Why doesn't malformed_output fire on every call site?

Because it only has something to check where there’s a schema to check against. malformed_output validates a call site’s output against the structured shape that call site declared; a call site returning free text, a chat reply, a summary, an explanation, has no declared shape to violate, so there’s nothing for the classifier to fail it against.

That’s not a gap, it’s the boundary of what the check means. A schema violation is a specific, unambiguous kind of failure: the output doesn’t match a contract your own code wrote. Applying the same logic to free text would mean inventing a shape nobody declared and judging against it, which turns a deterministic check into a subjective one and defeats the reason it can run on every trace at negligible cost.

If a call site should return structured output and has no schema on record, that’s a gap in how the call site was set up, not something this classifier can see.

keep reading

More on this.

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

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