Collect Public Data
Use Statcast, FanGraphs, and curated prospect records to build a reproducible analysis base.
Baseball Analytics Case Study
Public-data audit of player-development translation and roster-construction signals.
A reproducible Python case study using Statcast, FanGraphs, and a curated prospect cohort to examine where measurable inputs translated into MLB value and where they did not.
The project evaluates pitch-level vulnerabilities, prospect readiness, roster archetypes, baserunning, defense, and team-level value using public data, notebooks, reusable feature engineering, and written findings.
It is framed as an analytical audit, not a causal claim. The workflow is designed to make assumptions, uncertainty, and limitations visible.
- Data
- Statcast, FanGraphs batting/fielding data, and a curated 43-prospect cohort.
- Methods
- Vectorized pandas features, effect sizes, z-scores, regression, cross-validation, and notebooks.
- Outputs
- Twelve notebooks, generated figures, methodology notes, limitations, and a concise findings summary.
- Posture
- Observational public-data analysis, not access to private models, coaching decisions, or internal data.
2017-26
Analysis window
43
Prospects in cohort
16
Organizations compared
12
Case-study notebooks
94
Automated tests
8
Summary findings
Workflow
Engineer Features
Create pitch-type, count, velocity, plate-discipline, batted-ball, roster, baserunning, and fielding features.
Compare Cohorts
Evaluate player-development translation, organizational outcomes, and balanced roster archetypes.
Model Team Value
Test exploratory pressure, baserunning, and defense composites against public WAR outcomes.
Document Limits
Separate measured patterns from causality and account for sample size, public-data gaps, and ongoing seasons.
Analysis Areas
notebooks/01_translation_gap.ipynb
Translation Gap
Tools-to-production analysis for how measurable prospect inputs translated into MLB outcomes.
notebooks/02_pitch_diagnostics.ipynb
Pitch Diagnostics
Pitch-type and count-specific vulnerabilities across chase, whiff, and contact-quality signals.
notebooks/07_yankees_systemic.ipynb
Organization Comparison
Curated cohort comparison across target organizations, with sample-size caveats.
notebooks/08_yankees_case_studies.ipynb
Roster Case Studies
Baserunning, defense, home-run dependency, roster extremes, and construction tradeoffs.
notebooks/09_team_value_composite.ipynb
Team Value Composite
Exploratory pressure, baserunning, and defense composite tested against team WAR.
notebooks/10_rice_comparison.ipynb
Ben Rice Comparison
Internal counterexample showing a different readiness-gate profile and live validation case.
Data Analysis Proof Points
src/fire_fishman/
Reusable Data Helpers
Project code fetches and caches public data, then exposes reusable feature-engineering helpers for notebooks.
pitch-level features
Feature Engineering
Builds chase, whiff, velocity-tier, pitch-type, batted-ball, and count-context features from public data.
readiness gates
Validation Framework
Uses readiness gates to compare prospect profiles before updating the read with MLB outcomes.
notebooks/10_rice_comparison.ipynb
Ben Rice Check
Rice is the internal counterexample: lower-profile path, stronger readiness-gate profile, and better early validation.
notebooks/09_team_value_composite.ipynb
Model Sanity Checks
Regression and Bayesian checks test whether non-offensive components add descriptive value beyond offense.
docs/limitations.md
Interpretation Discipline
Separates measured public-data patterns from causal claims, private decision-making, and scouting certainty.
Findings
Pitch-Type Signals
Offspeed chase, breaking-ball chase, and high-velocity whiff created clearer separation than aggregate discipline.
Live Validation Case
Ben Rice passed 5/5 readiness gates, broke out in 2025, and opened 2026 on OPS/SLG leaderboards.
Open Evaluations
Dominguez and Volpe remain live cases; the project flags why neither should be treated as closed.
Organization Comparison
The curated cohort shows Yankees outcomes lagging top comparison organizations, with modest-sample caveats.
Baserunning Decline
Public BsR moved from +7.6 in 2017 to -17.2 in 2024, with cumulative -39.2 from 2018-2024.
Exploratory Composite
Pressure, baserunning, and defense added descriptive signal beyond offense, but not causal proof.
Scope And Limitations
Observational
The analysis identifies public-data patterns and missed value. It cannot fully observe internal decisions, coaching context, private models, or health.
Sample Size
Prospect cohorts and organizational comparisons are modest. Labels and live-player evaluations can change as seasons and careers evolve.
Public Metrics
Statcast and FanGraphs metrics are useful but incomplete proxies for player quality, fielding value, and organizational process.
Model Use
Regression and machine-learning checks support exploratory analysis. They are not scouting grades, causal models, or predictive guarantees.