Indexing rules

The indexer reads Zola's TOML +++ front matter properly and honours Zola's own conventions rather than approximating them. This page is the exact contract; chops-search docs shows you its output for your site.

What gets indexed

Every .md file under content, except:

_index.md files (section pages) are skipped entirely
draft = true pages
in_search_index = false pages (Zola's own opt-out convention)

Malformed front-matter TOML is a build error, not a silent default: Zola itself would refuse to build such a page, so indexing it anyway would diverge from the site.

What gets weighted

Keyword scoring is BM25F over four fields. Tags from [taxonomies] tags = [...] are the highest-weighted field (default 4): they carry more signal per byte than anything in the body, being the author's own statement of what a page is about. Titles weight 2, the front-matter description sits at parity with body (1), and each field is normalised by its own length before the weights apply, so a weight biases rather than inflates. How ranking works has the mechanics; the weights are configurable and baked into the index at build time.

URL reconstruction

Each page's URL is rebuilt the way Zola would build it:

Rule	Example
`path` front-matter override wins outright	`path = "/elsewhere"` → `/elsewhere/`
`slug` replaces the final segment	`slug = "custom"` → `/blog/custom/`
Page bundles collapse	`foo/index.md` → `/foo/`
`YYYY-MM-DD-` filename prefixes strip	`2024-06-04-my-post.md` → `/blog/my-post/` (a front-matter `date` overrides the date but not the stripping)
Every segment is slugified (`slugify.paths = "on"`, Zola's default)	`My Post.md` → `/my-post/`, `Café_Crème` → `/cafe-creme/`

Numeric prefixes that aren't full dates are kept, as Zola keeps them: 01-baseline stays 01-baseline.

Documented gaps

Stated rather than guessed at:

_index.md content is invisible to search. If your substantive prose lives in a section landing page (a series intro, say), it won't be findable. Move it into a page, or accept the gap.
Multilingual suffixes (foo.fr.md) are not handled.
Per-section path overrides in ancestor _index.md files are not applied to descendant URLs.
Non-Latin slug transliteration diverges from Zola's: Unicode alphanumerics are kept as-is where Zola transliterates to ASCII. Latin-script content is byte-identical.

If eval expectations 404, run chops-search docs and diff its URLs against your live site before assuming a ranking bug.