{
credential: {
name: "Provider Credential",
definition: "An API key for a model provider such as OpenAI or Anthropic. Encrypted at rest and never returned through the API."
},
dataset: {
name: "Dataset",
definition: "A CSV of real inputs. Each row becomes one test case."
},
prompt: {
name: "Prompt",
definition: "A versioned template with {{variable}} placeholders. Editing a prompt that has already been run creates a new version, so past results stay reproducible."
},
run: {
name: "Run",
definition: "One execution of a prompt against a dataset. Stores every output and the judge's scores."
},
response: {
name: "Response",
definition: "The model's output for a single dataset row, with the judge's reviews attached."
},
metric: {
name: "Metric",
definition: "An evaluation dimension with its own 1-5 rubric. The LLM judge scores every response against it."
}
}.freeze