A Next.js page streams locally but arrives all at once in production. How do you diagnose it?

In a multi-instance deployment I also check build IDs, shared cache behavior, sticky assumptions, deployment skew, and whether a CDN coalesces or buffers dynamic responses. A feature can remain correct without streaming, so the incident severity follows user latency and timeout impact rather than the missing animation alone. I add a production smoke probe that records first byte and completion timing for a controlled streaming route. The repair includes a rollback path and topology documentation because later proxy, compression, or CDN changes can reintroduce buffering.