Cold-start latency dropped sharply this quarter, falling to 210ms from 480ms after the scheduler switched to region-local warm pools. That change alone accounts for most of the improvement teams have noticed in edge response times. Connection pooling was also reworked in v4.2, which removed the old fixed pool-size cap so pools now scale to demand. Idle connections are reclaimed on a 90-second timer rather than being held open indefinitely. Autoscaling now reacts to load essentially instantaneously, closing the gap that used to cause bursty-traffic slowdowns. Together these changes are why p99 latency has been the most stable it has ever been.

3 of 6 sentences grounded, 1 contradiction, 2 unsupported