So I looked at this. It's a bigger problem in general and it's due to the way we handle "not matching". It's slightly complex, so bear with me.

so the query that query builder produces is this: https://w.wiki/unm (with some modifications):

SELECT ?item ?itemLabel ?instance ?instanceLabel WHERE {
  SERVICE wikibase:label { bd:serviceParam wikibase:language "[AUTO_LANGUAGE]". }
  ?item (p:P106/ps:P106/(wdt:P279*)) wd:Q2526255.
  ?item (p:P31/ps:P31/(wdt:P279*)) ?instance.
  FILTER(?instance != wd:Q5)
}
LIMIT 5

What is happening? it's basically going up the ladder of P279 for not matching, so P279 of Q5 and its P279 and so on, so each item becomes several rows (as wdqs is a graph db of a triples not items).
So the result would be:

And it only removes the last line (and leaves the rest) making the query both incorrect and full of duplicates.

I talked to @Lucas_Werkmeister_WMDE and came up with several solutions but each has pros and cons.

One:
https://w.wiki/unp

SELECT DISTINCT ?item ?itemLabel ?instance ?instanceLabel WHERE {
  SERVICE wikibase:label { bd:serviceParam wikibase:language "[AUTO_LANGUAGE]". }
  ?item (p:P106/ps:P106/(wdt:P279*)) wd:Q2526255.
  ?item (p:P31/ps:P31) ?class.
  MINUS { ?item (p:P31/ps:P31/(wdt:P279)*) wd:Q5. }
}
LIMIT 5

Basically take every one who has P31, and remove anything that has the Q5 in the P279 ladder

Pros:

• Correct

Con:

• It times out

Two: https://w.wiki/unu
The other way to handle it is to actually discard P279 ladder for "not matching" part.

SELECT DISTINCT ?item ?itemLabel ?instance ?instanceLabel WHERE {
  SERVICE wikibase:label { bd:serviceParam wikibase:language "[AUTO_LANGUAGE]". }
  ?item (p:P106/ps:P106/(wdt:P279*)) wd:Q2526255.
  MINUS {?item p:P31/ps:P31 wd:Q5. }
}
LIMIT 5

Pros:

• It's fast

Cons:

• It's limited, If I want to filter out galaxies from my result, it wouldn't exclude spiral galaxies, etc.

I don't know which way to go. I think @Lydia_Pintscher should decide here.

In T272140#6756245, @Ladsgroup wrote:

One:
https://w.wiki/unp

SELECT DISTINCT ?item ?itemLabel ?instance ?instanceLabel WHERE {
  SERVICE wikibase:label { bd:serviceParam wikibase:language "[AUTO_LANGUAGE]". }
  ?item (p:P106/ps:P106/(wdt:P279*)) wd:Q2526255.
  ?item (p:P31/ps:P31) ?class.
  MINUS { ?item (p:P31/ps:P31/(wdt:P279)*) wd:Q5. }
}
LIMIT 5

Basically take every one who has P31, and remove anything that has the Q5 in the P279 ladder

Pros:

• Correct

Con:

• It times out

You can rearrange that query to optimize (compare T166139):

SELECT DISTINCT ?item ?itemLabel ?instance ?instanceLabel WHERE {
  {
    SELECT DISTINCT ?item ?instance WHERE {
      ?item (p:P106/ps:P106/(wdt:P279*)) wd:Q2526255.
      ?item (p:P31/ps:P31) ?instance.
      MINUS { ?item (p:P31/ps:P31/(wdt:P279)*) wd:Q5. }
    }
    LIMIT 5
  }
  SERVICE wikibase:label { bd:serviceParam wikibase:language "[AUTO_LANGUAGE]". }
}

Then it doesn’t time out anymore (though it’s still expensive).

Edit: Note that this will still give you duplicate results when items have multiple P31 statements (fictional human + television character, for example).

Query tried during the discussion call:

SELECT DISTINCT ?item ?itemLabel ?instance ?instanceLabel WHERE {
  {
    SELECT DISTINCT ?item ?instance WHERE {
      ?item (p:P106/ps:P106/(wdt:P279*)) wd:Q2526255.
      ?item (p:P31/ps:P31) ?instance.
      MINUS { ?item (p:P31/ps:P31/(wdt:P279)*) wd:Q5. }
    }
    LIMIT 5
  }
  SERVICE wikibase:label { bd:serviceParam wikibase:language "[AUTO_LANGUAGE]". }
}

Filed T272490: Improve labelling performance in query builder

https://github.com/wmde/query-builder/pull/166