Datasets

Define test cases as YAML files with inputs, cached outputs, and ground truth for workflow evaluators

Datasets are YAML files that define test cases for your workflow evaluators. Each file contains a workflow input, optionally a cached output, and ground truth values that evaluators can check against.

Datasets live in tests/datasets/ within your workflow directory:

src/workflows/
  blog_generator/
    workflow.ts
    tests/
      evals/
        workflow.ts
        evaluators.ts
      datasets/
        happy_path.yml
        edge_case.yml
        stripe_blog.yml

Basic Dataset

The simplest dataset has a name, input, and cached output:

tests/datasets/basic_input.yml

name: basic_input
input:
  values:
    - 1
    - 2
    - 3
    - 4
    - 5
last_output:
  output:
    result: 15
  executionTimeMs: 100
  date: '2026-02-13T00:00:00.000Z'

The last_output contains the workflow’s cached result. When you run evals with --cached, the framework uses this output instead of re-executing the workflow.

Dataset with Ground Truth

For evaluators that need expected values, add a ground_truth section:

tests/datasets/stripe_blog.yml

name: stripe_blog
input:
  topic: "Stripe the payment processor"
  requirements: "Include a link to https://stripe.com/en-gb/pricing"
ground_truth:
  notes: "Known good case"
  evals:
    length_of_output:
      min_length: 100
    evaluate_topic:
      required_topic: "Stripe the payment processor"
    evaluate_content:
      required_content: "https://stripe.com/en-gb/pricing"
last_output:
  output:
    title: "Stripe: The Modern Payment Processing Platform"
    blog_post: |
      Stripe has revolutionized online payment processing...
  executionTimeMs: 5000
  date: '2026-02-16T00:00:00.000Z'

Dataset Fields

Field	Required	Description
name	Yes	Unique name for this test case
input	Yes	The workflow input (must match your workflow’s input schema)
ground_truth	No	Expected values for evaluators to check against
last_output	No	Cached workflow output (used with --cached flag)
last_eval	No	Cached evaluation results from the last run

Ground Truth Structure

Ground truth supports global values and per-evaluator overrides:

ground_truth:
  # Global values — available to all evaluators
  notes: "Known good case"
  min_length: 100

  # Per-evaluator overrides
  evals:
    evaluate_topic:
      required_topic: "Stripe the payment processor"
    evaluate_content:
      required_content: "https://stripe.com"

When an evaluator runs, the framework merges global ground truth with evaluator-specific values. Per-evaluator values override globals with the same key. In your evaluator, access them through context.ground_truth:

( { output, context } ) =>
  Verdict.gte( output.blog_post.length, Number( context.ground_truth.min_length ?? 100 ) )

Managing Datasets with the CLI

Listing Datasets

output workflow dataset list blog_generator

Generating Datasets

You can generate datasets from scenario files, trace files, or production traces:

# Generate a dataset from a scenario file
output workflow dataset generate blog_generator my_scenario --name new_case

# Generate from a trace file
output workflow dataset generate blog_generator --trace path/to/trace.json

# Download recent traces from S3 and generate datasets
output workflow dataset generate blog_generator --download --limit 10

Generating datasets from traces is useful when you want to test against real production inputs. Download traces from S3, pick interesting ones, and the CLI creates dataset YAML files with the input and output already filled in.

What’s Next

• Running Eval Workflows — Wire evaluators into an eval workflow and run them from the CLI
• Workflow Evaluators — Writing evaluators with verify(), Verdict helpers, and judge functions