> ## Documentation Index
> Fetch the complete documentation index at: https://docs.gegentic.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Ethics & Bias

> Daily automated evaluation of AI agent outputs for bias, fairness, and ethical compliance, run on a sample of traces.

<Info>This feature is experimental and may not be available on every plan.</Info>

Each night, Gegentic samples a portion of the previous day's traces for every project and scores them against six LLM-judge criteria: **toxicity**, **demographic bias**, **sentiment consistency**, **fairness**, **misinformation**, and **privacy leakage**. Evaluation runs asynchronously and never blocks live traffic.

## What's on the report

* **Overall Ethics Score** — a 0–100 gauge; scores below 60 on any evaluator flag the trace
* **Traces Sampled** — how many traces were evaluated, out of the total for that period
* **Flagged Traces** — count of flagged traces, broken out by high-severity findings
* **Agents Evaluated** — how many agents in the project were covered
* **30-Day Trend** — a rolling chart with reference lines at 80 (good) and 60 (risk)
* **Evaluator Breakdown** — scores per evaluator (toxicity, bias, etc.), clickable to filter the flagged-traces table
* **Agent Ethics Scores** — per-agent comparison; agents scoring below 60 are called out for review

## Methodology

* Sampling runs daily at 03:00 UTC against roughly 17% of the prior day's traces
* Six LLM-judge evaluators score each sampled trace from 0–100
* A score below 60 on any evaluator flags the trace for review
* Evaluation is fully asynchronous — it never adds latency to or blocks production traffic