A workflow that had been working now stalls in a waiting state until failing.

A job using runs-on: [gpu] depends on an available runner matching that label. If the workflow previously succeeded but now remains waiting, the runner ' s availability, registration, connectivity, or label assignment should be investigated. GitHub stores self-hosted runner application diagnostic logs in the runner installation ' s _diag directory; Runner_ log files provide details about runner startup, connectivity, registration, and update activity, making option A useful. Checking the Set up job information from a previous successful run can also identify which runner previously accepted the job, providing a concrete target for investigation. Rotating GITHUB_TOKEN does not repair runner scheduling because job assignment does not depend on manually rotating that workflow token. GitHub-hosted usage limits are similarly unrelated to a custom gpu runner label. Therefore, A and C are appropriate troubleshooting actions.

================