MiniPlay

The model self-trains. AUC went 0.975 → 0.9944 across generations without rewriting the feature set. Near-perfect separation. Engagement beats identity.

Gen 1–3: 4 separate detectors, each calling quarantine directly. No coordination. Race conditions. False-positive risk at every path. Gen 4: detectors stamp a dirty flag only. One unified verdict engine reads the recomputed score and applies all policy gates exactly once. One write path. Auditable. Reversible.

Gen 4 of MiniPlay's anti-sybil/bot model has been training on ~40M games played. KS separation: 0.982.
Claim-gate precision: 98.4%.
Recall: 92.8%. 310K+ accounts scored continuously. Here's a bit more of its progress 🧵

Training my brain on @MiniPlayStudio and hit level 2! Join with code CTWJFJUM for bonus XP. miniplay.studio #playonminipay #miniplay

The detection stack — four layers, self-improving, built on real production data from one of the hardest bot environments out there — is becoming its own thing. If you're building something that needs to know it's talking to a human, we're working on making this available. More soon.

This isn't a gaming problem. Grants. Airdrops. Waitlists. Content monetisation. Any system distributing value to humans faces the same non-human inception — bots, sybil farms, AI-generated personas competing against real people. We just happened to have one of the sharpest training grounds on the planet.

Big update on MiniPlay's Fair Play system. What started as bot/sybil detection for a crypto gaming platform is quietly becoming something much bigger. Here's the evolution 🧵

If you're building a reward economy onchain, the uncomfortable truth is this: the default composition of your user base will be adversarial until you make it expensive not to be. FairPlay's quick technical breakdown — training pipeline, feature weights, live cohort scores — is at miniplay.studio/fairplay Every reward that doesn't reach a bot reaches a real player. That's the product. Live within @MiniPay 🤝

8,692 accounts currently score ≥0.7 on the ML model but haven't triggered any rule-based flag yet. Those are the Gen 4 training labels — accounts the model suspects but the rules haven't caught. The next sweep starts there.

Announcing FairPlay — MiniPlay's ML-powered integrity system, now live in production. A logistic regression model trained on 281,036 real accounts. Test AUC 0.971. Active at every reward claim. Here's what it does, how we built it, and what it found. 🧵

cc @minipay @Celo

7/ The principle: anti-cheat shouldn't be a black box that swallows real players. Every detector logs its reasoning, every pause is reviewable, and when we get something wrong, we ship the fix the same day. Skill > Volume. That's the game we're running — every day, for every player who actually shows up. PS: this transparent thread with some specific details is intentional… 👀

6/ The second one was sneakier. A quarantine wave would wipe the top-100, then the surviving legit players appeared to "jump 100+ ranks" on the next refresh — and got quarantined too. A cascade. We added a guard: if 50+ accounts got paused in the last hour, the rank-jump rule sits out the next pass.