Skip to main content

Provisioned private-preview workspace required

Review an agent trace

Find one staging run, inspect the evidence around a decision, and record what should happen next.

Expected outcome

A reviewer can identify the relevant run, explain the failure or risk, and choose whether it should become a reusable test case.

Tempera is in private preview and provides private APIs across its product areas. Access is arranged through onboarding for an agreed evaluation, with product-specific credentials, endpoints, and environment details supplied directly to your team. API maturity ranges from preview to research, and no shared cross-product endpoint is implied.

Before you start

  • A provisioned Palette environment
  • A staging Palette trace; if it is associated with a Tempo session, an onboarding-configured correlation reference checked against both records
  • Read access to the agreed workspace

Steps

  1. 01

    Find the staging run

    List traces for the provisioned workspace and narrow the result to the workflow, time, or release under review.

  2. 02

    Read the decision in context

    Inspect the action sequence and the input and output that matter to the decision. Avoid treating a single final message as the whole run.

  3. 03

    Record a clear outcome

    Mark the behavior as acceptable, failed, or needing escalation. Add a short reason another reviewer can understand without repeating the investigation.

  4. 04

    Choose the next use

    If the case represents behavior that should remain stable, add it to the evaluation plan for the next agent version.

What success looks like

  • The reviewer can identify the run and the decision point.
  • The outcome and rationale are understandable to another team member.
  • The team knows whether the case should inform a later evaluation.

Canonical example

Use only after private access is ready

List and open a trace

TypeScript · private SDK surface

import { TemperaAuth, createTemperaClient } from "@tempera/sdk";

const auth = new TemperaAuth({
  issuerUrl: process.env.TEMPERA_ISSUER_URL!,
  apiKey: process.env.TEMPERA_API_KEY!,
});

const client = createTemperaClient({
  auth,
  environment: "staging",
});

const traces = await client.palette.listTraces({
  tenant_id: "org_example",
  project_id: "project_example",
  limit: 20,
});

const trace = await client.palette.getTrace({
  tenant_id: "org_example",
  trace_id: "trace_example",
});

This example reads Palette traces; it does not derive a relationship to a Tempo session. Use only an explicit reference configured during onboarding when that relationship is part of the evaluation. Replace the inert example identifiers with values from your private workspace.

If something is blocked

  • If no run appears, confirm the workspace and time range before changing permissions.
  • If sensitive fields are redacted, ask the workspace owner whether the evaluation requires approved access to those fields.
  • If two reviewers disagree, capture both reasons and agree on the expected outcome before turning the case into a quality check.