Context

Describe the background behind the problem.

Follow-up to INTPYTHON-1095, which added a fixed some_latency allowance to test_ttl to stop it racing MongoDB's TTL monitor (a single sleep(ttl + monitor_period) had zero margin). That fix used some_latency = 1 second on top of a monitor_period of 2 seconds - proportionally large, and the _release.yml failure that prompted it has only been observed once.

Fix

What was done.

Reduced some_latency from 1 to 0.5. Verified locally against a real local-Atlas container: 5/5 consecutive runs pass in ~3.6s each.

Pitfalls

What should the implementer watch out for? What are the risks?

None specific - contained to one test file, no production code changes.