Autoscale deployment healthcheck timeout regression — same config worked 2 days ago, now fails on every publish

My autoscale deployment was working correctly as of March 5. No changes were made to the deployment configuration. Since then, every republish fails the healthcheck at startup.

Run command:

gunicorn --bind=0.0.0.0:5000 --reuse-port app:app

What the logs show:

Gunicorn takes approximately 12 seconds to start and bind to 0.0.0.0:5000. The healthcheck begins probing 127.0.0.1:1104 immediately at startup, receives 500 responses for ~9 seconds (while the previous pod is still shutting down), then times out with context deadline exceeded after gunicorn starts — before it can serve a single request.

healthcheck failed: healthcheck / returned status 500  (×10, before gunicorn starts)

[INFO] Listening at: http://0.0.0.0:5000

[INFO] Booting worker with pid: 27

healthcheck failed: context deadline exceeded (×4+)

What I expect: The healthcheck should wait long enough for gunicorn to finish starting before marking the deployment as failed. This was the behavior 2 days ago on the same config.

Last working deployment commit: 172f79e (March 5)

Repl: attached-assets-mharriscdn.replit.app

Has the autoscale healthcheck startup grace period changed recently?