Layer Three: The Answer Stayed the Same. The Reason Didn’t.

Behavioural Audit – Layer Three - A practical guide to interpreting explanation stability in AI systems

Have you ever asked an AI system the same question twice and received the same answer… but the reasoning was different each time?

The decision may be stable. The reasoning may not be.

This guide introduces the third observable probe in behavioural audit: explanation stability.

What This Guide Will Help You Do

After reading, you will be able to:

• Recognise when explanations vary without the decision changing

• Identify shifts in reasoning, emphasis and justification

• Apply the Explanation Stability Test

• Understand how explanation can influence interpretation and oversight

• Treat explanation as behavioural output, not a fixed justification

Who It’s For

Practitioners, analysts, HR professionals, policy leads and managers using AI tools in real workflows.

What It Is Not

This is not a technical guide to explainability methods or model internals.

It introduces an observational method for assessing how reasoning behaves under repetition and constraint.

Also available as part of the Applied AI Guides bundle (Layers 1–3).

~30 pages