@@ -34,6 +34,7 @@ jobs:
         with:
           files: |
             dist/dump/*.jsonl
+            dist/dump/aliases.json
             dist/dump/datapackage.json
           fail_on_unmatched_files: true
           generate_release_notes: true
@@ -24,7 +24,7 @@ Husky installs on `npm install`: Prettier runs on staged files, `commitlint` enf
 - `src/content/docs/{association,standard,api,community,get-started}/` — site sections (autogenerated sidebar)
 - `src/content/docs/de/` — German mirror (see Localization)
 - `src/pages/id/{work,system,ref,mapping}/` — canonical record pages (`/id/.../` HTML and `.json` JSON-LD siblings)
-- `src/pages/reg/` — registry browser (index + paginated reference lists); does not host individual records
+- `src/pages/reg/` — registry browser (index + paginated reference lists), the registry-wide JSON-LD collections, and the per-work locator index (`/reg/work/{key}/aliases.json`); does not host individual records
 - `src/pages/cite/` — short-alias redirects (`/cite/{work}/{system}/{locator}` always; bare `/cite/{work}/{locator}` for a work's preferred citation system, and retargetable if that preference changes) → canonical `/id/` URL
 - `src/styles/brand.css` — brand tokens
 - `public/{logo.svg,favicon.svg,fonts/}` — brand assets
 
@@ -71,19 +71,19 @@ Prerequisites: Node 24 and npm.
 
 Configuration lives in `.env`; use [`.env.example`](./.env.example) as the starting point. `SITE_DOMAIN` controls Astro's canonical `site` URL and defaults to `textrefs.org` when unset.
 
-| Command                 | Action                                                                                                                                                              |
-| :---------------------- | :------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
-| `npm install`           | Install dependencies; wires git hooks (husky + lint-staged)                                                                                                         |
-| `npm run dev`           | Start local dev server at `localhost:4321`                                                                                                                          |
-| `npm run build`         | Build the production site to `./dist/`                                                                                                                              |
-| `npm run build:fast`    | Build the site against a tiny fixture registry, without compiling full data                                                                                         |
-| `npm run preview`       | Preview the build locally                                                                                                                                           |
-| `npm run compile:data`  | Read hand-authored YAML under `data/works/` and `data/systems/`, expand the in-memory registry, and emit JSONL resources plus `datapackage.json` under `dist/dump/` |
-| `npm run validate:data` | Validate every compiled record against the canonical Zod schemas                                                                                                    |
-| `npm run build:data`    | `compile:data` then `validate:data` — the contributor data pipeline                                                                                                 |
-| `npm run verify:fast`   | Fast local check using fixture registry data                                                                                                                        |
-| `npm run verify`        | Prettier + `astro check` + tests + production build — the CI gate                                                                                                   |
-| `npm run changelog`     | Regenerate `CHANGELOG.md` from git history (git-cliff)                                                                                                              |
+| Command                 | Action                                                                                                                                                                               |
+| :---------------------- | :----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `npm install`           | Install dependencies; wires git hooks (husky + lint-staged)                                                                                                                          |
+| `npm run dev`           | Start local dev server at `localhost:4321`                                                                                                                                           |
+| `npm run build`         | Build the production site to `./dist/`                                                                                                                                               |
+| `npm run build:fast`    | Build the site against a tiny fixture registry, without compiling full data                                                                                                          |
+| `npm run preview`       | Preview the build locally                                                                                                                                                            |
+| `npm run compile:data`  | Read hand-authored YAML under `data/works/` and `data/systems/`, expand the in-memory registry, and emit JSONL resources, the alias table, and `datapackage.json` under `dist/dump/` |
+| `npm run validate:data` | Validate every compiled record against the canonical Zod schemas                                                                                                                     |
+| `npm run build:data`    | `compile:data` then `validate:data` — the contributor data pipeline                                                                                                                  |
+| `npm run verify:fast`   | Fast local check using fixture registry data                                                                                                                                         |
+| `npm run verify`        | Prettier + `astro check` + tests + production build — the CI gate                                                                                                                    |
+| `npm run changelog`     | Regenerate `CHANGELOG.md` from git history (git-cliff)                                                                                                                               |
 
 Contributors edit the YAML under [`data/works/`](https://github.com/textrefs/registry/tree/main/works) and [`data/systems/`](https://github.com/textrefs/registry/tree/main/systems); the directory is a git submodule pointing at [`textrefs/registry`](https://github.com/textrefs/registry). Run `git submodule update --init --recursive` after cloning. The compiler expands the pinned submodule into the flat registry dump (works, systems, refs, mappings) under `dist/dump/`. See [`docs/get-started/authoring`](https://textrefs.org/get-started/authoring/) for the format. For documentation, styling, and route work, use `npm run verify:fast` locally; run the full `npm run verify` before PRs that touch registry data, release output, production build behaviour, or CI behaviour.
 
 
@@ -23,16 +23,20 @@ info:
     yet live.
 
     A client that does not yet know a key starts at `/reg/works.json` or
-    `/reg/systems.json`. `/dump/` carries the whole registry as
-    newline-delimited JSON for bulk consumers.
+    `/reg/systems.json`. A client that knows a work key, a citation system
+    key, and a locator reads the reference UUID from
+    `/reg/work/{key}/aliases.json`. It needs no UUIDv5 implementation of its
+    own. `/dump/` carries the whole registry as newline-delimited JSON, plus
+    the whole alias table as one JSON object, for bulk consumers.
 
     The site is static, and the host derives every `Content-Type` from the
     file extension. This contract therefore declares the media type that
     the host actually sends, not the media type that best describes the
-    payload. A `.json` body is JSON-LD by content and arrives as
-    `application/json`. A `.jsonl` body is newline-delimited JSON by
-    content and arrives as `application/octet-stream`. Parse by
-    documented shape, not by response header.
+    payload. A record or collection `.json` body is JSON-LD by content and
+    arrives as `application/json`. The two alias artifacts are plain JSON
+    and arrive the same way. A `.jsonl` body is newline-delimited JSON by
+    content and arrives as `application/octet-stream`. Parse by documented
+    shape, not by response header.
   license:
     name: AGPL-3.0-or-later
     url: https://www.gnu.org/licenses/agpl-3.0.html
@@ -47,13 +51,15 @@ tags:
     description: Canonical record URLs under `/id/`.
   - name: Collections
     description:
-      Registry-wide JSON-LD collections under `/reg/`. Entry points for a
-      client that does not yet know a key.
+      Registry-wide JSON-LD collections and per-work locator indexes under
+      `/reg/`. Entry points for a client that knows no key, or that knows a
+      locator but no reference UUID.
   - name: Bulk
     description:
       Whole-registry artifacts under `/dump/`, rebuilt on every deploy. The
-      host serves them as `application/octet-stream`; the bodies are
-      newline-delimited JSON.
+      `.jsonl` bodies are newline-delimited JSON, and the host serves them
+      as `application/octet-stream`. The `.json` bodies are single JSON
+      objects, and the host serves them as `application/json`.
 paths:
   /id/work/{key}/:
     get:
@@ -200,6 +206,28 @@ paths:
           content:
             application/json:
               schema: { $ref: '#/components/schemas/SystemCollection' }
+  /reg/work/{key}/aliases.json:
+    get:
+      tags: [Collections]
+      summary: Work locator index — JSON
+      operationId: getWorkAliasIndex
+      description: Maps every locator of one work to the UUID of its
+        CanonicalReference, grouped by citation system key. A client that
+        knows a passage as work, citation system, and locator reads the UUID
+        here, then fetches `/id/ref/{uuid}.json`. It needs no UUIDv5
+        implementation of its own. The body is plain JSON, not JSON-LD, and
+        carries no `@context`.
+      parameters:
+        - $ref: '#/components/parameters/FlatKey'
+      responses:
+        '200':
+          description:
+            Every reference of this work, in every status, keyed by citation
+            system key and then by locator.
+          content:
+            application/json:
+              schema: { $ref: '#/components/schemas/WorkAliasIndex' }
+        '404': { $ref: '#/components/responses/NotFound' }
   /dump/works.jsonl:
     get:
       tags: [Bulk]
@@ -252,6 +280,23 @@ paths:
           content:
             application/octet-stream:
               schema: { type: string }
+  /dump/aliases.json:
+    get:
+      tags: [Bulk]
+      summary: Alias table — JSON
+      operationId: getAliasesDump
+      description:
+        The complete alias table as one JSON object, with sorted keys and no
+        indentation. For a single lookup, use
+        `/reg/work/{key}/aliases.json` instead. This file serves a consumer
+        that mirrors the whole table.
+      responses:
+        '200':
+          description: One JSON object. Each key is an alias. Each value is the
+            canonical IRI that the alias resolves to.
+          content:
+            application/json:
+              schema: { $ref: '#/components/schemas/AliasTable' }
   /dump/datapackage.json:
     get:
       tags: [Bulk]
@@ -260,9 +305,10 @@ paths:
       responses:
         '200':
           description:
-            The Frictionless data-package descriptor that lists the four
-            JSONL resources with, for each, its byte count and its
-            `sha256:` hash, plus the registry version.
+            The Frictionless data-package descriptor that lists the five
+            resources — the four JSONL files and the JSON alias table —
+            with, for each, its byte count and its `sha256:` hash, plus the
+            registry version.
           content:
             application/json:
               schema: { type: object }
@@ -481,6 +527,46 @@ components:
         '@graph':
           type: array
           items: { $ref: '#/components/schemas/CitationSystem' }
+    WorkAliasIndex:
+      description:
+        The `/reg/work/{key}/aliases.json` locator index of one work. Plain
+        JSON, not JSON-LD; it carries no `@context`. The index states
+        identity alone. Read the CanonicalReference for the status and the
+        resolver targets.
+      type: object
+      required: [work_key, preferred_citation_system_key, refs]
+      properties:
+        work_key: { $ref: '#/components/schemas/FlatKey' }
+        preferred_citation_system_key:
+          allOf:
+            - $ref: '#/components/schemas/FlatKey'
+          description:
+            The system that the bare `/cite/{work_key}/{locator}` alias uses
+            (ADR-0005). A default for presentation, never an identity claim.
+            The other keys of `refs` have equal standing.
+        refs:
+          type: object
+          description: One entry per citation system key that this work carries
+            references under. An empty object means the work has no
+            references yet. The same locator under two systems denotes a
+            different passage, and a different UUID, each time.
+          additionalProperties:
+            type: object
+            description: Locator to CanonicalReference UUID.
+            additionalProperties:
+              type: string
+              pattern: '^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$'
+    AliasTable:
+      description:
+        The `/dump/aliases.json` table, with one entry per alias. A key is
+        either a `/cite/` alias path — `{work_key}/{locator}` or
+        `{work_key}/{citation_system_key}/{locator}` — or an external
+        identifier IRI carried by a MappingAssertion. A value is the
+        canonical IRI that the key resolves to — an `/id/ref/` IRI for a
+        `/cite/` alias, an `/id/work/` IRI for an external identifier. A
+        `://` in the key marks the second kind.
+      type: object
+      additionalProperties: { $ref: '#/components/schemas/Iri' }
     RegistryObject:
       oneOf:
         - { $ref: '#/components/schemas/Work' }
 
@@ -10,7 +10,13 @@ import { mkdtempSync, mkdirSync, writeFileSync, rmSync } from 'node:fs';
 import { tmpdir } from 'node:os';
 import { join } from 'node:path';
 import { v5 as uuidv5 } from 'uuid';
-import { compileRegistry, type CompiledRegistry } from './compile.js';
+import { createHash } from 'node:crypto';
+import {
+	compileRegistry,
+	dumpResources,
+	describeResource,
+	type CompiledRegistry,
+} from './compile.js';
 import { LanguageTag } from '../standard/schema/common.js';
 
 type RegistryFiles = {
@@ -671,3 +677,100 @@ test('regular grandfathered tags parse and irregular ones do not', () => {
 		[],
 	);
 });
+
+// --- Dump resources (#84) -------------------------------------------------
+
+// `workWithMappings` gives an alias table with both kinds of entry: `/cite/`
+// aliases targeting reference IRIs, and external identifiers targeting the work
+// IRI.
+const dumped = () => dumpResources(compileFixture(workWithMappings()));
+
+const aliasSpec = (specs: ReturnType<typeof dumpResources>) => {
+	const spec = specs.find((s) => s.name === 'aliases');
+	assert.ok(spec, 'the dump carries an alias resource');
+	return spec;
+};
+
+test('the dump carries the alias table beside the four JSONL resources', () => {
+	const specs = dumped();
+	assert.deepEqual(
+		specs.map((s) => s.name),
+		['works', 'citation-systems', 'references', 'mappings', 'aliases'],
+	);
+	const spec = aliasSpec(specs);
+	assert.equal(spec.filename, 'aliases.json');
+	assert.equal(spec.format, 'json');
+	assert.equal(spec.mediatype, 'application/json');
+});
+
+test('the alias resource body is the whole alias table', () => {
+	const reg = compileFixture(workWithMappings());
+	const spec = aliasSpec(dumpResources(reg));
+	const table = JSON.parse(spec.body) as Record<string, string>;
+	assert.deepEqual(table, reg.aliases);
+	// A `/cite/` alias resolves to a reference; an external identifier resolves
+	// to the work. A `://` in the key marks the second kind.
+	assert.match(
+		table['test.work/primary-section/5'],
+		/^https:\/\/textrefs\.org\/id\/ref\//,
+	);
+	assert.equal(
+		table['https://www.wikidata.org/entity/Q1'],
+		'https://textrefs.org/id/work/test.work',
+	);
+});
+
+test('alias keys are sorted, so the body does not depend on visit order', () => {
+	const keys = Object.keys(JSON.parse(aliasSpec(dumped()).body));
+	assert.ok(keys.length > 1);
+	assert.deepEqual(keys, [...keys].sort());
+});
+
+test('the alias body is minified and newline-terminated', () => {
+	const body = aliasSpec(dumped()).body;
+	assert.equal(body.split('\n').length, 2);
+	assert.ok(body.startsWith('{"'));
+	assert.ok(body.endsWith('}\n'));
+});
+
+test('every descriptor states the bytes and sha256 of its own body', () => {
+	for (const spec of dumped()) {
+		const d = describeResource(spec);
+		// Recomputed the way a downstream consumer verifies a download.
+		assert.equal(d.bytes, Buffer.byteLength(spec.body, 'utf8'));
+		assert.equal(
+			d.hash,
+			`sha256:${createHash('sha256').update(spec.body).digest('hex')}`,
+		);
+	}
+});
+
+test('every descriptor carries the Frictionless fields', () => {
+	for (const spec of dumped()) {
+		const d = describeResource(spec);
+		assert.equal(d.profile, 'data-resource');
+		assert.equal(d.encoding, 'utf-8');
+		assert.equal(d.path, spec.filename);
+		assert.equal(d.name, spec.name);
+	}
+});
+
+test('the JSONL bodies are unchanged by the alias-resource refactor', () => {
+	const reg = compileFixture(workWithMappings());
+	const specs = dumpResources(reg);
+	const bodyOf = (name: string) => specs.find((s) => s.name === name)?.body;
+	assert.equal(
+		bodyOf('works'),
+		reg.works.map((r) => JSON.stringify(r)).join('\n') + '\n',
+	);
+	assert.equal(
+		bodyOf('references'),
+		reg.references.map((r) => JSON.stringify(r)).join('\n') + '\n',
+	);
+	// An empty record array still yields an empty body, not a bare newline.
+	assert.equal(
+		dumpResources({ ...reg, mappings: [] }).find((s) => s.name === 'mappings')
+			?.body,
+		'',
+	);
+});