What change unlocks this pitcher?
Pittsburgh staff · 2026
Sorted by Keystone Stuff. Select a pitcher. Δ = Stuff change since 2025 (one model, each season on its own league); a drop is flagged only when a model that never saw him agrees.
| Pitcher | FIP | Stuff | Δ | Loc | Best pitch | Flags |
|---|---|---|---|---|---|---|
| 2027 minors | 7.01 | 115 | — | 89 | SL 114 | |
| 2027 bullpen | 3.36 | 114 | +2 | 79 | ST 117 | walks |
| 2027 bullpen Medical note | 3.73 | 113 | — | 75 | CU 121 | walks |
| Leaving (free agent) | 3.32 | 110 | 0 | 81 | ST 113 | |
| 2027 bullpen | 2.36 | 110 | −7 | 97 | CU 117 | down |
| 2027 call-up | 5.83 | 108 | — | 102 | 4S 101 | |
| 2027 rotation Medical note | 3.22 | 106 | — | 105 | SL 109 | |
| 2027 bullpen | 2.97 | 105 | 0 | 108 | 4S 105 | +ST |
| 2027 rotation | 3.33 | 104 | −10 | 108 | ST 119 | down |
| 2027 rotation Health: velocity check | 4.24 | 103 | — | 98 | ST 117 | |
| 2027 rotation Medical hold | 3.50 | 102 | −1 | 113 | CU 110 | |
| 2027 bullpen Medical note | 3.22 | 102 | — | 98 | SL 109 | |
| 2027 call-up | 4.32 | 100 | −7 | 93 | CU 106 | |
| 2027 bullpen Medical hold | 3.72 | 98 | −1 | 92 | 4S 99 | |
| 2027 call-up | 4.34 | 97 | — | 88 | 4S 99 | |
| 2027 bullpen | 3.75 | 94 | −9 | 105 | FS 100 | down |
| Leaving (free agent) Medical hold | 3.19 | 93 | −6 | 100 | 4S 96 | |
| 2027 rotation Medical hold | 4.54 | 93 | −1 | 105 | ST 108 | |
| 2027 bullpen | 3.67 | 92 | +1 | 97 | SL 96 |
Paul Skenes · arsenal
SP · 174.1 IP · 28.1% K · 6.9% BB
Stuff down 10 points since 2025 (each season against its league; 11 by a model that never saw him; 104 in 2026; fastball −1.4 mph; arm angle +2°), about 9.4 runs a season at his 2026 workload on the 10 points both agree on. Find the cause before anything else. Development plan· Extension brief
| Pitch | Use | Velo | IVB | HB | Stuff | Loc | Whiff | xRV/100 |
|---|---|---|---|---|---|---|---|---|
| Four-seam41% use | 41% | 96.8 | 13.0 | +12.6 | 99 | 108 | 30% | +0.97 |
| Sweeper16% use | 16% | 84.3 | 5.7 | −10.2 | 119 | 101 | 29% | +1.16 |
| Changeup12% use | 12% | 88.7 | 6.5 | +15.4 | 90 | 95 | 37% | −0.25 |
| Sinker12% use | 12% | 96.6 | 7.3 | +17.0 | 107 | 105 | 15% | +0.36 |
| Splitter10% use | 10% | 93.2 | 1.7 | +13.3 | 101 | 106 | 18% | −0.51 |
| Slider9% use | 9% | 86.7 | 1.9 | −3.3 | 108 | 110 | 36% | +1.27 |
Changeup grades 90 with below-average expected results; his sweeper grades 119. The best case is against lefties: the sweeper is worth +1.4 runs per 100 pitches more, 1.9 standard errors (136 and 289 pitches), short of the 2 needed. No mix change until it clears.
Stuff or command?
Pittsburgh pitchers with 300+ pitches. Up = better pitches; right = better locations; both lines at 100, the 2026 league average. Circle size = pitches thrown.
What the grades are for
- Stuff predicts about as well as results do. Graded by a model trained on 2024 only, a pitcher’s 2025 Stuff tracks his 2026 results at r 0.49, against 0.45 for his 2025 results themselves. For a pitcher the model has never seen it does worse: r 0.39, against 0.47 for pitchers it has seen at the same training size and 0.46 for their results. On a thin sample it’s an early read, not a better one.
- Location diagnoses walks. It predicts next season’s walk rate (r 0.46), a little less well than last season’s walk rate does (0.53), and adds nothing for runs (r −0.03).
- Pitching describes. Stuff and location together explain this season best, but looking forward they add nothing to Stuff (r 0.48 vs. 0.49).
- Time-ordered test, 324 pitchers with 500+ pitches in both seasons: Stuff and Pitching trained on 2024 pitches only; Location fit within 2025 on that season's other games (cross-fit by game), centered on 2025; it sees no 2026 pitch. The site’s season grades are cross-fit (each season graded by a model trained on the others), which is fair pitch by pitch but lets a “next season” test peek at the future; this one can’t. The 2026 league medians: Stuff 100, Location 99.
- Refit noise. Refit the model on different seasons and a pitcher’s 2026 Stuff moves about ±2.9 points (SD; 90% within 5.1), and his change since 2025 about ±2.1 (90% within 3.3), comparing the production model with a refit on 2025–26 pitches only (both saw his two seasons) across 469 pitchers with 500+ pitches. Read a gap of a few points as a tie. A Stuff alert whose agreed drop sits within 2.1 points of the −5 bar is marked borderline.
Command work: Evan Sisk (Location 75, 2nd percentile; 12.7% walks); Yohan Ramírez (Location 79, 3rd percentile; 12.2% walks). The rule: bottom-quarter Location and a walk rate clearly above the 8.6% median over 150+ batters, the same rule the development plans use.
How we knowThe coordinator’s work list
Stuff declines first (the cause comes before any fix), then pitch-mix changes, shape goals from the development plans and new pitches in runs a 2027 season, then command work, which carries a walk target because Location doesn’t forecast runs. Medical holds sort last; players leaving this winter are out.
- 1 Stuff alert Stuff 104, −10 since 2025 (−11 by a model that never saw him); fastball −1.4 mph; arm angle +2°: find the cause ≈9.4 runs at stake
- 2 Stuff alert Stuff 94, −9 since 2025 (−8 by a model that never saw him); fastball −1.0 mph: find the cause ≈5.5 runs at stake
- 3 Stuff alert Stuff 110, −7 since 2025 (−7 by a model that never saw him); fastball −0.2 mph; arm angle +3°: find the causeBorderline ≈2.5 runs at stake
- 4 Shape Get the four-seamer back to its 2024 shape (2″ more ride)Medical noteplan ≈1.9 runs/season if it takes
- 5 New pitch Sweeper (83.2 mph), +6 or better on his cutter under every model (median +8) 0.4 runs/season if it takes
- 6 Command Location 79 (3rd percentile), 12.2% walks walks vs. 8.6% median
- 7 Command Location 75 (2nd percentile), 12.7% walksMedical note walks vs. 8.6% median
Shape work and new pitches together: about 2.2 runs a season if all of them take (new pitches often don’t; a shape may not come back). Stuff declines are runs at stake, not runs to gain: a lost pitch may not come back.
How a mix change is picked: the model first (a pitch graded 96 Stuff or below, thrown 8%+ of the time, with a pitch of 104+ to take its usage); its expected results running below its family’s average are a secondary screen with no significance test of their own. The switch applies only against a batter side where the better pitch has beaten it by 2+ standard errors.
Considered, not queued (6)
- : Sinker: +21 on his splitter (+13 or better under every fit), but that would out-grade 100% of the comps’ real pitches, more likely the construction than the pitch (fragile).
- : Changeup grades 90 (the sweeper 119), but no batter side where the replacement is better by 2+ standard errors.
- : Cutter grades 95 (the slider 109), but no batter side where the replacement is better by 2+ standard errors.
- : Changeup grades 91 (the sweeper 108), but no batter side where the replacement is better by 2+ standard errors.
- : Curveball grades 92 (the sweeper 108), but no batter side where the replacement is better by 2+ standard errors.
- : Sinker grades 86 (the sweeper 108), but no batter side where the replacement is better by 2+ standard errors.
League Stuff leaders · 1,000+ pitches
Context for what elite looks like in 2026
- 1 Tyler Rogers TORa model that never saw him: 93 140
- 2 Mason Miller SD 133
- 3 Adrian Morejon SD 132
- 4 Jacob Misiorowski MIL 124
- 5 Chris Sale ATLa model that never saw him: 105 121
- 6 Louis Varland TOR 120
- 7 Alex Vesia LAD 120
- 8 Steven Okert HOU 119
- 9 Chase Burns CIN 119
- 10 Ryan Zeferjahn CHC 118
- 11 Bryan Hudson CWSa model that never saw him: 91 117
- 12 Aaron Ashby MILa model that never saw him: 92 117
Marked: the pitcher-held-out model (the one Stuff alerts must agree with) grades the pitcher 15+ points below his published grade. The model behind a published grade was trained on his other seasons, and for a slot or shape with few neighbours it is partly reading them, so treat that grade as an upper bound.