Count Every Hit, Rank More Candidates
The count is always the real number. A search with the word lookup counts every document that
holds the query's words, exactly, else within one edit, else within two, and returns it in
ToleranceCounts: on a broad query the count is not capped by how many documents are ranked.
CountedTolerance says which rung the list was taken from.
How many are ranked is the candidate bound: the rows asked for times
Candidates.CandidatesPerRow (4), at least 100 and at most 5 000. When the counted documents are more, the best of
them by pattern recognition are ranked, and Result.Count is then above Result.Ranked. To rank more of them, raise
CandidatesPerRow (up to the 5 000) or set MaxCandidates, which sets the bound outright, past the 5 000 too. The cost is linear in the candidates, so
benchmark before raising it on millions of documents.
The facet counts are counted over the same counted documents, the whole rung, so the facet rail and the count beside the list always agree.
var query = new Query("search terms", 50) { Candidates = { MaxCandidates = 5000 } };
var result = engine.Search(query);
// result.ToleranceCounts.Exact: every document holding the words exactly