Highlighting with Querylight TS

Highlighting in Querylight TS is a post-retrieval step.

That means you:

1. run a normal query and get back hits
2. choose which fields to highlight for those hits
3. ask the DocumentIndex to generate fragments from the original stored field text

This keeps ranking and highlighting separate, similar to how Elasticsearch/OpenSearch treat highlighting.

Basic usage

import { DocumentIndex, MatchQuery, RankingAlgorithm, TextFieldIndex } from "@tryformation/querylight-ts";

const index = new DocumentIndex({
  title: new TextFieldIndex(undefined, undefined, RankingAlgorithm.BM25),
  body: new TextFieldIndex(undefined, undefined, RankingAlgorithm.BM25)
});

index.index({
  id: "range-filters",
  fields: {
    title: ["RangeQuery Over Lexical Fields"],
    body: ["Range filters work well for sortable values and filtering interfaces."]
  }
});

const query = new MatchQuery("body", "range filters");
const hits = index.searchRequest({ query, limit: 5 });

const highlight = index.highlight(hits[0]![0], query, {
  fields: ["title", "body"],
  fragmentSize: 140,
  numberOfFragments: 1
});

What the highlighter returns

The result is grouped by field and each field contains one or more fragments.

Each fragment includes:

• fragment text from the original stored value
• highlighted parts ready to render as <mark> spans
• exact source-relative offsets for the highlighted spans

This is useful for:

• result-title highlighting
• short “why it matched” excerpts
• exact, phrase, prefix, and fuzzy evidence on prose fields

Current support

The current implementation is intentionally conservative:

• exact term highlighting
• phrase highlighting
• prefix highlighting
• fuzzy highlighting for approximate analyzers such as ngrams
• source offsets from stored field text

That makes it reliable for fields such as title, summary-style fields, and body.

Current limitations

This is not yet a full clone of all Elasticsearch/OpenSearch highlighter modes.

In particular:

• fragment selection is simple and works best on prose
• approximate matches highlight the containing token span rather than reconstructing token-filter output
• code-heavy fields are usually poor candidates for result-card snippets

For best results in the UI, highlight prose fields and keep code/example fields out of default result snippets.