I'm Geoff. My cofounder James and I built Cyberlete, a competitive gaming platform, and the part I'd most like feedback on is the anticheat, because we took a different approach than the rest of the industry.
Most anticheat ships a kernel driver that tries to block cheats in real time (Vanguard, EAC, BattlEye). It's invasive, it's an endless arms race, and it runs ring 0 code on your machine. We don't do any of that. Everything runs in user mode, and we are mouse, keyboard, security, metadata in real time to try to stop a cheat mid match. Additionally, We catch it and audit it after the fact.
Anticheat has also almost always been a closed, centralized black box. You get banned, you never find out why, and there's no real way to appeal. We wanted the opposite: a verdict you can actually inspect and verify.
How the detection works: the desktop app captures raw mouse and keyboard input and runs around 40 motor control detectors over the stream. Fitts' law timing, minimum jerk trajectories, entropy and spectral signatures of human versus synthetic movement, plus a 7 model ML ensemble. The detectors are compiled to both Python and Rust/WASM with byte for byte parity, so a verdict is fully reproducible. There are also about 14 user mode integrity scans (TPM and secure boot attestation, DLL cross validation, DMA device detection).
The compute that runs all of this doesn't just live on our bare metal servers. Players can opt in to contribute their own PC's spare compute, and that distributed network runs the scoring and validation for other people's matches. So the players genuinely run the anticheat. It sits on a custom L1 we built for exactly this, live on testnet today with Windows nodes, mainnet in progress.
Every verdict goes to a human reviewer. And every scored session, plus every node's contribution, produces a cryptographic proof committed to the chain, so two things can be verified independently after the fact: that the verdict is what we say it is, and that a node actually did the work it claims, without anyone having to take its word for it.
The proof layer is built on a STARK interface and that's the direction we're headed, but the proofs running today are Schnorr based, not full ZK STARKs yet. The migration is on the roadmap and the ABI is already in place. I'd rather say that up front than have someone find it.
I'm a Marine vet, self taught, and I wrote most of this stack myself, so I'd genuinely value people here trying to break it.
You can try it at cyberlete.net (Windows for now, Linux and mobile next).
Hi HN,
I'm Geoff. My cofounder James and I built Cyberlete, a competitive gaming platform, and the part I'd most like feedback on is the anticheat, because we took a different approach than the rest of the industry.
Most anticheat ships a kernel driver that tries to block cheats in real time (Vanguard, EAC, BattlEye). It's invasive, it's an endless arms race, and it runs ring 0 code on your machine. We don't do any of that. Everything runs in user mode, and we are mouse, keyboard, security, metadata in real time to try to stop a cheat mid match. Additionally, We catch it and audit it after the fact.
Anticheat has also almost always been a closed, centralized black box. You get banned, you never find out why, and there's no real way to appeal. We wanted the opposite: a verdict you can actually inspect and verify.
How the detection works: the desktop app captures raw mouse and keyboard input and runs around 40 motor control detectors over the stream. Fitts' law timing, minimum jerk trajectories, entropy and spectral signatures of human versus synthetic movement, plus a 7 model ML ensemble. The detectors are compiled to both Python and Rust/WASM with byte for byte parity, so a verdict is fully reproducible. There are also about 14 user mode integrity scans (TPM and secure boot attestation, DLL cross validation, DMA device detection).
The compute that runs all of this doesn't just live on our bare metal servers. Players can opt in to contribute their own PC's spare compute, and that distributed network runs the scoring and validation for other people's matches. So the players genuinely run the anticheat. It sits on a custom L1 we built for exactly this, live on testnet today with Windows nodes, mainnet in progress.
Every verdict goes to a human reviewer. And every scored session, plus every node's contribution, produces a cryptographic proof committed to the chain, so two things can be verified independently after the fact: that the verdict is what we say it is, and that a node actually did the work it claims, without anyone having to take its word for it.
The proof layer is built on a STARK interface and that's the direction we're headed, but the proofs running today are Schnorr based, not full ZK STARKs yet. The migration is on the roadmap and the ABI is already in place. I'd rather say that up front than have someone find it.
I'm a Marine vet, self taught, and I wrote most of this stack myself, so I'd genuinely value people here trying to break it.
You can try it at cyberlete.net (Windows for now, Linux and mobile next).
Ask me anything.