API: ranking/passages

better-web-search-mcp / ranking/passages

ranking/passages#

Interfaces#

Passage#

A scored span of text taken from a page.

Properties#

Property	Type	Description	Defined in
`end`	`number`	Character offset of the passage end within the source content.	src/ranking/passages.ts:26
`score`	`number`	BM25 score against the query. Higher is more relevant.	src/ranking/passages.ts:28
`start`	`number`	Character offset of the passage start within the source content.	src/ranking/passages.ts:24
`text`	`string`	The passage text, trimmed.	src/ranking/passages.ts:22

Variables#

MAX_PASSAGE_CHARS#

const MAX_PASSAGE_CHARS: 1200 = 1_200

Defined in: src/ranking/passages.ts:47

Hard ceiling on a single passage, in characters.

Paragraph splitting alone is not enough: plenty of extracted pages contain no blank lines at all, which made the whole page one passage. A cited “passage” was then the entire document, and web_research returned more text than reading the pages directly would have. Anything longer than this is windowed on sentence boundaries.

TARGET_PASSAGE_CHARS#

const TARGET_PASSAGE_CHARS: 500 = 500

Defined in: src/ranking/passages.ts:36

Target passage size in characters; paragraphs are packed up to this.

Functions#

rankPassages()#

rankPassages(passages, query): Passage[]

Defined in: src/ranking/passages.ts:153

Score passages against query with BM25 and return them ranked.

Passages that share no query term score 0 and are dropped: returning a passage with nothing in common with the question is what the old first-400-characters approach did.

Parameters#

Parameter	Type
`passages`	readonly `Passage`[]
`query`	`string`

Returns#

Passage[]

selectPassages()#

selectPassages(content, query, limit?): Passage[]

Defined in: src/ranking/passages.ts:218

Select the passages of content most relevant to query.

Parameters#

Parameter	Type	Default value	Description
`content`	`string`	`undefined`	The extracted page content.
`query`	`string`	`undefined`	The user’s question.
`limit`	`number`	`2`	Maximum passages to return.

Returns#

Passage[]

splitPassages()#

splitPassages(content): Passage[]

Defined in: src/ranking/passages.ts:71

Split content into passage-sized chunks on paragraph boundaries.

Markdown from the extraction pipeline is paragraph-separated, so blank lines are the natural split. Short paragraphs are packed together up to TARGET_PASSAGE_CHARS so a one-line heading is not its own passage, and an over-long paragraph is emitted whole rather than cut mid-sentence.

Parameters#

Parameter	Type
`content`	`string`

Returns#

Passage[]

tokenize()#

tokenize(text): string[]

Defined in: src/ranking/passages.ts:58

Lowercase alphanumeric tokens, stop words removed.

Parameters#

Parameter	Type
`text`	`string`

Returns#

string[]