← All projects
GroundTruthPlanned

Can an AI answer policy questions, and can I prove it?

Planned: an AI search assistant tested against known answers, so its accuracy is measured, not assumed. Work has not started.

  • Retrieval (planned)
  • Evaluation (planned)

Planned. Work has not started: there is no code or repository yet.

Overview

The question this project is meant to answer: can an AI assistant answer policy questions, and can that be proven?

The plan is a search assistant that is tested against questions with known answers, so its accuracy is something measured rather than assumed.

What it will cover

  • A set of questions with known answers
  • Retrieval over policy documents
  • Measured accuracy, reported together with the method used

Content checked against the local project folder, which is empty on 24 Oct 2026.