Evals and datasets

One LensClient creates the observer and reporter used by an evaluation run.

import { agentEvalTarget, runEvalSuite } from '@anvia/core/evals'
import { LensClient } from '@anvia/lens'

const lens = new LensClient({ serviceName: 'support-evals' })
const observer = lens.observer()
const reporter = lens.evalReporter({
  includeMetadata: true,
  includePayloads: false,
})

try {
  await runEvalSuite({
    name: 'support-regression',
    cases,
    target: agentEvalTarget<string>({
      agent,
      request: ({ input }) => ({ prompt: input }),
    }),
    metrics,
    reporters: [reporter],
  })
  await lens.flush()
} finally {
  await lens.close()
}

Attach observer to the evaluated agent. The reporter can correlate metric results with its traces; onMissingTrace decides whether an uncorrelated result is emitted, ignored, warned about, or rejected.

Read a managed dataset

const datasets = lens.datasetClient()
const dataset = await datasets.getDataset<string, string>({
  name: 'support-cases',
  version: 'v2',
})

The client reads published immutable versions and paginates automatically. Omitting version selects the latest published version. Pin a version in CI so repeated runs evaluate the same cases.

The public client is read-only: dataset drafting, publishing, archiving, comparison, and quality-gate configuration live in Lens.