Wildcard queries are not sent to the analyzers which means that the words in your query are not lower-cased and thus won't match any terms in the inverted index.
I agree that this is annoying and very confusing I will investigate to see if it's possible to force the analysis phase on wildcard tokens.
In the meantime I'd suggest to always use lowercase characters as a workaround:

• media*iki: https://www.mediawiki.org/w/index.php?search=media%2Aiki&title=Special:Search&go=Go
• wiki*edia: https://en.wikipedia.org/w/index.php?search=wiki%2Aedia&title=Special:Search&profile=default&fulltext=1

I'm curious to know if it's a recent regression that may have been caused by elasticsearch 5 migration. I'm tempted to say yes since it's the second bug report in the few days following elastic5 roll out.
I'll double check a restore the previous behavior if possible.

In T160829#3117467, @dcausse wrote:

I'm curious to know if it's a recent regression that may have been caused by elasticsearch 5 migration. I'm tempted to say yes since it's the second bug report in the few days following elastic5 roll out.

Yes, it is clearly a recent regression. I am currently using wildcard searches with uppercase charachters (looking for different spelling of people's last names more precisely) on a near-daily basis, and they worked correctly until last week.

Pulling this in to the current sprint to further investigate what is happening.

This dropped in priority due to other work and ongoing discussions. I've bumped it back up to the top of our list to investigate.

This will be resolved in the next version of ElasticSearch, moving to backlog for now

Moving to be with the epic (T163703)