This is not really anything to do with your search patterns or Kagi itself, just a general fact that the upstream sources we use are not guarenteed to return the same results for the "same" search terms, semantically. Even issuing the same search request back-to-back byte-for-byte can and will elicit different results in different orders; you are effectively just breaking our cache and forcing a retry.
It is also possible that one of our many sources failed on your first attempt, but succeeded on your second attempt, filling in the "missing" results. So even barring "noise" in the APIs that we use, this could still happen by other means.
We could put the effort into trying to cache both of these the same way, or "sorting" the tokens of your query, but it's effort for no benefit really, could possibly do more harm than good. So, suffice to say this is generally out of our control at the moment
Sorry for the confusion, and thanks for taking the time to report!
