Reference · Cheat sheet

Query vs Filter context

The compressed essence. Print it, pin it.

From lesson 0001See also glossary

	Query context	Filter context
Question asked	How well does it match?	Does it match — yes/no?
Produces _score?	Yes (BM25)	No (score irrelevant)
Cacheable?	No	Yes — node query cache (bitsets)
bool keys	`must`, `should`	`filter`, `must_not`
Use for	Full-text, relevance ranking	Exact constraints: status, dates, tags, ranges

Canonical shape

GET /articles/_search
{
  "query": {
    "bool": {
      "must":   { "match": { "title": "elasticsearch tuning" } },   // scored
      "filter": [
        { "term":  { "status": "published" } },                     // yes/no, cached
        { "range": { "published_at": { "gte": "2024-01-01" } } }
      ]
    }
  }
}

Rules of thumb

If it shouldn't affect ranking, filter it. Status, category, date ranges, permissions → filter.
Filters are cheaper on repeat. They skip scoring and cache as bitsets; reuse them across requests.
must vs filter = same matching, different scoring. Both must match; only must adds to _score.
must_not is filter context — exclusion never contributes to score.
minimum_should_match: defaults to 1 with no must/filter; 0 when a must/filter is present (should becomes a pure boost).
Want matching but a flat score? Wrap a filter in constant_score.

Common mistake

Putting exact constraints (e.g. term status=active) inside must. It still works, but you pay for scoring you throw away and lose the filter cache. Move it to filter.

← lesson 0001 · Source: Elastic Docs — Query and filter context