[est31]

Creating a correct implementation of a protocol is one thing, creating one that is indistinguishable from another implementation another. E.g. for TCP/IP you can find out the OS just from the additional information even though implementations are correct [1]. I guess they are doing something similar. And even if your protocol implementation were perfectly indistinguishable from the other one, usage patterns might reveal differences. E.g. if you have a bot then bots will most likely reply immediately after you issue a command. If you add a constant wait time, it's still distinguishable from humans typing at varying times. Same for uniformly randomly distributed wait times, I'm sure there is a distinction (I'd say it's correlated to message length for example). All of this is visible without looking at the message contents, which are obviously not available to the WhatsApp service.

[1]: https://en.wikipedia.org/wiki/TCP/IP_stack_fingerprinting