MAJORMar 28
Bot Detection v2 (multi-signal model)
Sharper accuracy across our updated behavioural model.
Public, voteable, kept fresh. If you don't see your problem here, request it — we read every entry.
The map
Workstreams across time. Hover any station for details.
Detail
Live in production
Sharper accuracy across our updated behavioural model.
Per-server suspect sessions, names hidden, signals exposed.
D1/D7/D30 retention by acquisition source.
End-to-end snapshot in under 1 s per cycle.
Being built right now
Scrub back to the moment a bot wave hit. Replay the input signals frame by frame.
Small communities and large servers shouldn't share a single decision boundary.
Each score gets an explicit ± margin. No more false certainty.
Queued for this quarter
Detect bot farms reusing the same identifier fingerprint across multiple servers.
Reduce false positives during off-peak hours by adapting the baseline.
Distinguish honest reconnects from patterns used to evade detection.
Research, no commitment
Group flagged sessions by behavioural signature to attribute them to operators.
Better handling of marathon roleplay players so they aren't penalised like bots.
Calibrate the surge-ratio bonus against expected daily growth curves per server.