Troubleshoot inference

Inference failures never turn missing interpretation into a successful result. Fix the provider, then start a fresh analysis attempt for a new evidence-bound receipt.

Inference not configured

Check that provider and model are both set. Gemini also requires GEMINI_API_KEY; Ollama requires a reachable base URL. Restart web and worker, then run Check inference.

Authentication failed

Replace the server-side Gemini key and restart Payne. The Connection response must never return the key.

Model unavailable

Run ollama list on the Ollama host. Pull the configured model explicitly or change PAYNE_INFERENCE_MODEL, restart Payne, and rerun the check.

Timed out or stopped at a limit

Inspect provider health and the failed receipt. No partial conclusion is used. Defaults are 6 steps, 90 seconds, 16,000 input tokens, 4,000 output tokens, 65,536 tool-result bytes, and 20 tool calls.

Tool or structured output could not be verified

Unknown evidence references, unsupported actions, authority claims, and invalid structured output fail validation. Preserve the deterministic summary, restore a compatible provider/model, and start a fresh attempt.