MLB The Show 26 — Player Card Generator

A reverse-engineered MLB The Show 26 rating system. Type a player name and a year, get back a Show-style attribute card built from real MLB stats.

Works for any season from the deadball era to today — Nap Lajoie 1897, Sandy Koufax 1965, Barry Bonds 1995, Bobby Witt Jr. 2025.

What it does

• Pulls real MLB stats from Baseball-Reference and Baseball Savant (Statcast)
• Maps them to MLB The Show 26 attributes (Vision, Discipline, Power L/R, Contact L/R, Speed, Fielding, Reactions, Arm, Pitcher Velocity, Break, Control, Stamina, etc.)
• Generates a full card with a 0–99 OVR badge, color-coded by tier
• Compare two players side by side
• Auto-detects pitcher vs hitter
• Pre-Statcast eras get era-appropriate fallbacks (Total Zone for fielding, SB-based speed, manual arsenal overrides for pre-2008 pitchers)

Quick start

git clone https://github.com/<you>/mlb-show-card-generator.git
cd mlb-show-card-generator
python -m venv venv
venv/Scripts/activate            # Windows
# source venv/bin/activate       # macOS/Linux
pip install -r requirements.txt
python app.py

Then open http://localhost:5000.

Defaults to localhost only. To reach it from your phone on the same WiFi, set MLBSHOW_EXPOSE_LAN=1 before running — this exposes write endpoints to your whole network.

CLI works too:

python generate_card.py "Juan Soto" 2025
python generate_card.py "Tarik Skubal" 2025 --pitcher
python generate_card.py "Barry Bonds" 1995 --position LF

How the ratings work

Each Show attribute is mapped to a real stat with the strongest correlation:

Attribute	Formula
Vision	-2.55 * K% + 121.3
Discipline	4.86 * BB% + 17.2
Contact R	374.6 * BA + 2.30 * Exit_Velo - 217.4 (Statcast era)
Contact L	347.6 * blended_BA - 13.7 (PA-trust-weighted vs LHP)
Power R/L	~336 * ISO + 4 / ~244 * ISO + 17
Speed	15.01 * Sprint_Speed - 353.5 or SB-based fallback
Stealing	0.6 * (1.88*SB + 18.5) + 0.4 * Speed
Fielding	OAA → DRS (Rdrs) → Total Zone (Rtot) → position baseline
Pitcher Velo	3.14 * FB_Velo - 208.2
Pitcher Ctrl	-4.44 * BB% + 104.9
Pitcher K/9	2.2 * K%_split + 15 per LHB/RHB split

The OVR formulas were fit by least squares against a hand-labeled calibration set of real Show 26 cards (regression refit 2026-04-07 after migrating off FanGraphs).

Data sources

• Baseball Savant (Statcast) — sprint speed, exit velocity, OAA, pitch arsenal, pitch-by-pitch splits. 2015+.
• Baseball-Reference — season batting/pitching/fielding leaderboards, including Total Zone (Rtot) and DRS (Rdrs) for the eras where they're available. Scraped directly via cloudscraper + pandas.read_html in bbref_scraper.py.
• Chadwick Bureau — player ID lookups (pybaseball.playerid_lookup).

FanGraphs is unreachable as of 2026-04 (Cloudflare interactive challenge). FG-only metrics like Spd, BsR, Def, and RngR are returned as NaN and the ratings code falls back to alternative inputs gracefully.

Pre-Statcast tiers

Era	Speed	Fielding	Contact	Pitch Arsenal
2015+	Sprint Speed	OAA	BA + Exit Velo	Statcast
2003–2014	SB-based fallback	DRS (BBRef Rdrs)	BA only	None
1953–2002	SB-based fallback	Total Zone (Rtot)	BA only	Manual override
Pre-1953	SB-based fallback	Position baseline	BA only	Manual override

Pitch arsenals before Statcast

For pitchers before the 2008 pitch-tracking era, real arsenals are stored in pitcher_arsenals.json. Cards display these with an "EST." badge so you don't mistake reference arsenals for measured data. Add your own by appending to the JSON — no code change needed.

Project structure

generate_card.py        # Core: data pulls, attribute formulas, OVR estimation
app.py                  # Flask web app (localhost:5000)
bbref_scraper.py        # Direct Baseball-Reference scraper
pitcher_arsenals.json   # Manual pre-2008 arsenals
recalibrate.py          # OVR coefficient refitter (NNLS)
templates/              # Web UI templates

Caveats

• The OVR formulas are fit on a small calibration set (~11 hitters, ~10 pitchers). Speed-first hitters are systematically under-predicted because the calibration set lacks variance separating speed from contact ability. Working on growing the set.
• Pre-2008 pitcher arsenals are reference estimates from scouting sources, not measured pitch data. Marked clearly in the UI.
• Pitcher Break is heuristic-based — no measurable stat correlates strongly with Show's Break rating.
• Pitcher Velocity overrates soft-tossers under ~93 mph.

Running tests

Install test dependencies (pytest is already in the shared venv, but this makes it explicit):

"C:/Users/Alex/Claude Context/Robert Stock - Pitching Grade Model/venv/Scripts/python.exe" -m pip install -r requirements-dev.txt

Run the suite with one command from the repo root:

"C:/Users/Alex/Claude Context/Robert Stock - Pitching Grade Model/venv/Scripts/python.exe" -m pytest -q

All 170 tests should pass. The configuration lives in pytest.ini; conftest.py adds the project root to sys.path automatically so no manual path setup is needed. CI runs the same command on every push and pull request via .github/workflows/tests.yml.

License

MIT. Built for fun. Not affiliated with Sony, San Diego Studio, or MLB.