I worked this summer in enabling a sudoku style webgame (Kludoku) and I designed it to self-fund with ad revenue (basically banner and modal during win screens), but it has been rejected twice by Adsense for light content.
Does anyone have a suggestion on how to get it approved? or it is a lost case?
Screenshots & trailer:
This started as a vacation project that got completely out of hand. I was in Spain this summer, sick for most of the trip, with a stack of murder-mystery logic puzzle books as my only entertainment. I loved them — but a book is finite. You solve the eighty cases and it's over, and the paper can't do the one thing I kept wishing for: let me test a hunch and walk it back. So between naps I started writing a generator, as an exercise, to see if I could build cases that weren't limited by what fits on a printed page.
The technical rabbit hole turned out to be the fun part, so let me lead with that.
The engine. Every case is procedurally generated and then proved to have exactly one logical solution before it's published. The solver is a bitset constraint engine: each suspect's possible squares are a bitmask, testimonies compile into unary and binary constraints, and a backtracking counter verifies uniqueness. But uniqueness isn't enough — a puzzle can be unique and still miserable — so a second, human-style solver replays the case using only propagation plus "what if?" assumption rounds, and the number of rounds it needs becomes the difficulty grade. A beginner case must solve with zero assumptions; an expert case is deliberately selected to need several. If a generated case doesn't land in its difficulty band, it's thrown away and regenerated.
The liar. The harder cases add an accomplice: one suspect whose every statement is false. Generating those means proving uniqueness across the space of "which testimony set do you trust" — the engine verifies that exactly one choice of liar yields exactly one solution, so working out who lies is itself a deduction, never a guess. Some cases also demand the motive, not just the name.
Even the furniture is verified. Decorative scenery is placed by the generator as real obstacles — and every piece is auditioned: placed, re-solved, and kept only if the measured difficulty doesn't move. A bookshelf is never allowed to accidentally turn a hard case into a medium one.
The backend, because I couldn't help myself. It runs on a small fleet of my own boxes across two continents — Postgres with replicas, geo-aware routing on a free Cloudflare plan (a hostname trick, since free plans can't route by geography), self-healing deploys shipped as versioned packages through a private npm registry, and ~90 automated tests that gate every release. Massively overbuilt for the traffic it gets, which is honestly part of the hobby.
How it actually plays: you get a floor plan, a cast, and one testimony per suspect — "she was in the same column as the piano," "he was exactly three rows south of the Gardener." Place everyone, exactly one per row and never two in a column, until only one person could have been alone with the victim. Tap a testimony and the board spotlights what it rules out. There are pencil marks, and a "what if?" mode that lets you place a whole hypothetical line and commit or discard it — the thing paper couldn't do.
The numbers, for the curious: five shared daily cases every day (same boards for everyone, so times are comparable), from a 5×5 beginner up to a 10×9 expert, across 23 hand-drawn crime scenes, playable in 15 languages (I'm Spanish, so that includes Catalan, Valencian, Galician and Basque). There are printable case sheets if you're a pencil person. It's free in the browser, no download, no account needed.
Play it here: https://kludoku.com
What I'd genuinely love from this sub: play a case or two and tell me what breaks or what feels unfair. Difficulty curve, clue phrasing, whether the accomplice cases feel like deduction or like a coin flip — anything. I ship fixes fast, usually same-day. Happy sleuthing 🔍
