Fair-Play Without False Accusations
A withhold-only ranked fair-play rule that protects the ladder without ever calling a person a cheater, backed by submission-time telemetry instead of self-reported clocks.
Research figure
Fair-Play Without False Accusations
The ranked rule withholds implausible rating gains instead of accusing anyone. Two boring checks, no docked rating, and a human kept in the loop for anything consequential.
Ranked practice only means something if the rating reflects real solving. But most "anti-cheat" turns into accusation: flagging people, docking points, and handling appeals nobody trusts. We wanted ranked integrity that protects the ladder without ever calling a person a cheater.
Withhold, do not accuse
The ranked fair-play rule is withhold-only. When a win looks implausible, the system declines to award the inflated rating gain. It never reduces a rating someone already earned, never marks an account, and never produces an accusation a human has to adjudicate. The worst case for a false positive is a single win that did not move the number: recoverable and quiet.
Two simple, boring checks
The rule combines two deliberately unremarkable signals:
We are intentionally not publishing the exact thresholds. The point is the shape of the rule, not a number to game.
Submission-time telemetry
Both checks lean on something we now capture at submission: how long the work actually took, derived from the session's own timing rather than a self-reported clock. That same signal feeds the broader integrity picture, not as a verdict, but as one input a human reviewer can weigh.
Why not auto-ban
Automatic punishment requires a trained, validated classifier and real labels. We do not have those yet, and shipping bans on an untested threshold would do more harm than the cheating it targets. Until then, the honest move is to withhold unearned gains, capture clean telemetry, and keep a human in the loop for anything consequential.