Build and audit a small LLM evaluation with Inspect

Define one behaviour with a threat model, write items, validate a model grader against your own labels, sample repeatedly and report intervals and failure categories.

The exercise happens on Self-directed project (Inspect framework). Canonical link: https://inspect.aisi.org.uk/. Link check: link responded (5 Oct 2026). Opened from this app 0 times; opening is not counted as an attempt.

Supports: Evaluation-design evidence for alignment, safety and applied-AI evaluation roles.