Read provisioning results

Inspect the job, generated sets, skipped work, and partial failures.

Read a provisioning job with the ID returned by creation:

http
GET /v2/provisioning/jobs/:id
X-API-Key: YOUR_API_KEY

Job fields #

FieldMeaning
id, organizationId, engineIdJob identity and owning organization and engine.
statusTagged Running, Completed, or Failed state. Failed includes message.
kindsOutcomes for requested glossary, ruleset, and brandVoice work. Unrequested kinds can be absent.
sourcesOptional source-reading progress or report.
websiteFirst supplied website URL; absent for text-only input.
createdAtISO creation timestamp.
finishedAtISO completion timestamp, present after settlement.

Per-kind states #

Inspect every requested kind, including when the top-level job is Completed:

kinds.<kind>._tagFields and meaning
PendingStill working. Optional step is reading, generating, or saving; startedAt identifies when that step began.
DonesetId identifies the attached set; itemCount is the number of generated items.
Skippedreason explains why the kind was left unchanged.
Failedmessage explains why the kind could not be filled.

For example, this result excerpt contains a generated glossary and a failed ruleset:

json
{
  "glossary": { "_tag": "Done", "setId": "YOUR_GLOSSARY_ID", "itemCount": 12 },
  "ruleset": { "_tag": "Failed", "message": "Could not generate rules from the supplied material." }
}

Each kind is saved independently. Successful guidance remains available when another kind fails. An engine can also gain an attachment while generation is running, causing an empty-only fill or brand-voice fill to be skipped.

Inspect the generated records #

Open the engine in Studio and follow the attached glossary, ruleset, or brand voice. For API reads, use:

  • GET /v2/glossaries/:id/terms for glossary entries.
  • GET /v2/rulesets/:id/rules for rules.
  • GET /v2/brand-voices/:id/variations for voice variations.

Use the corresponding setId, and follow each list's pagination. Check terminology, applicability by language, non-translatable names, and any assumptions the model added. An item count measures records created, not their correctness.

For a failed kind, use retry and recovery. Preserve completed sets instead of creating a replacement engine to repeat all work.