> ## Documentation Index
> Fetch the complete documentation index at: https://sen-0597ab10.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.
# Anomaly Score
## How Anomaly Score Is Calculated
The profiler evaluates these signals against
`behaviour-profiler.weights` :
1. max. horizontal movement
2. avg. horizontal movement
3. max. airborne sequence
4. CPS
5. APS
6. max. reach
7. Scaffold placement angle
8. max. yaw snap
9. Low rotation variance
10. Total anticheat VL
For ordinary over-threshold signals no score is added while t he value is at or below `threshold` Above the threshold the contribution rises gradually until it reaches the signals configured `weight: ` . Low variance works in the opposite direction: Variance below its threshold adds score but only when roatation average reaches `min-average`.
All signal contributions are added together. When the final value reaches `behavior-profiler.alert-score`, the profiler submits a `BehaviorAnomaly` suspicious sample. The normal `checks.behavior-anomaly.buffer`, `violation`, and `kick-vl` settings then control whether it alerts, adds VL, or executes punishment commands.
`behavior-profiler.alert-cooldown-seconds` prevents repeated BehaviorAnomaly submissions for the same player during the cooldown. Samples continue to be saved during that time.
## Interpretation and Safety Notes:
* Anomaly score is not a cheating probability.
* A high score means several configured signals exceeded their baselines. it does not establish intent or identify a specific hack.
* Always compare suspicious rows with ping, TPS, active effects, server mechanics, normal Anticheat flags, and staff observation.
* Do not automatically ban solely from a behavior label or one high anomaly sample.
* Labels are staff annotations and can be wrong. Use consistent review standards when creating a labeled dataset.
* If behavior data is exported or used for model training, remove data that is not necessary and follow applicable privacy requirements.