Developers
Start with a dataset. Integrate at your own pace.
Use structured exports in your existing evaluation workflow. API development and self-hosted tooling are separate from the dataset purchase.
Use the packaged export
The 500-case proof package contains JSONL, original provenance, historical grades, a manifest, and an internal-use license. Keep reference answers out of model prompts and review the grading rules before interpreting results.
Run a public sample →Work with the development API
The repository includes bounded verification and evaluation APIs with operator-issued credentials. Managed API onboarding is not a public self-service product.
Keep API credentials server-side. Dataset checkout does not include hosted Runtime access.
Development workflow
POST /v1/projects POST /v1/proof-evals POST /v1/evals GET /v1/reports/:id GET /v1/failures POST /v1/evals/:id/remediation
Availability depends on the deployed environment, configured sources, and supported claim types. The repository quickstart documents authentication, idempotency, and current bounds.