LinedatLinedat
Beta

Capabilities

The Linedat capability catalogue, anchored to regulation and with their limits

AI Governance

Bias Detection with p-value

We do not tell you there is bias: we prove it, with a p-value.

Framework / standard: ISO/IEC 38507 §6.4 · EU AI Act Art. 15 · ISO/IEC 42001 A.6.2.6

Saying "we believe there is no bias" does not survive the first question from an expert. Linedat does not claim the absence of bias: it measures it with a statistical test and gives you a p-value.

The monitor analyses human acceptance of AI-generated content by group (action type, provider, entity type), detecting disparity between groups and drift against a baseline.

A z-test, not a gut feeling

The analysis uses a 2-proportion z-test with a recent window against a 30-day rolling baseline per group, and three gates (significance p<0.05, effect size and minimum N) before issuing a signal. It detects drift, not just level. When the threshold is crossed, a drift alert is generated and a snapshot is persisted as evidence.

Snapshots to track the trend

Each analysis persists a snapshot, so you can see the evolution of fairness over time, not just today's snapshot. It is traceable evidence for management review and for an AI audit.

The limits (what we do not claim)

This is a rigorous statistical proxy of human acceptance across AI USAGE categories, NOT fairness over protected groups of people — that is the work of the FRIA (AI Act Art. 10), and we keep them deliberately separate. The monitor generates alerts and persists snapshots; automatic incident creation on drift is a follow-up item and is not active today. Groups with a sample below the minimum do not produce an alert (conservative behaviour).

How Linedat helps

Linedat turns bias into a statistical signal with p-value and confidence intervals — the difference between "we saw no bias" and "we measured and proved it".

FAQ

Respuestas sobre implementación y capacidades

No. It measures, with statistical rigour, human acceptance of AI content by usage categories. Bias over protected groups of people is evaluated in the FRIA (AI Act Art. 10); they are distinct things and we keep them separate.

Govern your data and your AI with Linedat

Connect your sources and in minutes you have catalog, quality, lineage and AI governance. Free to start.