Confirmed its happening upstream on secure.phabricator.com as well.

Reported upstream in their discourse environment.

I've filed this upstream in https://secure.phabricator.com/T13340. See that task for more details.

Although part of the goal here is to move these results to the bottom of the list, the strikethru rendering in global results and autosuggest results is a side effect of the internals accomplishing this in a slightly messy way, and I think unnecessarily confusing.

The intent is to: sort these results to the bottom; and render them in a way that makes the fact that the result is special in some way (here, the user account is a "Bot" account) clear to the user. The normal tokenizer accomplishes both these goals:

...but the global typeahead and the inline autosuggest render "closed" attributes differently, in a way that leads to visually conflating "disabled" and "bot":

Thank you @Peachey88 and @epriestley !

In T227963#5330775, @Aklapper wrote:

Looks like a valid design decisions to me: Why do you search for a non-human user with which you won't be able to interact?

I started typing the name of a bot because I was searching for tasks that involved it.
Then I got the autocompleted result, and initially assumed it was an Archived Tag (because those also display with strikeout and faded).
Finally I clicked through and realized it was a userprofile of an active bot.
(Yes, I should've mentioned all that in my description!).

However, even if someone was trying to find a bot, any active bot should have contact-info for the bot-owner in the userprofile description; hence someone might be discouraged by the formatting of that search autocomplete and not click through.

It might make more sense to switch the small to or such in the dropdown proposals