Inspiration
You line up a shot, you hit it, and someone types “aimbot” in chat. Cheating ruins competitive games, but so do false accusations. And the people who have to make that call, like moderators and tournament organizers, usually only have a clip and a gut feeling. I wanted to give them something better, real evidence they can check for themselves.
What it does
You upload a clip from any FPS or battle royale, and FairPlay gives you a suspicion score, a timeline of suspicious moments, and the frames that back each one up. It only needs the video. No game access, no anti-cheat driver.
Analyze: Upload a clip and watch the analysis run live. The report scores it on a scale from Clean to Suspicious to Likely to Confirmed. Evidence you can check: Every finding comes with a timestamp, the actual frames, the raw numbers, and a plain-English reason. For example, “a 115° view change in 100 ms.” A reviewer can look at it and decide for themselves. Tournament mode: Upload up to 10 clips and FairPlay ranks players by risk, so organizers know which clips to look at first. Trust Score: Every clip you analyze adds to a player profile. Clean play raises the score, flagged clips lower it, and it can bounce back over time. Cheat Academy: Four short lessons on aimbot, wallhack, speed hack, and recoil hack, plus a quiz to test your eye.
How we built it
The idea is simple: people have arms, and software doesn’t. A real flick speeds up, slows down, overshoots a little, and settles with a tiny wobble. A cheat snaps to the target in a frame or two and stays perfectly still. A Python analyzer built with FastAPI measures how the camera turns between frames and runs six checks for that kind of movement. The score comes from the video, not from AI. Gemini only writes the plain-English explanation and can’t change the result. The front end is Next.js, and it streams progress live as the clip is analyzed.
Challenges we ran into
Game HUDs and black bars made the camera look like it was barely moving, so I measured only the center of the screen. I also set all thresholds in milliseconds so clips from any frame rate are judged the same, and I report ultra-fast flicks as “at least” a speed instead of dropping them.
Accomplishments that we're proud of
A complete workflow where one clip becomes a shareable report, a tournament ranking, and a long-term trust profile. Every number traces back to frames a person can check. AI explains the result but never decides it, so the same clip always gets the same score. A modular detector. Adding a new signal is just adding a function and a weight.
What we learned
Good measurement matters more than adding more checks. And a result people can check, with timestamps and real frames, is worth more than a mystery number.
What's next for FairPlay
Calibrate the scoring on verified, labeled clips from several sources. Turn on the enemy-detection signals (target lock and reaction time) that are already built. Add field-of-view and HUD presets for each game. Add detectors for wallhack, speed hack, and recoil hack. The report is already organized by cheat type, so they have a place to land. Let reviewers click agree or disagree on each finding, and use those clicks as new labeled data.
