CatCraft Anti-Cheat A1: A Server-Side Multi-Signal Corroboration Anti-Cheat System for Counter-Strike 2 (Technical Report)

Cheat detection in first-person shooter (FPS) games has long relied on client-side anti-cheat and black-box machine-learning models, while dedicated servers are often reduced to replay storage and manual review. This report presents CatCraft Anti-Cheat A1 (v0.6.0-b7, CCAC-A1), a server-side multi-signal corroboration anti-cheat system deployed on a Counter-Strike 2 (CS2) dedicated server. The data plane combines an engine ProcessUsercmds hook, server-side entity ground truth, game events, client cvars, and network statistics, and implements 19 detection modules with 25 independent judges (plus one aim sub-judge). Three engineering mechanisms are proposed: (1) three-tier punishment with sub-module corroboration, where a conjunction judge may only escalate to a permanent ban when all bound sub-modules confirm; (2) kinematic corroboration using event-window vs player-baseline variance, plus an input-output causal consistency sub-judge based on mousedx/mousedy; and (3) stability-first online protection against client message flooding. The system has been running on a live community server in DryRun mode, backed by a MySQL evidence store and per-map SourceTV demos. We report the design, observed judge behaviour, false-positive corrections, and limitations. This is a technical report based on preliminary deployment experience on a single server; it is not a controlled experiment and makes no claim about detection or false-positive rates. This record contains the English technical report (primary) and the Chinese original (additional file). It is not a peer-reviewed publication.