SkillEngine (v1.0)

Python analytics pipeline for evaluating MLB player skill using Statcast leaderboard data and percentile-based scoring models.

SkillEngine is a controlled baseball analytics project focused on core player skill only.

It produces percentile-based skill scores for qualified hitters and pitchers using Baseball Savant leaderboard data and exports ranked lists including Top 50 outputs.

The system is implemented as a reproducible Python pipeline capable of processing multiple seasons of data.

---

Purpose

• Build a resume-ready analytics engineering project with clear scope boundaries, repeatable outputs, and versioned changes.
• Provide a reusable core skill model that can serve as the baseline layer for future matchup/context modeling.
• Establish a reproducible pipeline that can process historical seasons and future seasons without code changes.

---

Tech Stack

• Python
• Pandas
• CSV data processing
• Baseball Savant leaderboard exports
• Deterministic ranking algorithms

---

Scope Boundary (Important)

SkillEngine intentionally does not include:

• Home/away splits
• Weather
• Pitch-type matchups
• Ballpark factors
• DFS logic
• Betting odds

Those features will be implemented later in a separate layer called MatchupEngine.

SkillEngine evaluates baseline player skill independent of matchup context.

---

Data Sources

SkillEngine uses Baseball Savant Custom Leaderboard exports.

Each season requires two raw files:

{season}_batting.csv
{season}_pitching.csv

Example:

2024_batting.csv
2024_pitching.csv

Raw files must be placed in:

01_data/raw/

The exact export process is documented in:

04_docs/SAVANT_EXPORT_GUIDE.md

This document ensures the raw data can always be reproduced consistently.

---

Pipeline Execution

The full pipeline runs using:

python 02_src/run_skillengine.py {season}

Example:

python 02_src/run_skillengine.py 2024

Pipeline stages:

1. build_master_dataset.py

   • Column mapping
   • Data validation
   • Player name normalization
   • Innings conversion
   • WHIP calculation
   • Qualification filtering

2. score_hitters_v1.py

   • Percentile calculations
   • SkillScore generation for hitters

3. score_pitchers_v1.py

   • Percentile calculations
   • SkillScore generation for pitchers

4. finalize_rankings_v1.py

   • Deterministic ranking
   • Top 50 export generation

---

Qualification Rules

To focus on meaningful sample sizes, the master datasets apply qualification filters.

Hitters

PA ≥ 200

Pitchers

IP ≥ 80

These thresholds ensure the SkillScore is calculated using players with sufficient playing time.

---

SkillScore v1.0

All component metrics are first converted to percentiles within the qualified player pool, then combined using weighted formulas.

Hitters

SkillScore =

• 0.45 × OBP percentile
• 0.35 × SLG percentile
• -0.20 × K-rate percentile

Pitchers

SkillScore =

• 0.35 × K-rate percentile
• -0.20 × BB-rate percentile
• -0.25 × WHIP percentile
• -0.20 × ERA percentile

Interpretation:

Higher SkillScore indicates a stronger underlying skill profile relative to the qualified player pool.

---

Outputs

For each processed season, the pipeline generates the following outputs.

Master Files

{season}_hitters_master.csv
{season}_pitchers_master.csv

Scored Files

{season}_hitters_scored_v1.csv
{season}_pitchers_scored_v1.csv

Ranked Files

{season}_hitters_ranked_v1.csv
{season}_pitchers_ranked_v1.csv

Top 50 Files

{season}_hitters_top50_v1.csv
{season}_pitchers_top50_v1.csv

---

Ranking Rules (Deterministic Sorting)

To ensure rankings are reproducible, ties are resolved using deterministic tie-breakers.

Hitters Tie-Break Order

1. Higher SkillScore
2. Higher OBP
3. Higher SLG
4. Higher PA

Pitchers Tie-Break Order

1. Higher SkillScore
2. Higher K-rate
3. Lower WHIP
4. Higher IP

---

How to Use This

• Use ranked files to evaluate the full qualified player pool.
• Use top50 files as a quick shortlist for draft research or high-level analysis.
• Treat SkillEngine as the baseline skill evaluation layer.

Future systems (MatchupEngine) will adjust these baseline ratings using matchup context.

---

Versioning & Change Rules

SkillEngine follows semantic-style versioning.

Major Version (2.0, 3.0, etc.)

Structural or conceptual model changes.

Examples:

• Changing percentile methodology
• Multi-year weighting
• New core architecture

Minor Version (1.1, 1.2, etc.)

Metric additions, weight adjustments, validation improvements, or ranking refinements.

Patch Version (1.0.1)

Bug fixes that do not alter scoring logic.

---

Weight Adjustment Policy

Weights may only change if:

• Evaluation metrics show persistent mis-weighting.
• Changes are documented in CHANGELOG.md.
• Post-change validation is recorded.

No silent weight changes.

---

Change History

See:

CHANGELOG.md

for versioned model changes, validation notes, and architectural updates.