while True: learn()

while True: learn() is a node-based puzzle game dressed up as a machine learning career simulator. You play a developer whose cat accidentally outperforms every neural network you build, so you set out to crack feline logic by constructing data-pipeline flowcharts using drag-and-drop visual blocks. Each level gives you a dataset and an expected output, and your job is to connect classifiers, filters, decision trees, loops, and preprocessing nodes until the output matches. It is closer to a logic puzzle collection than a traditional sim, but the career wrapper, the client contracts, and the slow drip of new node types give it real progression bones. The core mechanic is genuinely well designed. Luden.io uses the puzzle format to introduce real ML ideas: supervised versus unsupervised learning, k-nearest neighbours, random forests, confusion matrices. You are not doing actual coding, but the conceptual scaffolding is honest. A computer science student will recognize the structures immediately. A curious newcomer will finish the game with a clearer mental model of machine learning than most YouTube explainers provide. The tutorial paces itself responsibly, introducing one new node type per puzzle cluster rather than front-loading a manual. That is rarer than it should be in this genre. Where the game loses momentum is in the mid-to-late stretch. Puzzle difficulty plateaus for long runs, and several contracts feel like light reskins of mechanics you already mastered. The node palette grows wide but not always deeper, and the AI opponent clients provide little friction once you understand the dominant pipeline patterns. Completionists chasing the optional research upgrades and cat cosmetics will squeeze another few hours out, but the last quarter of the campaign does coast. The mod ecosystem on Steam is small, which limits replayability beyond the base content. For strategy and sim players specifically, think of it as a systems-puzzle game rather than a management sim. There is no resource scarcity or macro pressure, no late-game crisis you have to survive. The satisfaction is cleaner and narrower: find the optimal graph topology, hit the accuracy threshold, get paid, unlock a cat hat. If you want complexity per hour, this is not a deep well. If you want a focused, well-tutorialized puzzle experience with genuine educational upside and a 10-to-15 hour runtime, it delivers that reliably. The 91-percent positive rating on over 8,000 Steam reviews reflects a player base that knew what they were buying and was not disappointed. Diego, Scout Team