This sum is wrong. You're adding all the different ints that satisfy modelCoverage(package, _, _, type, n), so if two kinds happen to have the exact same number of models, then those numbers won't be added. You'll need to add string kind in the range. Another minor thing: If you change sum to strictsum, then it'll bind package and type so you can drop the first line.

Thanks for the explanation. The kind thing is tricky, that's hard to see.

Regarding strictsum - do you think it might be worth mentioning that on https://codeql.github.com/docs/ql-language-reference/expressions/ ?
Currently it just says strictconcat, strictcount, and strictsum: These aggregates work like concat, count, and sum respectively, except that they are strict. That is, if there are no possible assignments to the aggregation variables that satisfy the formula, then the entire aggregation fails and evaluates to the empty set (instead of defaulting to 0 or the empty string). This is useful if you’re only interested in results where the aggregation body is non-trivial..

No, the missing kind in the range has nothing to do with the switch from sum to strictsum. The explanation of strictsum that you linked seems perfectly adequate.