Replay evaluation cases on a candidate
With feedbackCaseId, prepare conditional guidance from that kept correction and test it with the Agent’s newest kept cases, always including the source. Do not combine feedbackCaseId with candidate or caseIds. This does not publish. Requires conversation read access. Voice Agents use a spoken-content rehearsal with an independent model (evaluationMode voice_text), not the live voice model, audio, interruptions or telephony. A person must review its applicability to calls before approval. Otherwise measure a candidate version on its kept evaluation cases before publishing it: the WhatsApp Agent rewrites each case's reply with the candidate's instructions, procedures, integrationProcedures, model and reasoningEffort (omitted ones stay as the live Agent has them), without running any integration tool (each answers with what it answered in the original conversation) and without sending anything; then a judge model compares each new reply with what the team expects. Replays every kept case, or caseIds, at most 50. It calls paid models, so it needs confirmCost: true. Follow it with getEvalRun; with the Agent's settings.evalGate, updateAgent publishes exactly this content with evalRunId. EVAL_CASES_UNAVAILABLE: no kept case to replay. Not an AI tool.
- Scope:
raiz:agents:write; some inputs also needraiz:conversations:read,raiz:conversations:write - Credentials: an API key, a client-credentials token or a person's access token
- Effect: writes
- Limits: request body up to 512 KB, response up to 2 MB
- CLI:
arbol raiz eval-runs create· not available on MCP
https://api.getarbol.com/v1/raiz/eval-runsAuthorization
Request body JSON required
Returns
200
Successful operation.
404
NOT_FOUND: The requested operation does not exist.RESOURCE_NOT_FOUND: The requested resource does not exist in this organization.
409
EVAL_CASES_UNAVAILABLE: The Agent has no kept evaluation case to replay, or a named case is not one of its kept cases.INVALID_STATE: The resource is not in a state that allows this operation.
422
AGENT_PROCEDURES_INVALID: A procedure is incomplete, references a tool, document or procedure the Agent does not have, or breaks the structured step rules.issuessays where.AGENT_SETTINGS_INVALID: The settings do not apply to this Agent's modality, or two settings contradict each other.INTEGRATION_PROCEDURE_UNAVAILABLE: A procedure does not exist in this organization's connections, Raíz cannot run its connection, or the Agent is not a text Agent.
502
OUTCOME_UNKNOWN: The operation may have been applied; check its result before repeating it.
503
UPSTREAM_UNAVAILABLE: The upstream service could not complete the request.FEEDBACK_PROPOSAL_UNAVAILABLE: A usable improvement could not be prepared. The Agent has not changed. Try again later.INTEGRATIONS_UNAVAILABLE: The organization's integrations cannot be read right now. Try again shortly.