Pitch grades do not make the development decision.
A strong shape score says what tends to work across the league. It does not establish whether asking this pitcher to change this pitch is feasible, durable, or better than leaving the arsenal alone.
Independent baseball analytics research
When does changing a pitch actually help?
ShapeShift evaluates pitch-design interventions—not just pitch quality. It asks whether a pitcher should Keep, Reshape, or De-emphasize an offering, how much value the action may create, and what evidence would invalidate the recommendation.
Successful redesign stories are selected after the fact.
ShapeShift detects physical changes without looking at outcomes, retains failed and abandoned experiments, and compares treated pitches with similar not-yet-treated controls.
One recommendation, translated for three audiences.
Leadership receives a decision memo, coaches receive checkpoints and stop rules, and analysts receive the complete validation and sensitivity trail.
Research contract
Designed to survive skepticism.
- Outcome-blind detection.Shape changes are identified before their results are examined.
- Temporal separation.2021–23 develops, 2024 validates, and 2025 stays locked.
- Explicit alternatives.Keep and de-emphasize must lose before reshape can win.
- Withhold is valid.No recommendation when overlap, stability, or support fails.
How to read this site
From question to action.
- 01DecisionCompare all three actions and uncertainty.
- 02EvidenceInspect effects, support, and stress tests.
- 03Coach planTranslate evidence into a testable bullpen.
- 04MethodologyAudit the pipeline and failure criteria.
RV / 100
Expected pitcher-positive runs per 100 pitches.
Durability
Estimated probability the new shape persists for 60 days.
Common support
Comparable historical pitches exist for the counterfactual.
Command stress
Recommendation remains useful if location consistency worsens.
Locked research resultAggregate 2025 matched ATE is +0.02 RV/100 (near null). Featured case: Muñoz, Andrés slider matched effect -0.29 → De-emphasize.
Decision brief · locked 2025 evaluation
Muñoz, Andrés · RHP · slider · change date 2025-08-09
De-emphasize this pattern.
Do not pursue arm side break changes like the Muñoz, Andrés slider pattern.
Current viewDe-emphasizeModerate confidence
Expected lift-0.29runs saved / 100 pitches
80% interval-0.37 to -0.24clustered bootstrap
Durability60%probability at 60 days
Decision confidenceModeratepublic-data evidence
Compare the alternatives
One decision. Three explicit options.
Choose an option to inspect its decision case.
Why this wins
The locked average effect is +0.02 RV/100 — too small to justify broad redesign programs. The Muñoz slider analog shows a matched -0.29 RV/100 after an arm-side break and velocity shift, so the actionable call on that pattern is De-emphasize, while Keep remains the organizational default.
Evidence
Does the change hold up?
Matched, outcome-blind interventions with not-yet-treated controls. The diagnostic is designed to reject weak recommendations.
01
Matched event study
Estimate 80% interval
Event week relative to detected shape change
Locked 2025 matched event study · ATE +0.02 RV/100 · featured case -0.29
02
Feasible shape target
Induced vertical break (in)
Glove-side break (in)
Current Target band
Target constrained to comparable RHP sliders with matched release profiles
Pre-trend gapPass
Treated vs control pre-trend gap +0.01
OverlapPass
381 matched pairs within pitch-family support
Aggregate ATENull
Locked effect +0.02 (80% CI includes ~0)
Featured caseAct
Muñoz, Andrés matched -0.29 over 10 controls
Coach plan
Turn the finding into a bullpen.
No model language. Three cues, measurable checkpoints, and stop rules that protect the rest of the arsenal.
Player goal
Stop chasing the Muñoz slider shape pattern that lost matched value.
De-emphasize trial
01
Change
- Return toward the pre-change arm-side break band.
- Protect velocity; do not trade mph for extra sweep.
- Re-test only after shape stabilizes for two sessions.
02
Protect
- Keep primary fastball intent unchanged.
- Hold zone rate near the pre-change baseline.
- Avoid stacking multiple shape experiments at once.
03
Stop if
- Matched-style value stays negative after 40 pitches.
- Velocity remains down more than 1.5 mph.
- Arm-side break drifts another inch away from baseline.
Observed break shift-2.3 inarm-side break
Velocity shift-1.7 mphsession means
Matched effect-0.29RV / 100
Controls10same-family matches
Methodology
Built to be challenged.
ShapeShift separates intervention detection, outcome modeling, effect estimation, and recommendation. Each layer has a failure criterion.
01
Ingest
Immutable weekly Statcast Parquet with provenance and QA.
02
Detect
Outcome-blind physical changepoints by pitcher and pitch.
03
Estimate
Matched event study with not-yet-treated controls.
04
Decide
Keep / Reshape / Abandon under feasibility constraints.