What is Tessary's groundedness classifier?

It’s a built-in classifier that checks, for every LLM call in a trace, whether what the agent said is actually supported by the context it was given. It runs at negligible per-event cost on every trace, the same as Tessary’s other built-in classifiers, so nothing needs to be sampled to catch it.

What makes it worth having is the shape of failure it targets: an answer that asserts something its context doesn’t support while reading as fluent and confident. That’s the hallucination that’s hard to catch by eye, because nothing about the sentence looks wrong.

A flag becomes a finding, not an alert. Triage rules on whether it’s a real issue, and an ungrounded claim that keeps recurring becomes one case with the triggering traces attached, instead of a pile of separate flags nobody can act on.

keep reading

More on this.

Can a factually correct answer still get flagged as ungrounded?Yes. Groundedness checks whether an answer is supported by its context, not whether the answer is true, and those are different questions.answer →Does groundedness check every LLM call, or just the final answer?Every LLM call. It scores each call against its own context as the call happens, not just the last message a trace sends back to the user.answer →Does the groundedness classifier check whether the context itself is correct?No. It checks whether the answer matches the context given, not whether that context is accurate, so a wrong or stale source can still score grounded.answer →Everything on tessary groundednessThe definition, and every question answered under tessary groundedness.concept →

Send us the traces you already emit.

book a demo