# Refs, inputs, outputs, and provenance ## Capability discovery `GET /capabilities` lists runnable lab versions. `GET /capabilities/{namespace}/{name}` returns the latest accessible version plus declared input and output schemas. `GET /compute-profiles` lists available CPU/GPU resources, limits, and credit rates. Only accessible lab packages in an executable registry state are returned. Private packages are isolated to their owner. ## References Use either: ```text namespace/name namespace/name@version ``` The service resolves an unpinned reference once, before execution, and persists the exact version and package digest. Pinned references are recommended for production. ## Inputs and units Input keys use the lab's declared public names, not internal module aliases. Values can be raw JSON: ```python inputs={"initial_cells": 10} ``` Or carry an explicit unit: ```python inputs={"temperature": {"value": 37, "unit": "degC"}} ``` Unknown inputs, missing required values, incompatible types, and unsupported units return structured `422` errors before credits are reserved. ## Outputs The result flattens declared lab outputs back to their public names: ```python result.outputs["population"] ``` Internal module aliases and wiring paths are not part of the public result contract. ## Provenance Run provenance includes the requested and resolved package refs, package SHA-256 digest, lab commit, scientific runtime version, compute profile, and execution provider. Runtime/solver configuration, model resolutions, and a random seed are retained when present in the submitted lab configuration.