Since the issue with conference_auto_record is happening intermittently, it’s likely due to timing, resource conflicts, or certain call scenarios. To troubleshoot, try adding a small delay (1-2 seconds) in the sched_api command to ensure the conference is fully set up before recording starts. Compare channel variables for successful and failed calls using uuid_dump to spot inconsistencies, especially with the conference_auto_record variable. Turn on detailed logging to catch any errors or unusual behavior during the process. Also, check your system resources (CPU, memory, disk) to rule out performance issues, and test under higher call volumes to see if the problem only happens under load. Finally, make sure the variable is being passed reliably, using export if needed, and enable conference event logging to confirm when recordings are being triggered. These steps should help you pinpoint the cause.