CODING_FAILED — Coding step failed mid-run

Summary

Stage 3 of the pipeline — assigning codes to verbatims — failed. The job’s data.error field has the specific cause. Any verbatims coded before the failure are still available and only those were charged.

HTTP status

500 Internal Server Error. Standard envelope.

Example response

{
  "success": false,
  "error": {
    "code": "CODING_FAILED",
    "message": "Coding failed at batch 7 of 12. 600 of 1000 responses were coded successfully.",
    "request_id": "req_01HXJZK4ABCDEF",
    "doc_url": "https://docs.surveycoder.io/errors/coding-failed"
  }
}

Why this happens

An upstream LLM provider hit a transient outage.
A specific verbatim triggered a content-policy refusal that wasn’t retried in time.
The codebook was edited mid-run (rare — usually blocked by INVALID_STATE).

How to fix it

Fetch the job: GET /v1/jobs/{job_id}. Partial results are in data.partial_result.
To finish the remaining verbatims, call POST /v1/code again on the same question — the API skips already-coded responses by default.
If the same job keeps failing at the same batch, isolate the offending verbatims (data.failed_response_indices) and either reword or skip them.
Email support@surveycoder.io with the request_id if you see persistent failures across different questions.