persistent_id, as an eval grades it. Read it before you correct it with Set ground truth.
document_packet_id is the dataset copy’s id: the one Upload to dataset returns, or the one GET …/evals/{eval_id}/accuracy/ lists in by_document_packet.
Omit version to read the workflow’s current version, the one an eval grades by default; pass a version_id from List workflow versions to read another. node_id is the extract node’s id in the workflow graph, and you need it only when the version has more than one extract node; without it such a version answers 422 EXTRACT_NODE_UNRESOLVED. The response names the version_id and node_id it was read from. ground_truth is null until ground truth is first set for that version and node.
Any key with the read scope can read. Every 404 NOT_FOUND names what is missing in params.resource_type: document_packet for an unknown packet or one in another organization, and workflow_version for an unknown version. A production packet answers 409 NOT_IN_DATASET: only a dataset document packet holds ground truth.

