I believe this behavior is “by design” but I agree that it is confusing.
As for in-reply-to suppression on immediate replies, we tried this both ways and it is very noisy when you don’t suppress the indicator in the following conditions:
when there is exactly 1 reply
when that reply is directly under the post it is replying to
Locality means a lot in conversations, and when one person stops talking and another starts talking, the vast, vast, vast majority of the time they are responding to what was just said.
This is now a site setting, so if you want reply indicator to show up in all cases, flip it on for your Discourse instance. Problem solved.