I had a quick look. This behavior is – unfortunately – just what CirrusSearch gives us.

• When you type "Spiel" without space, this is what happens: Special:Search/Spiel*
• When you type "Spiel " with space: Special:Search/Spiel

Vorlage: Spiel des Jahres appears in both cases, just on different positions (#76 in my test). The reasons are manifold. There is just sooooo much about games in the template namespace. And that particular template is rarely used.

The fact that the space at the end of "Spiel " makes a difference at all (and is not trimmed) is a little "hidden feature" we added intentionally exactly because of situations like this.

Continuing to type "Spiel des" moves the template you are looking for way up to position #2.

I suggest to leave this behavior as it is.

Ok, I understand that there are dozens of templates containing the phrase "Spiel". At the moment, I get the following result if I type in "Spiel"

Tabelle RoundRobin-10Spieler-1Ergebnis-4weiter
Navigationsleiste Spiel der Spiele
Navigationsleiste Spiele von Key (Spieleentwickler)
Spielerstatistik
Pointstreak Spiel
WNBA-Spielerprofil
Spielfigur
Spielwiese
BBL-Spielerprofil
Infobox Spiel

Only one of the first ten hits is a *documented* template. This list appears unsorted to me, but I understand now that the results are sorted by frequency of use, which is welcome. It is conceivable to offer other sort criteria (e.g. preferring prefix search results, or using Levenshtein distance), but this would be too complicated for the normal use case.

The primary reason why I didn't find the template was that I omitted the spaces when continuing the search. This misconception leaded to the search string "SpielDesJahres" with zero results. It's a little bit weird that the program still finds 10 results if I type in "SpielD", and no results, if I type in "SpielDe".

Is it possible to indicate to the user that there are more than 10 results, e,g. by displaying the total number of results? If I know that there are 76 possible hits or even more, then I will think a little bit more about variations of search items to find the correct one.

Another idea I have is to introduce a flag to indicate that the user is searching for documented main templates only, and not for e.g. un-documented sub-templates. Recently, I used an undocumented template and told the author of the template about a problem of using the template. The author instructed me not to use the template because the template was intentionally not documented because of internal use. In this sense, it might be useful to neglect undocumented templates in the search.

Taken as a whole, I would agree to leave the behavior as it is, as long as there is no real improvement on the horizon.