This felt impossible. What?! Is this the brainrot everyone is talking about? I mean I still read code every day, I just don't write it anymore. I could read and review the solution easily, but writing it myself felt like speaking a language I only understand, but cannot speak.
Being the data hoarder I am, I immediately wanted numbers, so I built Muscle Memory to help me feed them.
It's like leetcode but better. Challenges are more realistic, and I'd like to think the UX is more straightforward. It covers 12 languages and your code runs in a network-denied sandbox in your browser. No account needed.
I want it to be a place to keep your hands sharp: short tasks, clear prompts, and difficulty that comes from solving the problem rather than figuring out what the question means.
About the data (that I'm curious about): With your consent, Muscle Memory records your practice as raw events, including keystrokes, runs and submissions. Hoping to build an open dataset of how we write code in the AI era. That could help us understand how we approach problems, where we get stuck, and how we well do we retain our craft during this transition.
Your individual results stay private unless you share them. “Stealth” mode keeps your data out of published findings, and you can export or delete everything.
The data will be published as a report/notebook on the site itself, and be available raw under MIT license (parquet) once we reach a critical mass of sorts, so it makes sense to share.
The code is also MIT-licensed on GitHub (https://github.com/ekkuleivonen/muscle-memory). I wrote more about why I built it in the manifesto here: https://musclememory.run/manifesto.
I'd especially like to hear what surprised you about your first attempt.
Try it out here: https://musclememory.run