ML-based anti-cheat data collector and scorer for the OneLiteFeather network.
A Velocity plugin passively records player packets via PacketEvents, an ingest service stores them pseudonymised in a data lake (Parquet on S3/MinIO, catalog in PostgreSQL), curated pools feed model training, and a CPU-only scoring service publishes a per-player "likely cheating" score.
Corvus never sanctions players on its own; the score is a decision aid for staff.
| Path | Content |
|---|---|
proto/ |
Protobuf schemas shared by Java and Python |
plugin/ |
Velocity capture plugin (Java) |
services/ |
Python services: ingest, enrich, score |
ml/ |
Python: evaluation, training, pilot analysis |
ui/ |
Management UI (Nuxt, Material Design 3) |
deploy/ |
Docker Compose (pilot/standalone), Helm |
Requirements and architecture decisions (ADR-CORV-*) live in the OneLiteFeather Outline wiki.
Apache License 2.0, see LICENSE.