mindsweeper

Minesweeper has stayed almost unchanged for over 30 years because the core loop, read the numbers, deduce the safe tile, is genuinely good. Most variants change the skin, not the puzzle. Mindsweeper changes the puzzle itself: the board actively works against the plan you just made. You’re not just solving a fixed layout anymore, you’re solving a layout that reacts to your own progress, which means memorization and static solving strategies stop working and real-time adaptation starts mattering. Every board size was independently tested with thousands of simulated games using an actual deductive solver, not a guess, to make sure the drifting-mine twist stays a fair, winnable challenge and doesn’t turn into random luck.# AI Tools Used Claude (Claude Code) and Game Designer Mind. # How AI was integrated into the development process Game Designer Mind was a real, ongoing design partner across the whole build, not a single one-off prompt. Mind conceived the core mine-relocation mechanic and revised it across several real rounds: a first version with open questions, a second version rewritten after real playtest data showed the original probe budget was mathematically unwinnable, and a third version that folded in pushback from our own testing (dropping an unnecessary rule, keeping first-click safety, keeping the standard mine-counting convention). Mind also designed the four-tier sizing system and proposed the daily-challenge groundwork that shipped in this version. Claude handled implementation: writing the game code, then independently verifying every board size with a real deductive solver playing thousands of simulated games, not just eyeballing it. Two separate adversarial testing passes were run against the finished game specifically to try to break it, and one real exploit was found this way (mass-flagging could freeze the drifting-mine mechanic entirely) and fixed, then re-verified with the solver to confirm the fix held.