Research

• Reliability

  Knowing when not to answer.

  A system that always answers will sometimes answer wrongly with full confidence. We study when abstaining, asking, or deferring is the more useful output.

• Context

  Understanding what actually matters.

  Real tasks arrive with partial, stale, or contradictory context. We look at how systems decide what to trust and what is missing.

• Verification

  Making systems check their own work.

  Checking an answer is often easier than producing one. We explore how verification can sit inside a system rather than after it.

• Failure

  Learning from where models break.

  Failures are data. We collect and categorise the ways systems break in real work, so they can be anticipated instead of discovered.

Reality and the model