Fighting Game · Guides & deep dives

V2 Nous Anti-Cheat Classifiers

A guides & deep dives document in the Fighting Game library.

0sections1 minread
Markdown source
On this page

@v2/nous-anti-cheat-classifiers wires the V2 fair-play classifier suite to Nous:

  • V2 publishes labeled training examples on @oshun/event-bus.
  • @nous/training owns the training loop for smurf, win-trading, coordinated-throw, and geographic-anomaly classifiers.
  • @nous/safety hosts the Model Card at V2/docs/ai/model-cards/anti-cheat-classifier-suite.md.
  • The suite is review-only: classifier output may prioritize manual fair-play queues, but it cannot issue automatic discipline and cannot influence rollback.
  • The suite publishes a canonical v2.anti_cheat.review_published event through @oshun/audit-platform. V2 publishes the review-only anti-cheat record; Oshun retains and exports the shared forensic audit trail.

The V2 bridge returns the training plan and Model Card together so CI can assert event-bus topic coverage, complete classifier coverage, human review, appeal coverage, and rollback isolation.