What Goes Into PSI+
Measures how often a pitcher misses bats when it matters most. Two-strike whiffs count double. First-pitch misses count half. The heaviest component in PSI+.
Pitches are weighted by count leverage before calculating whiff rate.
Captures the top-end speed a pitcher can reach when the moment demands it. Not average velocity — the high gear they can access in big counts.
95th percentile of release speed across four-seam fastballs, sinkers, and cutters — measuring the top-end speed a pitcher can reach, not their average.
Measures how flat or steep a fastball enters the strike zone. Flatter angles are harder for hitters to square up. A smaller component, but stable year over year.
Mean vertical approach angle of fastballs at the front of home plate. More negative values indicate a flatter plane into the zone.
The same count-leverage logic as CLW. Applied only to secondary pitches — breaking balls, changeups, and off-speed offerings. Only included when a pitcher has thrown at least 50 secondary pitches.
Applies count-leverage multipliers to whiffs on breaking balls, changeups, and off-speed pitches. New in PSI+ v2.
−6.92° approach angle (98th percentile) · CLW 98th percentile · 96.6 mph velocity. A sinker-ball pitcher outsmarting hitters rather than overpowering them. The clearest example of what PSI+ finds that raw K% misses.
Rolling PSI+ across starts: 97.7 → 97.0 → 95.8 → 93.6 → 92.2. Velocity slipping from 95.2 to 94.8 mph. PSI+ picked up the slide before the K rate did.
Leaderboard
Minimum 200 pitches. Scored within role. 100 = league average.
Rolling PSI+ Over Time
Track how a pitcher's strikeout ability has evolved across starts. Searches 2026 qualifying pitchers.
Where K% Missed, PSI+ Didn't
Pitchers where PSI+ disagreed with their raw strikeout rate. The following season showed who was right.
PSI+ flagged underrated pitchers correctly 58.8% of the time when the K% rose the next year. The overrated signal was correct 79.1% of the time. At larger divergences greater than 1.0 standard deviation, overall accuracy rose to 74.5% across 94 cases.
Does It Actually Work?
PSI+ was built entirely on pre-2025 data. Then we tested it on 2025 pitchers the model had never seen. Every result below is from that blind test.
How We Found the Signal
We tested 51 different pitcher stats to find which ones best predict future strikeout rate. Most well-known stats fell short. The ones that made it into PSI+ are the ones that actually held up.
Count-leveraged whiff rate outperforms every publicly available strikeout predictor we tested. r = 0.5818 vs. 0.4892 for CSW%. Missing bats matters. Missing them in two-strike counts matters more.
How the Weights Were Chosen
We tested every combination of CLW, velocity, and VAA weights against the 2025 holdout data. That optimization informed v1. PSI+ v2 adds a fourth component and adjusts the weights accordingly.
v1 winner: CLW = 60%, Velocity = 30%, VAA = 10%. Every combination that gave VAA more than 10% weight underperformed. VAA and velocity are correlated, so over-weighting VAA was essentially counting the velocity signal twice.
v2 adds SLWR as a fourth component. Secondary pitch quality was missing from v1. The original three weights were adjusted to make room.
When a pitcher has fewer than 50 secondary pitches, SLWR is excluded and the remaining three weights are rescaled proportionally.
PSI+ runs on two separate weighting schemes depending on the use case.
Uses the core weights above. Optimized for predicting strikeout rate year over year. This is what appears on the leaderboard.
Uses a separate set of weights not published here. Optimized for a different objective. The two versions will diverge on some pitchers.
What Didn't Work
Two approaches that looked promising on paper and failed in testing. Showing them here because credibility means showing what didn't work, not just what did.
Measured how similar two pitches look to a hitter before they break in different directions.
Whether throwing one type of pitch made the next pitch harder to hit. We analyzed over 2 million consecutive pitch pairs with adjusted baselines.
Both had intuitive appeal. Both failed validation. What the data consistently rewarded was simpler: count leverage + stuff quality.