Can a factually correct answer still get flagged as ungrounded?

Yes. Groundedness checks whether an answer is supported by the documents it was given, not whether the answer happens to be true, and those are different questions.

An agent can land on a correct answer using what it already knows, with nothing in the retrieved documents backing it up. From the classifier’s side, that answer has no support to check against, so it fires even though the content is right. That’s intended: an answer grounded in nothing works right up until the model’s prior is wrong on a similar question, and then it fails the same way, with nothing in the trace explaining why.

The classifier can’t tell you an answer is false, only that it can’t find support for it in the documents it was given. A true fact the agent got from a tool call can be flagged too, since tool results aren’t read as documents.

keep reading

More on this.

Can Tessary's groundedness classifier check a very long answer?Up to a point. It reads the answer and its documents in one pass capped at 8,192 tokens; an answer long enough to leave no room for a document isn't scored at all.answer →Does groundedness check every LLM call, or just the final answer?Neither. It checks every model call at a call site shaped rag_answer, summarize, or extract, wherever it sits in the trace. Planning and tool-choice calls aren't checked.answer →Does the groundedness classifier check whether the context itself is correct?No. It checks whether the answer is supported by the documents it was given, not whether those documents are 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 →

Two ways to run Tessary.

Tessary is an open-source agent reliability platform. Cloud and self-hosted run the same workflow on the OpenTelemetry traces your agent already emits.

Tessary Cloud

We host it for you. Send your first trace with nothing to deploy and no model key.

what's includedper organization
traces
10,000 per calendar month
stored trace data
1 GB
retention
30 days
model credit
$10, one-time, for triage and root-cause analysis
credit card
not required
start for free

Self-hosted Tessary

Run the open-source code on your own infrastructure with one command. Add your own model key for triage and root-cause analysis.

Self-host Tessary for me by following https://github.com/tessaryai/tessary/blob/main/setup.md

docker compose -f oci://docker.io/tessaryai/tessary:compose up -d -y