What This Actually Is
MLB PC is a standalone baseball simulation engine that runs on Windows and lets you build, run, and visualize full MLB seasons with detailed player progression, park effects, and weather systems. It is not a game in the traditional sense — it is a calculator with a front end. The community around it has accumulated a ton of documentation over the years, and if you do not know where to start, you will waste an afternoon digging through dead links and broken config files. I have been running simulations on this engine since 2019, and the core issue most people hit is not installation — it is parameter tuning. The default config assumes you are playing for casual fun, not competitive analysis or season projection work. I learned this the hard way after running a full simulated season where every starting pitcher threw a 75-pitch average because the fatigue curve was flatlined at the factory settings. My workaround was to open the fatigue_params.json file and change the decay_rate from 0.02 to 0.065, which matched the actual strikeout decline I was seeing against real 2018-2023 data. That single tweak cut my average game length from a padded 9 innings of high-scoring nonsense to roughly 7.2 innings per contest, which is much closer to actual MLB distribution. The engine itself is modular. You download a base installer, then layer in the database pack — either the official MLBAM snapshot or a community-curated version that backfills minor league stats. The official pack is more accurate but lags behind by about six weeks. Community packs fill that gap but occasionally contain copied errors. I cross-reference both and flag any player whose wRC+ drops below 40 above league average — that is almost always a data entry mistake.
Installation and Setup
Grab the base executable from the main project page. It is a single self-contained Windows binary, roughly 400 MB including the starter database. Extract it to a folder with no spaces in the path — I put mine at C:\sim\base\ because mixing Unicode paths with the config loader caused save corruption once and it only shows up when you are deep into a 300-game sim. After extraction, run the config wizard. It asks for your database source, difficulty preset, and whether you want replay export enabled. For difficulty, pick Manual or Advanced. Easy mode disables weather and park effects, which sounds fine until you realize you are not actually simulating baseball anymore — you are running a batting average spreadsheet. Advanced mode turns on everything: humidity, wind velocity, UV index affecting ball trajectory, and altitude correction for Coors. The engine handles all of this internally using a modified Bill James Pythagenpat formula with a Lincecum adjustment factor. The replay export is optional but highly recommended if you plan to publish results. It writes each pitch as a JSON line so you can replay games later or feed them into another analysis tool. I store mine in a separate directory and compress weekly with 7z — a full 162-game season at detailed replay is about 1.4 GB uncompressed. Compressed it comes to roughly 340 MB.
Running Your First Season
Once the config is set, select a database snapshot and hit Run Season. The engine cycles through 162 games per team, calculating box scores in real time. You can watch each game or skip to the end. Skipping loses replay data unless you exported it first. I recommend watching at least five games manually before going full auto — it helps you calibrate whether the RNG is behaving in a way that makes sense to you. One thing nobody mentions upfront: the engine does not auto-balance rosters. If you load a 2024 roster dump, the Dodgers will dominate because they actually are the best team. If you want parity, run the balance tool before starting. It adjusts payroll and minor league depth so no single team has more than three players above a 120 wRC+ threshold. This is important if you are trying to predict standings rather than just watch your favorite team crush everyone.
Get the Full Details
Common Pitfalls
The biggest failure mode is loading mismatched data years. If your player database is 2023 but your park database is 2019, the park dimensions and wind corridors will not align with current defensive shifts, and your left-field games at Oakland will throw outlier numbers. I caught this once when three consecutive home games at SFP were producing 15-run affairs. The park database had not been updated since the AT&T Park rename era, and the shift configuration in the player DB was modern. Mismatched metadata like that corrupts the outcome logic silently — the engine does not warn you. Another issue is memory limits. The engine loads the full player pool into RAM at startup. A complete database with minor leaguers included can consume 2.1 GB. If you are on a system with 8 GB total and run a browser alongside the sim, you will hit swap and the run time will triple. Close unnecessary applications or switch to the lightweight database pack that excludes MiLB callups.
Advanced Usage
Once you are comfortable, you can inject custom rules — designated hitter variants, pitching changes that respect real manager tendencies, or even experimental formats like 7-inning doubleheaders. The config file supports JSON overrides for most systems. I have run a version where relief pitchers must face at least two batters before being removed, modeled after the old AL rule before the universal DH took over. It changes bullpen management significantly and produces fewer low-leverage matchup events. You can also write simple Python scripts that query the replay JSON and output win probability graphs, leverage-adjusted ERA projections, or park-factor adjustments per stadium. There is a community repo on GitHub with a few starter notebooks. They assume you have pandas and numpy installed. The data export format is consistent enough that reverse-engineering it is straightforward if you already have access to the schema docs.
What It Cannot Do
Be honest about the limitations. This is not MLB The Show. There is no controller support, no graphics beyond a scorecard view, and no franchise mode with player morale or contract negotiations. It simulates outcomes, not experiences. If you want to feel like you are managing a club, you will be disappointed. If you want to understand how a rotation performs over 162 games with proper regression to mean, this is genuinely useful. The accuracy ceiling is around 78% on win-probability calibration against actual MLB results when all parameters are tuned correctly. That is good for a spreadsheet wearing a UI, not great for anything else. For visualization-heavy work, pair it with a separate tool like Tableau or even a basic R Shiny app. The JSON export plays nicely with both. I do this for all published projections — run the sim in the background, feed the results into a dashboard script, and generate charts automatically. The whole pipeline from sim start to published PDF takes about 12 minutes on a modern CPU.
