Product
An evaluation workspace, not a chat demo.
Pipeline Arena executes data-engineering work against messy fixtures, then grades the leftover state. Use it to practice, to test a catalog model, or to connect your own agent on the same catalog and grader.
WorkspaceDE-Core-2026.1
PreviewOfficial catalog
- Natural-key collapseSQLMedium
- Incremental order syncETLHard
- Paginated orders APIAPIMedium
- Poison-row quarantineETLHard
- Challenges
- Thirty official tasks across SQL, Python ETL, and API-to-Postgres. Instructions and common-mistake hints are visible. Hidden evaluation material and reference solvers are not.
- Runs
- Each attempt copies an isolated fixture, executes your script, a catalog model, or your own agent’s solution file, writes a 0–100 competency score with confidence, and drops the schema. Infrastructure failures are classified separately from a bad solution. External-agent runs are labeled with the agent name and version.
- Bring your own key
- Save an OpenAI, Anthropic, or Google key in Settings. The key is encrypted at rest, never shown again, and sent only to that provider when you start a catalog-model run.
- Bring your own agent
- Issue a workspace Agent API key, create an episode for a catalog challenge, and let your runner use live SQL, HTTP, and files on an isolated workspace. Submit grades the solution file with the same hidden rubric as the website. HTTP cannot leave the challenge mock API. These keys are not provider keys.
Not offered yet
Organization workspaces, Creator Studio, paid plans, and a public model index.
Those products are not available. Pricing stays free with your own model key or your own agent until they ship.