Azure dev ops pipelines jobs not starting with self-hosted agent

I'm having trouble with Azure Pipelines—jobs never start, even though everything seems correctly configured. I'm using a Managed DevOps Pool (cicd-pipeline-devops-pool2) with Ubuntu 22.04. The YAML pipeline is valid, runs on Python 3.11, installs dependencies, and uses pytest. Jobs get queued but hang indefinitely with logs like “We are provisioning an agent…” and “Pool provider seems slow…”, until they time out.

What I’ve checked:

• YAML triggers correctly; branch and task setup (UsePythonVersion, venv, pytest) are fine.
• No custom demands (e.g., Agent.OSVersion) are present.
• I’ve avoided vmImage because I don’t have parallelism for Microsoft-hosted agents (request pending for 3+ weeks).

On the Managed Pool side:

• Manual scaling window (Wed 12–16h) with 100% standby agents.
• Azure Portal shows agents going Ready → Allocated → PendingReturn → Returned, often without jobs running.
• Sometimes a job is allocated to an agent but never progresses.
• No manual cancellations. No hard errors. No PoolProviderTimeout.

What I suspect:

• Agents are provisioned, but jobs don’t bind in time or agents are retired too quickly.
• Could still be a hidden capability mismatch, but I can’t view agent capabilities in a Managed Pool.

Looking for:

• Any known issues with agent/job orchestration in Managed Pools?
• Whether auto-scaling is more reliable than manual standby?
• A way to inspect reported agent capabilities in a Managed Pool?
• And how to escalate parallelism approval for Microsoft-hosted agents?

Any help would be greatly appreciated.