Billing granularity (minimum billable time)

I’m starting to work on a project that will use the DO API to spin up large VM instances for short process runs. The idea is I will spin up the VM and provide a script that will run the task at hand. That script will do some intensive compute and upload the results somewhere. The VM will then power itself off, and a monitor process on my end will check every minute; once the VM is powered off it will be destroyed via the API. I have one question though: I’m not clear on how granular the Digital Ocean billing system is.

I’ve experimented with the lowest spec VM, but its per-hour billing is less than $0.01, so it doesn’t matter if I use it for one minute or for one hour, I’m still going to be billed $0.01.

However, if I spin up a very expensive droplet (such as a 256GB memory-optimized droplet) which runs a script doing some intensive processing that takes only around 15 minutes, am I billed for one hour of that VM, or am I billed for 15 minutes of it? The largest memory-optimized VM costs $2.97/hour, so does this mean the minimum billable time period is $2.97 even if I spin up the VM and immediately destroy it?

Similarly, what if the VM runs for one hour and thirty seconds? Am I then billed $5.94, or am I billed $3.02 ($2.97 + ~5c for the extra minute) This scenario could get expensive and ugly if billing is strictly hourly; suppose my process finished at 59:45 after the VM was created, but the script I am running to check for and destroy the VM when it is finished only checks once per minute, and it doesn’t grab the status and destroy the VM until 01:00:05 - I’d then be billed an extra $2.97 for five seconds during which the VM was powered off.

If there is more granularity, what level is it? Per minute? Per second? etc.?