Three ways to classify content, each with its own job. Tags are free-form page taxonomy; labels and categories are defined, reader-facing classifiers rendered in preview and published output.

Tags

Free-form keywords on a page:

---
title: Reverse Proxy Setup
tags: [nginx, deployment]
---

Tags render in the page's meta row on the published site and are searchable: a query matching a tag finds the page even when the word never appears in the text.

Categories are additive: a page can have several, in authored order. Documents store stable IDs, not localized display names:

---
label: new
categories:
  - beta-testing
  - cloud_services
---

The definition files use keyed identities. definitions/categories.yml owns optional global color; content/<locale>/definitions/categories.yml owns localized name and optional tooltip. Categories render below the page title or heading. Use them for broad areas and tags for specific topics; a page may have one label and several categories.

Labels

Root definitions/labels.yml owns keyed IDs and global color; locale files own localized name, optional compact, and optional tooltip. IDs match [a-z0-9_-]+ and are referenced from frontmatter or heading attributes. Labels render above the page title or heading.

# definitions/labels.yml
labels:
  new:
    color: "#10B981"

# content/en/definitions/labels.yml
labels:
  new:
    name: New
    compact: NEW
    tooltip: Recently added

Heading assignments

Any H1-H6 heading accepts a terminal attribute block. Use a scalar label and a typed, comma-separated categories list; formatting is canonical comma-space:

## Billing {#billing label="new" categories="beta-testing, cloud_services"}

{#billing} is the explicit anchor. Anchors must be unique in the rendered page; duplicate anchors are errors with rename/remove quick fixes. Missing IDs are errors with a create-definition quick fix. Repeated labels or category IDs warn and keep the first occurrence.

Archive pages

A tag or a category with at least one published entry gets its own generated page: /tags/<slug>/ and /categories/<id>/, always under those fixed English segments, on both site types, paginated like any other listing. They are never authored - an authored page whose route starts with tags/ or categories/ is NAV_013, and the build refuses to overwrite a generated archive rather than losing a page silently.

A tag's identity is its slug, so every spelling that normalizes to one slug is one tag with one page. Chips keep the spelling you wrote; the archive is titled by the first spelling in the site's own order; more than one spelling reports MD_015; and a tag that leaves no URL segment at all reports MD_016.

Archives reach sitemap.xml but never llms.txt, llms-full.txt, or the .md mirrors: they index content rather than carry it.

Catalog workflow

Catalog creation asks for localized Name and tooltip, an editable global ID, optional global Color, and label-only compact text. IDs are suggested by Unicode transliteration and normalized to [a-z0-9_-]+, but remain editable before confirmation. Creating from a selected locale writes the root identity and that locale payload atomically. Editing an ID renames all locale keys and references; deleting can remove only the selected locale or the entity everywhere. Removing the last assignment never deletes definition data.