Hanzo

Scores

Evaluation scores from feedback, graders, and review — attached to traces, observations, and sessions.

Scores

A score is a judgment attached to a trace, observation, or session: a thumbs-up from a user, a grade from an automated evaluator, or a rating from a human reviewer. Scores are the measurable output of evaluation. Served by Hanzo O11y, tenant-scoped by org.

Anatomy of a score

Every score has a data type and a source, and points at what it grades. Its shape is governed by a score config referenced through configId.

FieldDescription
nameThe metric, e.g. helpfulness, hallucination
valueNumeric value (also used for boolean 0/1 and category codes)
stringValueThe category label or free-text value
dataTypeNUMERIC, CATEGORICAL, BOOLEAN, or TEXT
sourceAPI (feedback), EVAL (graders), ANNOTATION (review)
traceId / observationId / sessionIdWhat the score is attached to
commentOptional reviewer note

Where scores come from

  • API — end-user feedback (thumbs up/down, star ratings) sent from your app.
  • EVAL — automated graders, including LLM-as-judge, produced by experiment runs.
  • ANNOTATION — human review through annotation queues.

How is this guide?

On this page