Troubleshoot Payne

Start with the observable symptom. Preserve durable evidence and run a read-only check before a reset or retry.

Symptom or code	Impact	First diagnostic	Recovery guide
`gms-unreachable`, `authentication-failed`, `permission-denied`, `mcp-unavailable`	Current catalog evidence is unavailable	Open Connection and rerun the capability check	DataHub setup
`capability-missing`, Capability limited	Some evidence-dependent actions are unavailable	Inspect the named missing capability	Connection limitations
Inference not configured, authentication, model, timeout, tool, or structured-output error	Agent interpretation or artifact generation cannot complete	Run Check inference	Inference troubleshooting
PostgreSQL unavailable or migrations missing	Durable state and jobs cannot be trusted	Open Health and inspect database readiness	Database recovery
`lease-expired`, `worker-recovery-exhausted`	Queued work is delayed or failed	Check worker health and latest attempt	Worker leases
Monitoring degraded	Prior approval is not currently verified	Inspect latest monitoring run and dependencies	Monitoring recovery
Notification delivery failed	The in-app delivery record is unsent	Inspect attempt count and last error	Notification outbox
Artifact validation or integrity failure	Patch cannot progress to review/application	Inspect the first failed row or checksum	Migration artifact recovery
Sandbox baseline, application, or re-ingestion failure	Accepted work is not verified as applied	Inspect the failed stage before retrying	Sandbox recovery
Write-back rejected or fresh read mismatch	DataHub result remains incomplete	Inspect the immutable mutation step	DataHub write-back

When reporting an unresolved issue, include the safe error code, route, Payne version, operating mode, latest attempt timestamp, and redacted service status. Never include tokens, keys, credential-bearing URLs, raw private queries, or provider payloads.