@tomerlichtash/md-registry

Entity registry for markdown corpora — the thing Astro's reference() does for content collections, but framework-free. One canonical record per entity (author, translator, subject), an alias map from display strings to ids, polymorphic byline resolution, and a per-call-site policy for unknown refs.

Core has zero dependencies and does no file I/O: it takes a plain entries array from any loader (@tomerlichtash/md-collections, your own, mocks).

import { createRegistry } from '@tomerlichtash/md-registry';

const people = createRegistry(entries, { nameField: 'name', aliasField: 'aka' });

// frontmatter `author:` may be a legacy free string, an id, or a list of ids
people.resolveDisplay(page.author, { join: ' & ', override: page.display_author });
people.idsOf(page.author);                    // → ['ada-quill']

// unknown-ref policy per call site
people.resolveRef('rio-marsh', { onUnknown: 'error' });        // typo fails the build
people.resolveRef('rio-marsh', { onUnknown: 'passthrough' });  // legacy strings flow through
people.resolveRef('rio-marsh', { onUnknown: 'unresolved' });   // placeholder for unknowable identities

Names can be plain strings or per-locale records ({ en, fr }); pass defaultLocale and/or a locale per call. An alias shared by two entities resolves to both — the combined-credit split. The building blocks (buildAliasMap, resolveDisplay, idsOf) are also exported as pure functions.

./reference — zod helper

import { reference } from '@tomerlichtash/md-registry/reference';

refs: z.array(reference(() => people)); // parse-time build fail on a typo'd id

zod is a peer dependency of this subpath only. Only want validation? z.string().refine((id) => knownIds.has(id)) covers it without this package.