5% | 69.4% | 66.5% | +2.1 |
| 2019 | 5,426 | 71.3% | 70.5% | 66.4% | +0.8 |
| 2021 | 5,274 | 71.5% | 67.9% | 66.7% | +3.6 |
| 2022 | 5,210 | 71.3% | 68.9% | 66.6% | +2.4 |
| 2023 | 5,341 | 70.8% | 68.2% | 65.8% | +2.5 |
| 2024 | 5,402 | 70.7% | 69.0% | 65.9% | +1.7 |
| 2025 | 5,350 | 71.4% | 69.4% | 67.1% | +2.0 |
Calibration
When it says 80%, it means 80%
A pick is only half the answer. Each game also gets a win probability, and those probabilities hold up: when the model gave the favorite 80-90 percent, the favorite won 86% of the time. Every band lands within 2 points of its forecast.
That is the difference between a ranking and a forecast. A ranking says who is better. A forecast says how often the better team wins, which is what you need to set a pitching plan or judge a weekend.
Where the edge comes from
Biggest in March, still there in the postseason
In March the RPI has three weeks of results to work with. The ratings also know how good each team was last season, pulled a quarter of the way back to average, so they start closer to the truth.
By May both have seen plenty, and the gap narrows. The model still beats the RPI in the NCAA tournament: 76.1% to 74.3% over 1,737 games.
Head to head
When they disagree, the ratings win 56% of the time
In 10,115 games the ratings and the RPI picked different winners. The ratings were right in 56.5% of them. In the 2025 NCAA tournament they split 11 times, and the ratings were right in 8.
Several of these were near coin flips, and the table shows the model's number so you can judge them that way.
| Date | Final | Ratings picked | RPI picked | Right |
|---|---|---|---|---|
| May 16 | Michigan 4, UCF 3 | Michigan (50%) | UCF | Ratings |
| May 17 | Stanford 14, Oregon 1 | Oregon (58%) | Stanford | RPI |
| May 17 | UCF 10, Michigan 8 | Michigan (52%) | UCF | RPI |
| May 18 | Oregon 15, Stanford 5 | Oregon (55%) | Stanford | Ratings |
| May 18 | Stanford 7, Oregon 10 | Oregon (58%) | Stanford | Ratings |
| May 23 | UCLA 2, South Carolina 9 | UCLA (61%) | South Carolina | RPI |
| May 24 | South Carolina 4, UCLA 5 | UCLA (63%) | South Carolina | Ratings |
| May 25 | UCLA 5, South Carolina 0 | UCLA (61%) | South Carolina | Ratings |
| May 29 | Texas 3, Florida 0 | Texas (50%) | Florida | Ratings |
| May 29 | Ole Miss 0, Texas Tech 1 | Texas Tech (59%) | Ole Miss | Ratings |
| May 30 | Florida 3, Tennessee 11 | Tennessee (54%) | Florida | Ratings |
Plain talk
What a 2-point edge is worth
Softball is noisy, and no model gets near 80% over a full schedule. Over one team's season, 1.8 points is about one more game called right. Across D1 it is roughly 101 more games a season.
The bigger value is what the RPI cannot give you at all: a win probability, an expected run margin, and a separate offense and defense rating for every team. Those are what the scouting reports are built on.
How it was tested
Method
- Ratings: a Poisson model of runs scored with separate offense and defense ratings for every D1 team, a home edge, and shrinkage so thin schedules do not swing the numbers.
- Walk-forward: each Monday from March 1, refit on every game played before that day, then predict the D1-vs-D1 games of the next seven days. Repeat through the World Series.
- Prior season: last year's final ratings, pulled 25% back toward average, are the starting point. 2012 had no prior season in the archive.
- Settings (shrinkage, prior weight, probability scaling) were chosen on 2013 to 2015 only. On 2016 to 2025, which never touched the tuning, the ratings called 71.0% of 48,188 games against the RPI's 69.1%.
- RPI: 25% winning percentage, 50% opponents' winning percentage (games against the team removed), 25% opponents' opponents', from D1 games before the cutoff. Ties go to the home team.
- Women's College World Series games are treated as neutral site. 2020 is left out because the season stopped in March. Tie games are skipped.
- Data: every final score in the softballR-data archive of stats.ncaa.org.