> For the complete documentation index, see [llms.txt](/llms.txt).
> A full single-fetch corpus is available at [llms-full.txt](/llms-full.txt).
---
title: "Experiments"
description: "Create experiments, trigger runs against a dataset, and read run results, failures, and scores."
last_verified: 2026-07-27
---


**Endpoints**: experiment CRUD and run management under `/api/v1/projects/:projectId/experiments`; run analysis (summary, failures, items, scorers, CI, scores) under `/api/v1/projects/:projectId/eval/...` — see the endpoint list below
**Auth**: `X-API-KEY` or `Authorization: Bearer <token>` (required)
**Error envelope**: 4xx/5xx return `{"error": string, "message": string, "code": int, "hint"?: string}`


An **experiment** runs a component, deployment, or prompt against a [dataset](/docs/api-reference/datasets.md) version and scores each result with one or more [scorers](/docs/api-reference/scorers.md). Each execution of an experiment is a **run**.

## Analyze a run

Once a run completes, a separate set of endpoints — nested under `/eval` rather than the experiment — surface its per-item results, failures, and scores.

## Next steps

* [Datasets](/docs/api-reference/datasets.md): the dataset versions an experiment runs against
* [Scorers](/docs/api-reference/scorers.md): define the scoring logic an experiment applies to each result
* [Quality Cases](/docs/api-reference/quality-cases.md): track a persistent failure discovered by a run