SemanticDrift

Words are closer than you think. OCEAN sits right next to SEA, which is no surprise. But APPLE next to BLIGHT? MUSIC leading to SILENCE? SemanticDrift is a word puzzle built on that closeness. Two words are close when they're used in similar ways, turning up in the same kinds of sentences, like DOCTOR and NURSE. The game measures that as a distance you can see, and every guess has to land close enough to your last word to count. You're working toward a target word that starts out too far to reach directly. So you hop from word to word, each step close enough to be allowed, watching the distance shrink until the target is finally in range. Some paths are obvious. Most aren't. The best ones make you say, "Wait, that worked?" DAILY CHALLENGE One puzzle a day, with the exact same words for everyone. Compare your path on the Game Center leaderboards. A new puzzle drops at midnight. PUZZLE LIBRARY Thousands of curated word pairs across multiple difficulty levels. The easy ones build your intuition. The hard ones wreck it. FREE DRIFT A relaxed solo mode with the target word shown from the start. Choose a difficulty and chart your own route to it, a direct line or the scenic way. Add a timer if you want the pressure, or wander with no clock at all. MULTIPLAYER Best Path: Same start and target, entirely different routes. Build your path, then vote on whose logic is the best or weirdest. Speed Race: Throw words as fast as they come to you and see what sticks. WORD MAP Pick any word and see what lives nearby. Poke around its semantic neighborhood and discover connections you didn't know existed. FREE WITH NO ADS Play the daily challenge, plus three Puzzle Library games and three Free Drift games, every single day. No ads, no tracking, and no data collection. Not now, not ever. PREMIUM (One-Time Purchase) Unlimited puzzles. Host multiplayer games. Sync with iCloud. Zero subscriptions. WHAT'S ACTUALLY HAPPENING SemanticDrift runs on Apple's NaturalLanguage framework, using word embeddings: a machine learning technique that also powers predictive text and search. Every word gets a fixed position in a high-dimensional space, set by the company it tends to keep. Each hop you make is a nearest-neighbor search in that space, the same kind of lookup that RAG (retrieval-augmented generation) uses to find relevant text for an AI assistant. Here you run it by hand, one word at a time. That's why OCEAN is close to SEA, and why some paths will still surprise you. All of this math runs locally on your device, with no internet needed for solo play.