It is impossible to pause a cluster that was recently resume.

The documentation indicates it is a voluntary feature (https://www.mongodb.com/developer/products/atlas/pause-resume-atlas-clusters/) to allow the Atlas service to apply any pending changes or fixes to the cluster that may have accumulated while it was paused.

I agree with this, but it cause a problem for automation (for example sometimes we launch a platform exclusively for testing purpose and then shut it down, it is complicated to add a delay for pausing the Atlas cluster) and even manually for example at the end of the day of work this requires either staying later just to replay the pause or leaving the cluster running all night for example.

I want to suggest you to improve this feature by recording the pause request and applied it once you finished your maintenances. You might introduce a new cluster state (terminating ?) to show to everyone that the cluster will be paused in few time.