Deletes _node_document_ids

This brings users without an embedding backend configured to closer
parity with those who do.  Reuse the search backend to locate similar
documents and use them to provide the LLM with the better suggestion pool
to draw from

.github/workflows/issue-bot.yml

@@ -72,7 +72,7 @@ jobs:
                 'You are welcome to open a new issue that describes the problem you observed in your own words.'
               : 'This issue was automatically closed because it was not opened using our bug report form. ' +
                 'Issues have to be created through the form so that the details we need to investigate are included.\n\n' +
-                `If the problem is still there, please [open a new issue](${newIssue}) using the form. No other action is needed here.\n\n' +
+                `If the problem is still there, please [open a new issue](${newIssue}) using the form. No other action is needed here.\n\n` +
                 'If any part of your report was written by an AI tool or agent, you must say so: undisclosed AI-generated ' +
                 `contributions are a violation of our [Code of Conduct](${codeOfConduct}).`;
 

.github/workflows/pr-bot.yml

@@ -25,6 +25,10 @@ jobs:
   pr-bot:
     name: Automated PR Bot
     runs-on: ubuntu-latest
+    # Runs after Anti-slop so the welcome comment can see whether the PR was closed
+    # instead of racing it. Still runs if that job fails, so labeling is not lost.
+    needs: Anti-slop
+    if: ${{ !cancelled() }}
     permissions:
       contents: read
       pull-requests: write
@@ -99,8 +103,25 @@ jobs:
         uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
         with:
           script: |
-            const pr = context.payload.pull_request;
+            const user = context.payload.pull_request.user.login;
-            const user = pr.user.login;
+
+            // Re-read the PR: Anti-slop may have closed and labeled it after the webhook
+            const { data: pr } = await github.rest.pulls.get({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              pull_number: context.payload.pull_request.number,
+            });
+
+            if (pr.state === 'closed') {
+              core.info('Skipping comment: PR is already closed');
+              return;
+            }
+
+            const labels = pr.labels.map((label) => (typeof label === 'string' ? label : label.name));
+            if (labels.includes('ai')) {
+              core.info('Skipping comment: PR is labeled ai');
+              return;
+            }
 
             const { data: members } = await github.rest.orgs.listMembers({
               org: 'paperless-ngx',

src-ui/messages.xlf

@@ -501,15 +501,43 @@
           <context context-type="linenumber">30</context>
         </context-group>
       </trans-unit>
+      <trans-unit id="7057826840809102816" datatype="html">
+        <source>This value overrides <x id="INTERPOLATION" equiv-text="{{option.config_key}}"/>, which is set outside Paperless.</source>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/admin/config/config.component.html</context>
+          <context context-type="linenumber">35</context>
+        </context-group>
+      </trans-unit>
+      <trans-unit id="7221396516204435584" datatype="html">
+        <source><x id="INTERPOLATION" equiv-text="{{option.config_key}}"/> is set outside Paperless. Enter a value here to override it.</source>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/admin/config/config.component.html</context>
+          <context context-type="linenumber">37</context>
+        </context-group>
+      </trans-unit>
+      <trans-unit id="8318849619178340389" datatype="html">
+        <source>Use the externally configured value</source>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/admin/config/config.component.html</context>
+          <context context-type="linenumber">42</context>
+        </context-group>
+      </trans-unit>
+      <trans-unit id="6032629623003430385" datatype="html">
+        <source>Reset to external</source>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/admin/config/config.component.html</context>
+          <context context-type="linenumber">43</context>
+        </context-group>
+      </trans-unit>
       <trans-unit id="7808756054397155068" datatype="html">
         <source>Reset</source>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/admin/config/config.component.html</context>
-          <context context-type="linenumber">34</context>
+          <context context-type="linenumber">46</context>
         </context-group>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/admin/config/config.component.html</context>
-          <context context-type="linenumber">35</context>
+          <context context-type="linenumber">47</context>
         </context-group>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/admin/settings/settings.component.html</context>
@@ -520,7 +548,7 @@
         <source>Enable</source>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/admin/config/config.component.html</context>
-          <context context-type="linenumber">43</context>
+          <context context-type="linenumber">56</context>
         </context-group>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/common/profile-edit-dialog/profile-edit-dialog.component.html</context>
@@ -531,7 +559,7 @@
         <source>Cancel</source>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/admin/config/config.component.html</context>
-          <context context-type="linenumber">67,68</context>
+          <context context-type="linenumber">80,81</context>
         </context-group>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/admin/settings/settings.component.html</context>
@@ -610,7 +638,7 @@
         <source>Save</source>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/admin/config/config.component.html</context>
-          <context context-type="linenumber">70,71</context>
+          <context context-type="linenumber">83,84</context>
         </context-group>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/admin/settings/settings.component.html</context>
@@ -681,42 +709,42 @@
         <source>Error retrieving config</source>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/admin/config/config.component.ts</context>
-          <context context-type="linenumber">117</context>
+          <context context-type="linenumber">118</context>
         </context-group>
       </trans-unit>
       <trans-unit id="1172622527269118932" datatype="html">
         <source>Invalid JSON</source>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/admin/config/config.component.ts</context>
-          <context context-type="linenumber">143</context>
+          <context context-type="linenumber">144</context>
         </context-group>
       </trans-unit>
       <trans-unit id="5103146006962696736" datatype="html">
         <source>Configuration updated</source>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/admin/config/config.component.ts</context>
-          <context context-type="linenumber">187</context>
+          <context context-type="linenumber">193</context>
         </context-group>
       </trans-unit>
       <trans-unit id="1664963291286452273" datatype="html">
         <source>An error occurred updating configuration</source>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/admin/config/config.component.ts</context>
-          <context context-type="linenumber">192</context>
+          <context context-type="linenumber">198</context>
         </context-group>
       </trans-unit>
       <trans-unit id="2653081282186526824" datatype="html">
         <source>File successfully updated</source>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/admin/config/config.component.ts</context>
-          <context context-type="linenumber">214</context>
+          <context context-type="linenumber">220</context>
         </context-group>
       </trans-unit>
       <trans-unit id="5902783625859504265" datatype="html">
         <source>An error occurred uploading file</source>
         <context-group purpose="location">
           <context context-type="sourcefile">src/app/components/admin/config/config.component.ts</context>
-          <context context-type="linenumber">219</context>
+          <context context-type="linenumber">225</context>
         </context-group>
       </trans-unit>
       <trans-unit id="4804785061014590286" datatype="html">

src-ui/src/app/components/admin/config/config.component.html

@@ -23,18 +23,31 @@
                                     <div class="col">
                                         <div class="card bg-light">
                                             <div class="card-body">
-                                                <div class="card-title d-flex align-items-center">
+                                                <div class="card-title d-flex align-items-center flex-wrap">
                                                     <h6 class="mb-0">
                                                       {{option.title}}
                                                     </h6>
                                                     <a class="btn btn-sm btn-link" title="Read the documentation about this setting" i18n-title [href]="getDocsUrl(option.config_key)" target="_blank" referrerpolicy="no-referrer">
                                                         <i-bs name="info-circle"></i-bs>
                                                     </a>
+                                                    @if (isExternallyConfigured(option.config_key)) {
                                                       @if (isSet(option.key)) {
+                                                        <span class="badge rounded-pill bg-body-secondary text-dark fw-normal" title="This value overrides {{option.config_key}}, which is set outside Paperless." i18n-title>Overrides external</span>
+                                                      } @else {
+                                                        <span class="badge rounded-pill bg-body-secondary text-dark fw-normal" title="{{option.config_key}} is set outside Paperless. Enter a value here to override it." i18n-title>Set externally</span>
+                                                      }
+                                                    }
+                                                    @if (isSet(option.key)) {
+                                                      @if (isExternallyConfigured(option.config_key)) {
+                                                        <button type="button" class="btn btn-sm btn-link text-danger ms-auto pe-0" title="Use the externally configured value" i18n-title (click)="resetOption(option.key)">
+                                                            <i-bs class="me-1" name="x"></i-bs><ng-container i18n>Reset to external</ng-container>
+                                                        </button>
+                                                      } @else {
                                                         <button type="button" class="btn btn-sm btn-link text-danger ms-auto pe-0" title="Reset" i18n-title (click)="resetOption(option.key)">
                                                             <i-bs class="me-1" name="x"></i-bs><ng-container i18n>Reset</ng-container>
                                                         </button>
                                                       }
+                                                    }
                                                 </div>
                                                 <div class="mb-n3">
                                                     @switch (option.type) {

src-ui/src/app/components/admin/config/config.component.spec.ts

@@ -163,6 +163,19 @@ describe('ConfigComponent', () => {
     expect(component.configForm.get('barcodes_enabled').value).toBeNull()
   })
 
+  it('should identify externally configured options', () => {
+    component.externallyConfiguredVariables = new Set([
+      'PAPERLESS_OCR_LANGUAGE',
+    ])
+
+    expect(
+      component.isExternallyConfigured('PAPERLESS_OCR_LANGUAGE')
+    ).toBeTruthy()
+    expect(
+      component.isExternallyConfigured('PAPERLESS_OCR_OUTPUT_TYPE')
+    ).toBeFalsy()
+  })
+
   it('should group options into sections within a category, or not', () => {
     const sections = component.getCategorySections(ConfigCategory.OCR)
     expect(sections).toEqual([null, ConfigSection.RemoteOCR])

src-ui/src/app/components/admin/config/config.component.ts

@@ -69,6 +69,7 @@ export class ConfigComponent
   public configForm = new FormGroup({})
 
   public errors = {}
+  public externallyConfiguredVariables = new Set<string>()
 
   get optionCategories(): string[] {
     return Object.values(ConfigCategory)
@@ -152,6 +153,9 @@ export class ConfigComponent
   }
 
   private initialize(config: PaperlessConfig) {
+    this.externallyConfiguredVariables = new Set(
+      config.externally_configured_variables ?? []
+    )
     if (!this.store) {
       this.store = new BehaviorSubject(config)
 
@@ -162,7 +166,9 @@ export class ConfigComponent
           this.configForm.patchValue(state, { emitEvent: false })
         })
 
-      this.isDirty$ = dirtyCheck(this.configForm, this.store.asObservable())
+      this.isDirty$ = dirtyCheck(this.configForm, this.store.asObservable(), {
+        excludeKeys: ['externally_configured_variables'],
+      })
     }
     this.configForm.patchValue(config)
 
@@ -227,6 +233,10 @@ export class ConfigComponent
     return this.configForm.get(key).value != null
   }
 
+  public isExternallyConfigured(configKey: string): boolean {
+    return this.externallyConfiguredVariables.has(configKey)
+  }
+
   public resetOption(key: string) {
     this.configForm.get(key).setValue(null)
   }

src-ui/src/app/data/paperless-config.ts

@@ -422,6 +422,7 @@ export const PaperlessConfigOptions: ConfigOption[] = [
 ]
 
 export interface PaperlessConfig extends ObjectWithId {
+  externally_configured_variables: string[]
   output_type: OutputTypeConfig
   pages: number
   language: string

src/documents/serialisers.py

@@ -674,6 +674,9 @@ class TagSerializer(MatchingModelSerializer, OwnedObjectSerializer):
             ordering = ordering or (Lower("name"),)
             children = children.order_by(*ordering)
 
+        if not children:
+            return []
+
         serializer = TagSerializer(
             children,
             many=True,

src/documents/tests/test_api_app_config.py

@@ -35,6 +35,7 @@ class TestApiAppConfig(DirectoriesMixin, APITestCase):
         THEN:
             - Existing config
         """
+        with patch.dict("os.environ", {}, clear=True):
             response = self.client.get(self.ENDPOINT, format="json")
 
         self.assertEqual(response.status_code, status.HTTP_200_OK)
@@ -45,6 +46,7 @@ class TestApiAppConfig(DirectoriesMixin, APITestCase):
             response.data[0],
             {
                 "id": 1,
+                "externally_configured_variables": [],
                 "output_type": None,
                 "pages": None,
                 "language": None,
@@ -91,6 +93,31 @@ class TestApiAppConfig(DirectoriesMixin, APITestCase):
             },
         )
 
+    def test_api_get_config_reports_external_configuration_without_values(self) -> None:
+        with patch.dict(
+            "os.environ",
+            {
+                "PAPERLESS_OCR_LANGUAGE": "eng",
+                "PAPERLESS_REMOTE_OCR_API_KEY": "secret-value",
+                "PAPERLESS_FUTURE_SETTING": "future-value",
+                "UNRELATED_SETTING": "unrelated-value",
+            },
+            clear=True,
+        ):
+            response = self.client.get(self.ENDPOINT, format="json")
+
+        self.assertCountEqual(
+            response.data[0]["externally_configured_variables"],
+            [
+                "PAPERLESS_FUTURE_SETTING",
+                "PAPERLESS_OCR_LANGUAGE",
+                "PAPERLESS_REMOTE_OCR_API_KEY",
+            ],
+        )
+        self.assertNotContains(response, "secret-value")
+        self.assertNotContains(response, "future-value")
+        self.assertNotContains(response, "UNRELATED_SETTING")
+
     def test_api_get_ui_settings_with_config(self) -> None:
         """
         GIVEN:

src/locale/en_US/LC_MESSAGES/django.po

@@ -2,7 +2,7 @@ msgid ""
 msgstr ""
 "Project-Id-Version: paperless-ngx\n"
 "Report-Msgid-Bugs-To: \n"
-"POT-Creation-Date: 2026-09-07 20:47+0000\n"
+"POT-Creation-Date: 2026-09-08 15:31+0000\n"
 "PO-Revision-Date: 2022-02-17 04:17\n"
 "Last-Translator: \n"
 "Language-Team: English\n"
@@ -1631,49 +1631,49 @@ msgstr ""
 msgid "workflow runs"
 msgstr ""
 
-#: documents/serialisers.py:524 documents/serialisers.py:878
+#: documents/serialisers.py:524 documents/serialisers.py:881
-#: documents/serialisers.py:2838 documents/views.py:314 documents/views.py:2624
+#: documents/serialisers.py:2841 documents/views.py:314 documents/views.py:2624
 #: paperless_mail/serialisers.py:156
 msgid "Insufficient permissions."
 msgstr ""
 
-#: documents/serialisers.py:714
+#: documents/serialisers.py:717
 msgid "Invalid color."
 msgstr ""
 
-#: documents/serialisers.py:2315
+#: documents/serialisers.py:2318
 #, python-format
 msgid "File type %(type)s not supported"
 msgstr ""
 
-#: documents/serialisers.py:2359
+#: documents/serialisers.py:2362
 #, python-format
 msgid "Custom field id must be an integer: %(id)s"
 msgstr ""
 
-#: documents/serialisers.py:2366
+#: documents/serialisers.py:2369
 #, python-format
 msgid "Custom field with id %(id)s does not exist"
 msgstr ""
 
-#: documents/serialisers.py:2383 documents/serialisers.py:2393
+#: documents/serialisers.py:2386 documents/serialisers.py:2396
 msgid ""
 "Custom fields must be a list of integers or an object mapping ids to values."
 msgstr ""
 
-#: documents/serialisers.py:2388
+#: documents/serialisers.py:2391
 msgid "Some custom fields don't exist or were specified twice."
 msgstr ""
 
-#: documents/serialisers.py:2535
+#: documents/serialisers.py:2538
 msgid "Invalid variable detected."
 msgstr ""
 
-#: documents/serialisers.py:2894
+#: documents/serialisers.py:2897
 msgid "Duplicate document identifiers are not allowed."
 msgstr ""
 
-#: documents/serialisers.py:2924 documents/views.py:4626
+#: documents/serialisers.py:2927 documents/views.py:4626
 #, python-format
 msgid "Documents not found: %(ids)s"
 msgstr ""

src/paperless/serialisers.py

@@ -1,4 +1,5 @@
 import logging
+import os
 from io import BytesIO
 
 import magic
@@ -212,6 +213,7 @@ class ProfileSerializer(PasswordValidationMixin, serializers.ModelSerializer[Use
 class ApplicationConfigurationSerializer(
     serializers.ModelSerializer[ApplicationConfiguration],
 ):
+    externally_configured_variables = serializers.SerializerMethodField()
     user_args = serializers.JSONField(binary=True, allow_null=True)
     barcode_tag_mapping = serializers.JSONField(binary=True, allow_null=True)
     llm_api_key = ObfuscatedPasswordField(
@@ -227,6 +229,12 @@ class ApplicationConfigurationSerializer(
 
     OBFUSCATED_FIELDS = ("llm_api_key", "remote_ocr_api_key")
 
+    def get_externally_configured_variables(
+        self,
+        instance: ApplicationConfiguration,
+    ) -> list[str]:
+        return sorted(name for name in os.environ if name.startswith("PAPERLESS_"))
+
     def run_validation(self, data):
         # Empty strings treated as None to avoid unexpected behavior
         if "user_args" in data and data["user_args"] == "":

src/paperless_ai/ai_classifier.py

@@ -4,21 +4,24 @@ from django.conf import settings
 from django.contrib.auth.models import User
 
 from documents.models import Document
-from documents.permissions import get_objects_for_user_owner_aware
+from documents.permissions import permitted_object_ids
+from documents.permissions import restrict_queryset_to_visible
+from documents.permissions import user_is_unrestricted
 from paperless.config import AIConfig
 from paperless_ai.base_model import ClassificationSuggestions
 from paperless_ai.base_model import TaxonomyChoiceDict
 from paperless_ai.base_model import classification_suggestions_to_model
 from paperless_ai.client import AIClient
 from paperless_ai.db import db_connection_released
-from paperless_ai.indexing import _node_document_ids
 from paperless_ai.indexing import retrieve_similar_nodes
 from paperless_ai.indexing import truncate_content
 from paperless_ai.prompts.context import ClassificationPromptContext
 from paperless_ai.prompts.context import LocalizationPromptContext
 from paperless_ai.prompts.context import RagContextPromptContext
 from paperless_ai.prompts.render import render_prompt
+from paperless_ai.taxonomy import SimilarDocument
 from paperless_ai.taxonomy import TaxonomyCandidates
+from paperless_ai.taxonomy import _node_document_weights
 from paperless_ai.taxonomy import build_taxonomy_candidates
 from paperless_ai.taxonomy import empty_taxonomy_candidates
 from paperless_ai.taxonomy import format_taxonomy_for_prompt
@@ -37,6 +40,48 @@ logger = logging.getLogger("paperless_ai.rag_classifier")
 TAXONOMY_CANDIDATE_TOP_K = 15
 
 
+def _fulltext_similar_documents(
+    document: Document,
+    user: User | None,
+    top_k: int,
+) -> list[SimilarDocument]:
+    """Rank-based fallback when no embedding backend is configured. Uses
+    Tantivy's "More Like This" (term-overlap similarity) instead of vector
+    similarity - cruder, but far better than no candidates at all.
+    more_like_this_ids returns only a ranked ID list, no scores, so weight is
+    synthesized from rank (descending from top_k) rather than claiming a
+    similarity magnitude that doesn't exist. An unrestricted user (none, or an
+    active superuser - see user_is_unrestricted) is normalized to ``None``
+    before calling, since the backend's permission filter has no superuser
+    short-circuit of its own. Results are re-checked with
+    restrict_queryset_to_visible() since Tantivy's indexed permission fields
+    lag the DB via async reindexing.
+    """
+    from documents.search import get_backend
+
+    unrestricted = user_is_unrestricted(user)
+    search_user = None if unrestricted else user
+    backend = get_backend()
+    similar_ids = backend.more_like_this_ids(
+        document.pk,
+        user=search_user,
+        limit=top_k,
+    )
+    if not unrestricted:
+        allowed_ids = set(
+            restrict_queryset_to_visible(
+                Document.objects.filter(pk__in=similar_ids),
+                user,
+                "view_document",
+            ).values_list("pk", flat=True),
+        )
+        similar_ids = [doc_id for doc_id in similar_ids if doc_id in allowed_ids]
+    return [
+        SimilarDocument(document_id=doc_id, weight=float(top_k - rank))
+        for rank, doc_id in enumerate(similar_ids)
+    ]
+
+
 def get_language_name(language_code: str) -> str:
     normalized_language_code = language_code.lower()
     for code, name in settings.LANGUAGES:
@@ -136,43 +181,52 @@ def get_taxonomy_context(
     user: User | None = None,
     max_docs: int = 5,
 ) -> tuple[TaxonomyCandidates, str]:
-    """One retrieval feeds both taxonomy candidates and RAG text context.
+    """One retrieval feeds both taxonomy candidates and RAG text context. Uses
-    On any retrieval failure, degrades to empty candidates/context rather than
+    vector similarity when an embedding backend is configured, otherwise
-    propagating the exception - a vector-store outage should not block
+    falls back to Tantivy full-text "More Like This" similarity - see
-    classification, only its RAG-assisted enrichment.
+    _fulltext_similar_documents. On any retrieval failure, degrades to empty
+    candidates/context rather than propagating the exception - neither a
+    vector-store outage nor a search-index issue should block classification,
+    only its context-assisted enrichment.
     """
+    ai_config = AIConfig()
     try:
-        # None means "no restriction" to retrieve_similar_nodes. A superuser
+        if ai_config.llm_embedding_backend:
-        # (like no user at all) can see every document, so skip materializing
+            # None means "no restriction" to retrieve_similar_nodes. An
-        # every visible pk into a Python list and passing it through as an IN
+            # unrestricted user (no user at all, or an active superuser -- see
-        # filter: for a large library that is a wasted quadratic scan in the
+            # user_is_unrestricted) can see every document, so skip
-        # vector store at best, and past ~32,763 documents a hard
+            # materializing every visible pk into a Python list and passing it
-        # sqlite3.OperationalError (SQLite's bound-parameter limit) at worst.
+            # through as an IN filter: for a large library that is a wasted
-        # get_objects_for_user_owner_aware() would return every Document for a
+            # quadratic scan in the vector store at best, and past ~32,763
-        # superuser anyway (guardian's own with_superuser shortcut), so this
+            # documents a hard sqlite3.OperationalError (SQLite's
-        # changes nothing about which documents are considered -- only how we
+            # bound-parameter limit) at worst.
-        # get there.
+            # permitted_object_ids() has its own superuser shortcut that would
+            # return every Document's id anyway, so this changes nothing about
+            # which documents are considered -- only how we get there.
             visible_document_ids = (
                 None
-            if user is None or user.is_superuser
+                if user_is_unrestricted(user)
-            else list(
+                else list(permitted_object_ids(user, Document, "view_document"))
-                get_objects_for_user_owner_aware(
-                    user,
-                    "view_document",
-                    Document,
-                ).values_list("pk", flat=True),
-            )
             )
             nodes = retrieve_similar_nodes(
                 document,
                 top_k=TAXONOMY_CANDIDATE_TOP_K,
                 document_ids=visible_document_ids,
             )
+            similar_documents = _node_document_weights(nodes)
+        else:
+            # See _fulltext_similar_documents: it applies its own permission
+            # filter via `user`, so no visible-document-id list is needed here.
+            similar_documents = _fulltext_similar_documents(
+                document,
+                user,
+                top_k=TAXONOMY_CANDIDATE_TOP_K,
+            )
 
-        candidates = build_taxonomy_candidates(nodes, user)
+        candidates = build_taxonomy_candidates(similar_documents, user)
 
-        # ``nodes`` are already ordered by descending vector similarity; don't lose it.
+        # similar_documents is already ordered by descending weight; don't lose it.
-        similar_document_ids = list(dict.fromkeys(_node_document_ids(nodes)))
+        similar_document_ids = [s["document_id"] for s in similar_documents]
         similar_documents_by_id = Document.objects.in_bulk(similar_document_ids)
         similar_docs = [
             similar_documents_by_id[document_id]
@@ -186,8 +240,8 @@ def get_taxonomy_context(
             context_blocks.append(f"TITLE: {title}\n{text}")
     except Exception:
         logger.exception(
-            "Failed to retrieve RAG neighbours for document %s; continuing "
+            "Failed to retrieve similar-document context for document %s; "
-            "without taxonomy candidates or similar-document context.",
+            "continuing without taxonomy candidates or similar-document context.",
             document.pk,
         )
         return empty_taxonomy_candidates(), ""
@@ -241,7 +295,6 @@ def get_ai_document_classification(
 ) -> ClassificationSuggestions:
     ai_config = AIConfig()
 
-    if ai_config.llm_embedding_backend:
     candidates, context = get_taxonomy_context(document, user)
     prompt = build_prompt_with_rag(
         document,
@@ -249,9 +302,6 @@ def get_ai_document_classification(
         candidates=candidates,
         context=context,
     )
-    else:
-        candidates = empty_taxonomy_candidates()
-        prompt = build_prompt_without_rag(document, ai_config, candidates=candidates)
 
     client = AIClient()
     # Hand the pooled DB connection back while the (slow) LLM query runs so it

src/paperless_ai/indexing.py

@@ -721,20 +721,3 @@ def retrieve_similar_nodes(
             continue
         filtered.append(node)
     return filtered
-
-
-def _node_document_ids(nodes: list["NodeWithScore"]) -> list[int]:
-    document_ids: list[int] = []
-    for node in nodes:
-        document_id = node.metadata.get("document_id")
-        if document_id is None:  # pragma: no cover
-            # See the matching guard in retrieve_similar_nodes() above.
-            continue
-        try:
-            document_ids.append(int(document_id))
-        except ValueError:  # pragma: no cover
-            logger.warning(
-                "Skipping LLM index result with invalid document_id %r.",
-                document_id,
-            )
-    return document_ids

src/paperless_ai/taxonomy.py

@@ -31,6 +31,11 @@ class TaxonomyCandidate(TypedDict):
     weight: float
 
 
+class SimilarDocument(TypedDict):
+    document_id: int
+    weight: float
+
+
 class TaxonomyCandidates(TypedDict):
     tags: list[TaxonomyCandidate]
     document_types: list[TaxonomyCandidate]
@@ -49,10 +54,10 @@ def empty_taxonomy_candidates() -> TaxonomyCandidates:
     )
 
 
-def _node_document_weights(nodes: list["NodeWithScore"]) -> dict[int, float]:
+def _node_document_weights(nodes: list["NodeWithScore"]) -> list[SimilarDocument]:
-    """document_id -> that node's similarity score, summed if a document_id
+    """Sum each node's similarity score into its document_id (a document can
-    appears more than once across the retrieved nodes (e.g. multiple chunks
+    appear via multiple chunks/nodes) and return one SimilarDocument per
-    of the same source document)."""
+    distinct document_id."""
     weights: dict[int, float] = defaultdict(float)
     for node in nodes:
         document_id = node.metadata.get("document_id")
@@ -65,7 +70,14 @@ def _node_document_weights(nodes: list["NodeWithScore"]) -> dict[int, float]:
             weights[int(document_id)] += float(node.score or 0.0)
         except (TypeError, ValueError):  # pragma: no cover
             continue
-    return weights
+    return sorted(
+        (
+            SimilarDocument(document_id=document_id, weight=weight)
+            for document_id, weight in weights.items()
+        ),
+        key=lambda similar: similar["weight"],
+        reverse=True,
+    )
 
 
 def _visible_ranked_candidates(
@@ -101,21 +113,26 @@ def _visible_ranked_candidates(
 
 
 def build_taxonomy_candidates(
-    nodes: list["NodeWithScore"],
+    similar_documents: list[SimilarDocument],
     user: User | None,
 ) -> TaxonomyCandidates:
-    """Resolve each neighbour node's document_id to a live Document, read its
+    """Resolve each similar document's id to a live Document, read its
-    *current* tags/type/correspondent/storage_path via the ORM (never the
+    *current* tags/type/correspondent/storage_path via the ORM (never any
-    possibly-stale names cached in vector-index node metadata), weight each
+    possibly-stale names an adapter's source might have cached), weight each
-    distinct taxonomy object by aggregate neighbour similarity, permission-filter
+    distinct taxonomy object by aggregate similarity weight, permission-filter
     against what ``user`` can see, and return each category ranked by weight
-    and capped.
+    and capped. ``similar_documents`` may come from either the vector-RAG
+    adapter or the full-text fallback adapter - both produce this same shape.
     """
-
+    if not similar_documents:
-    document_weights = _node_document_weights(nodes)
-    if not document_weights:
         return empty_taxonomy_candidates()
 
+    # Both adapters guarantee at most one SimilarDocument per document_id, so
+    # this never silently drops a duplicate's weight.
+    document_weights: dict[int, float] = {
+        s["document_id"]: s["weight"] for s in similar_documents
+    }
+
     # Only .tags.all() needs prefetching (a reverse M2M, one extra query for
     # the whole batch). document_type/correspondent/storage_path are read
     # below via their *_id columns (neighbour.document_type_id, etc.), which

src/paperless_ai/tests/test_ai_classifier.py

@@ -1,4 +1,5 @@
 import datetime
+from collections.abc import Generator
 from types import SimpleNamespace
 from unittest.mock import MagicMock
 from unittest.mock import patch
@@ -6,18 +7,24 @@ from unittest.mock import patch
 import pytest
 import pytest_mock
 from django.test import override_settings
+from guardian.shortcuts import assign_perm
+from guardian.shortcuts import remove_perm
 
 from documents.models import Document
+from documents.search import TantivyBackend
 from documents.tests.factories import DocumentFactory
 from documents.tests.factories import TagFactory
 from documents.tests.factories import UserFactory
 from paperless.config import AIConfig
+from paperless_ai.ai_classifier import TAXONOMY_CANDIDATE_TOP_K
+from paperless_ai.ai_classifier import _fulltext_similar_documents
 from paperless_ai.ai_classifier import build_localization_prompt
 from paperless_ai.ai_classifier import build_prompt_with_rag
 from paperless_ai.ai_classifier import build_prompt_without_rag
 from paperless_ai.ai_classifier import get_ai_document_classification
 from paperless_ai.ai_classifier import get_language_name
 from paperless_ai.ai_classifier import get_taxonomy_context
+from paperless_ai.taxonomy import SimilarDocument
 from paperless_ai.taxonomy import TaxonomyCandidate
 from paperless_ai.taxonomy import TaxonomyCandidates
 
@@ -220,12 +227,10 @@ def test_use_rag_if_configured(
 
 @pytest.mark.django_db
 @patch("paperless_ai.client.AIClient.run_llm_query")
-@patch("paperless_ai.ai_classifier.build_prompt_without_rag")
+@patch("paperless_ai.ai_classifier.build_prompt_with_rag")
-@patch("paperless_ai.ai_classifier.AIConfig")
 @override_settings(LLM_BACKEND="ollama", LLM_MODEL="some_model")
-def test_use_without_rag_if_not_configured(
+def test_use_rag_prompt_even_without_embedding_backend(
-    mock_ai_config,
+    mock_build_prompt_with_rag,
-    mock_build_prompt_without_rag,
     mock_run_llm_query,
     mock_document,
 ):
@@ -235,13 +240,13 @@ def test_use_without_rag_if_not_configured(
     WHEN:
         - get_ai_document_classification() is called
     THEN:
-        - The non-RAG prompt builder is used
+        - The RAG-context prompt builder is still used (fed by the full-text
+          fallback's context/candidates instead of the vector store's)
     """
-    mock_ai_config.return_value.llm_embedding_backend = None
+    mock_build_prompt_with_rag.return_value = "Prompt with RAG"
-    mock_build_prompt_without_rag.return_value = "Prompt without RAG"
     mock_run_llm_query.return_value = NESTED_SUGGESTIONS
     get_ai_document_classification(mock_document)
-    mock_build_prompt_without_rag.assert_called_once()
+    mock_build_prompt_with_rag.assert_called_once()
 
 
 @pytest.mark.django_db
@@ -320,6 +325,7 @@ def test_build_localization_prompt_preserves_unicode_characters():
 
 
 @pytest.mark.django_db
+@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
 def test_get_taxonomy_context_assembles_rag_text_and_candidates():
     """
     GIVEN:
@@ -354,6 +360,7 @@ def test_get_taxonomy_context_assembles_rag_text_and_candidates():
 
 
 @pytest.mark.django_db
+@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
 def test_get_taxonomy_context_preserves_similarity_order_and_distinct_documents():
     """
     GIVEN:
@@ -424,6 +431,7 @@ def test_get_taxonomy_context_preserves_similarity_order_and_distinct_documents(
 
 
 @pytest.mark.django_db
+@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
 def test_get_taxonomy_context_no_similar_docs():
     """
     GIVEN:
@@ -447,6 +455,67 @@ def test_get_taxonomy_context_no_similar_docs():
     }
 
 
+@pytest.mark.django_db
+def test_get_taxonomy_context_uses_fulltext_fallback_when_no_embedding_backend(
+    mocker: pytest_mock.MockerFixture,
+) -> None:
+    """
+    GIVEN:
+        - No LLM embedding backend is configured (the default test settings)
+    WHEN:
+        - get_taxonomy_context() is called
+    THEN:
+        - _fulltext_similar_documents() is called with the document, the user
+          and TAXONOMY_CANDIDATE_TOP_K
+        - retrieve_similar_nodes() (the vector path) is never called
+    """
+    document = DocumentFactory.create(content="Some content")
+    mock_fulltext = mocker.patch(
+        "paperless_ai.ai_classifier._fulltext_similar_documents",
+        return_value=[],
+    )
+    mock_retrieve = mocker.patch("paperless_ai.ai_classifier.retrieve_similar_nodes")
+
+    get_taxonomy_context(document, user=None)
+
+    mock_fulltext.assert_called_once_with(
+        document,
+        None,
+        top_k=TAXONOMY_CANDIDATE_TOP_K,
+    )
+    mock_retrieve.assert_not_called()
+
+
+@pytest.mark.django_db
+@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
+def test_get_taxonomy_context_uses_vector_path_when_embedding_backend_configured(
+    mocker: pytest_mock.MockerFixture,
+) -> None:
+    """
+    GIVEN:
+        - An LLM embedding backend is configured
+    WHEN:
+        - get_taxonomy_context() is called
+    THEN:
+        - retrieve_similar_nodes() (the vector path) is called
+        - _fulltext_similar_documents() (the no-embedding-backend fallback)
+          is never called
+    """
+    document = DocumentFactory.create(content="Some content")
+    mock_retrieve = mocker.patch(
+        "paperless_ai.ai_classifier.retrieve_similar_nodes",
+        return_value=[],
+    )
+    mock_fulltext = mocker.patch(
+        "paperless_ai.ai_classifier._fulltext_similar_documents",
+    )
+
+    get_taxonomy_context(document, user=None)
+
+    mock_retrieve.assert_called_once()
+    mock_fulltext.assert_not_called()
+
+
 class TestGetTaxonomyContextVisibility:
     """get_taxonomy_context must not materialize every visible document id
     for a user who can already see the whole library: a superuser (like no
@@ -459,6 +528,7 @@ class TestGetTaxonomyContextVisibility:
     """
 
     @pytest.mark.django_db
+    @override_settings(LLM_EMBEDDING_BACKEND="huggingface")
     def test_skips_permission_lookup_for_superuser(
         self,
         mocker: pytest_mock.MockerFixture,
@@ -477,17 +547,18 @@ class TestGetTaxonomyContextVisibility:
             "paperless_ai.ai_classifier.retrieve_similar_nodes",
             return_value=[],
         )
-        mock_get_objects = mocker.patch(
+        mock_permitted = mocker.patch(
-            "paperless_ai.ai_classifier.get_objects_for_user_owner_aware",
+            "paperless_ai.ai_classifier.permitted_object_ids",
         )
         user = UserFactory.create(is_superuser=True)
 
         get_taxonomy_context(document, user)
 
-        mock_get_objects.assert_not_called()
+        mock_permitted.assert_not_called()
         assert mock_retrieve.call_args.kwargs["document_ids"] is None
 
     @pytest.mark.django_db
+    @override_settings(LLM_EMBEDDING_BACKEND="huggingface")
     def test_skips_permission_lookup_when_no_user(
         self,
         mocker: pytest_mock.MockerFixture,
@@ -506,16 +577,17 @@ class TestGetTaxonomyContextVisibility:
             "paperless_ai.ai_classifier.retrieve_similar_nodes",
             return_value=[],
         )
-        mock_get_objects = mocker.patch(
+        mock_permitted = mocker.patch(
-            "paperless_ai.ai_classifier.get_objects_for_user_owner_aware",
+            "paperless_ai.ai_classifier.permitted_object_ids",
         )
 
         get_taxonomy_context(document, None)
 
-        mock_get_objects.assert_not_called()
+        mock_permitted.assert_not_called()
         assert mock_retrieve.call_args.kwargs["document_ids"] is None
 
     @pytest.mark.django_db
+    @override_settings(LLM_EMBEDDING_BACKEND="huggingface")
     def test_restricts_to_visible_documents_for_non_superuser(
         self,
         mocker: pytest_mock.MockerFixture,
@@ -526,7 +598,7 @@ class TestGetTaxonomyContextVisibility:
         WHEN:
             - get_taxonomy_context() is called
         THEN:
-            - The user's visible document ids are looked up and passed to
+            - The user's permitted document ids are looked up and passed to
               retrieve_similar_nodes() as a restriction
         """
         document = DocumentFactory.create(content="Some content")
@@ -534,21 +606,232 @@ class TestGetTaxonomyContextVisibility:
             "paperless_ai.ai_classifier.retrieve_similar_nodes",
             return_value=[],
         )
-        mock_queryset = mocker.MagicMock()
+        mock_permitted = mocker.patch(
-        mock_queryset.values_list.return_value = [1, 2, 3]
+            "paperless_ai.ai_classifier.permitted_object_ids",
-        mock_get_objects = mocker.patch(
+            return_value=[1, 2, 3],
-            "paperless_ai.ai_classifier.get_objects_for_user_owner_aware",
-            return_value=mock_queryset,
         )
         user = UserFactory.create(is_superuser=False)
 
         get_taxonomy_context(document, user)
 
-        mock_get_objects.assert_called_once_with(user, "view_document", Document)
+        mock_permitted.assert_called_once_with(user, Document, "view_document")
         assert mock_retrieve.call_args.kwargs["document_ids"] == [1, 2, 3]
 
 
 @pytest.mark.django_db
+class TestFulltextSimilarDocuments:
+    """_fulltext_similar_documents is the no-embedding-backend fallback: it
+    asks the Tantivy full-text index for "More Like This" neighbours instead
+    of the vector store, and synthesizes a rank-based weight since Tantivy's
+    more_like_this_ids returns only an ordered id list, no scores.
+    """
+
+    @pytest.fixture
+    def fulltext_backend(
+        self,
+        mocker: pytest_mock.MockerFixture,
+    ) -> Generator[TantivyBackend, None, None]:
+        """An in-memory Tantivy backend, wired up as the module-level
+        singleton _fulltext_similar_documents resolves via get_backend()."""
+        backend = TantivyBackend(path=None)
+        backend.open()
+        mocker.patch("documents.search.get_backend", return_value=backend)
+        try:
+            yield backend
+        finally:
+            backend.close()
+
+    def test_ranks_by_rank_based_weight_descending(
+        self,
+        fulltext_backend: TantivyBackend,
+    ) -> None:
+        """
+        GIVEN:
+            - A source document and two similar documents indexed in Tantivy
+        WHEN:
+            - _fulltext_similar_documents() is called
+        THEN:
+            - Each result's weight reflects its rank (first result weighted
+              higher than the second), not a raw similarity score
+        """
+        source = DocumentFactory.create(content="quarterly financial report details")
+        first = DocumentFactory.create(content="quarterly financial report details")
+        second = DocumentFactory.create(content="financial report")
+        for doc in (source, first, second):
+            fulltext_backend.add_or_update(doc)
+
+        result = _fulltext_similar_documents(source, user=None, top_k=5)
+
+        assert len(result) == 2
+        weight_by_id = {s["document_id"]: s["weight"] for s in result}
+        assert weight_by_id[first.pk] > weight_by_id[second.pk]
+
+    def test_excludes_source_document(
+        self,
+        fulltext_backend: TantivyBackend,
+    ) -> None:
+        """
+        GIVEN:
+            - A source document indexed in Tantivy with no other documents
+        WHEN:
+            - _fulltext_similar_documents() is called
+        THEN:
+            - An empty list is returned - the source document is never its
+              own similar document
+        """
+        source = DocumentFactory.create(content="unique unrelated content")
+        fulltext_backend.add_or_update(source)
+
+        result = _fulltext_similar_documents(source, user=None, top_k=5)
+
+        assert result == []
+
+    def test_empty_index_returns_empty_list(
+        self,
+        fulltext_backend: TantivyBackend,
+    ) -> None:
+        """
+        GIVEN:
+            - A document that has never been indexed (fresh/empty Tantivy index)
+        WHEN:
+            - _fulltext_similar_documents() is called
+        THEN:
+            - An empty list is returned rather than raising
+        """
+        source = DocumentFactory.create(content="never indexed")
+
+        result = _fulltext_similar_documents(source, user=None, top_k=5)
+
+        assert result == []
+
+    def test_respects_top_k_limit(
+        self,
+        fulltext_backend: TantivyBackend,
+    ) -> None:
+        """
+        GIVEN:
+            - A source document and four similar documents indexed
+        WHEN:
+            - _fulltext_similar_documents() is called with top_k=2
+        THEN:
+            - At most 2 results are returned
+        """
+        source = DocumentFactory.create(content="shared overlapping keyword text")
+        fulltext_backend.add_or_update(source)
+        for _ in range(4):
+            fulltext_backend.add_or_update(
+                DocumentFactory.create(content="shared overlapping keyword text"),
+            )
+
+        result = _fulltext_similar_documents(source, user=None, top_k=2)
+
+        assert len(result) == 2
+
+    def test_result_shape_is_similar_document(
+        self,
+        fulltext_backend: TantivyBackend,
+    ) -> None:
+        """
+        GIVEN:
+            - A source document and one similar document indexed
+        WHEN:
+            - _fulltext_similar_documents() is called
+        THEN:
+            - Each result is a SimilarDocument (document_id + weight only)
+        """
+        source = DocumentFactory.create(content="shared content phrase")
+        other = DocumentFactory.create(content="shared content phrase")
+        fulltext_backend.add_or_update(source)
+        fulltext_backend.add_or_update(other)
+
+        result = _fulltext_similar_documents(source, user=None, top_k=5)
+
+        # rank 0 (the only/best result) with top_k=5 -> weight = top_k - rank = 5.0,
+        # per the "first result gets top_k, the last gets 1" formula.
+        assert result == [SimilarDocument(document_id=other.pk, weight=5.0)]
+
+    def test_superuser_sees_other_users_documents(
+        self,
+        fulltext_backend: TantivyBackend,
+    ) -> None:
+        """
+        GIVEN:
+            - A source document owned by one user and a similar document
+              owned by a different user, with no sharing between them
+        WHEN:
+            - _fulltext_similar_documents() is called with a superuser
+        THEN:
+            - The other user's document is still returned as a similar
+              document - a superuser must not be narrowed by the backend's
+              owner-based permission filter
+        """
+        owner = UserFactory.create()
+        other_owner = UserFactory.create()
+        superuser = UserFactory.create(is_superuser=True)
+        source = DocumentFactory.create(
+            content="shared content phrase",
+            owner=owner,
+        )
+        other = DocumentFactory.create(
+            content="shared content phrase",
+            owner=other_owner,
+        )
+        fulltext_backend.add_or_update(source)
+        fulltext_backend.add_or_update(other)
+
+        result = _fulltext_similar_documents(source, user=superuser, top_k=5)
+
+        assert [s["document_id"] for s in result] == [other.pk]
+
+    def test_excludes_stale_permitted_document_for_regular_user(
+        self,
+        fulltext_backend: TantivyBackend,
+    ) -> None:
+        """
+        GIVEN:
+            - A regular (non-superuser) user
+            - A similar document the user is permitted to view, and another
+              similar document indexed while the user still had view
+              permission but which has since had that permission revoked in
+              the database, i.e. the Tantivy index has stale permission data
+        WHEN:
+            - _fulltext_similar_documents() is called with that user
+        THEN:
+            - Only the still-permitted document is returned - the DB
+              re-check via restrict_queryset_to_visible() must catch the
+              document Tantivy's stale index still thinks is visible
+        """
+        owner = UserFactory.create()
+        viewer = UserFactory.create(is_superuser=False)
+        source = DocumentFactory.create(
+            content="shared content phrase",
+            owner=owner,
+        )
+        permitted = DocumentFactory.create(
+            content="shared content phrase",
+            owner=owner,
+        )
+        now_private = DocumentFactory.create(
+            content="shared content phrase",
+            owner=owner,
+        )
+        assign_perm("view_document", viewer, permitted)
+        assign_perm("view_document", viewer, now_private)
+        fulltext_backend.add_or_update(source)
+        fulltext_backend.add_or_update(permitted)
+        fulltext_backend.add_or_update(now_private)
+
+        # Revoke access after indexing, without reindexing: the index still
+        # carries viewer as a permitted viewer for `now_private`.
+        remove_perm("view_document", viewer, now_private)
+
+        result = _fulltext_similar_documents(source, user=viewer, top_k=5)
+
+        assert [s["document_id"] for s in result] == [permitted.pk]
+
+
+@pytest.mark.django_db
+@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
 @patch("paperless_ai.ai_classifier.retrieve_similar_nodes")
 def test_get_taxonomy_context_retrieval_failure_degrades_to_no_hints(mock_retrieve):
     """
@@ -575,6 +858,7 @@ def test_get_taxonomy_context_retrieval_failure_degrades_to_no_hints(mock_retrie
 
 
 @pytest.mark.django_db
+@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
 @patch("paperless_ai.ai_classifier.build_taxonomy_candidates")
 @patch("paperless_ai.ai_classifier.retrieve_similar_nodes")
 def test_get_taxonomy_context_candidate_building_failure_degrades_to_no_hints(

src/paperless_ai/tests/test_ai_indexing.py

@@ -1188,9 +1188,7 @@ class TestRetrieveSimilarNodesAgainstRealIndex:
 
         nodes = indexing.retrieve_similar_nodes(a, document_ids=[b.id])
 
-        assert all(
+        assert all(int(node.metadata["document_id"]) == b.id for node in nodes)
-            document_id == b.id for document_id in indexing._node_document_ids(nodes)
-        )
 
     def test_excludes_self(
         self,
@@ -1212,7 +1210,7 @@ class TestRetrieveSimilarNodesAgainstRealIndex:
 
         nodes = indexing.retrieve_similar_nodes(a, top_k=5)
 
-        assert set(indexing._node_document_ids(nodes)) == {b.id}
+        assert {int(node.metadata["document_id"]) for node in nodes} == {b.id}
 
     def test_excludes_self_with_multiple_chunks(
         self,
@@ -1235,4 +1233,4 @@ class TestRetrieveSimilarNodesAgainstRealIndex:
 
         nodes = indexing.retrieve_similar_nodes(a, top_k=3)
 
-        assert set(indexing._node_document_ids(nodes)) == {b.id}
+        assert {int(node.metadata["document_id"]) for node in nodes} == {b.id}

src/paperless_ai/tests/test_taxonomy.py

@@ -1,5 +1,4 @@
 import json
-from types import SimpleNamespace
 
 import pytest
 import pytest_mock
@@ -10,14 +9,14 @@ from documents.tests.factories import DocumentTypeFactory
 from documents.tests.factories import StoragePathFactory
 from documents.tests.factories import TagFactory
 from documents.tests.factories import UserFactory
+from paperless_ai.taxonomy import SimilarDocument
 from paperless_ai.taxonomy import TaxonomyCandidates
 from paperless_ai.taxonomy import build_taxonomy_candidates
 from paperless_ai.taxonomy import format_taxonomy_for_prompt
 
 
-def make_node(document_id: int, score: float) -> SimpleNamespace:
+def make_similar(document_id: int, weight: float) -> SimilarDocument:
-    """A stand-in for NodeWithScore: only ``.metadata``/``.score`` are read."""
+    return SimilarDocument(document_id=document_id, weight=weight)
-    return SimpleNamespace(metadata={"document_id": str(document_id)}, score=score)
 
 
 @pytest.mark.django_db
@@ -53,9 +52,9 @@ class TestBuildTaxonomyCandidates:
         doc_a.tags.add(tag)
         doc_b = DocumentFactory.create()
         doc_b.tags.add(tag)
-        nodes = [make_node(doc_a.pk, 0.9), make_node(doc_b.pk, 0.4)]
+        similar_documents = [make_similar(doc_a.pk, 0.9), make_similar(doc_b.pk, 0.4)]
 
-        result = build_taxonomy_candidates(nodes, user=None)
+        result = build_taxonomy_candidates(similar_documents, user=None)
 
         assert len(result["tags"]) == 1
         assert result["tags"][0]["id"] == tag.pk
@@ -80,9 +79,9 @@ class TestBuildTaxonomyCandidates:
         document.tags.add(tag)
         tag.name = "New Name"
         tag.save()
-        nodes = [make_node(document.pk, 0.5)]
+        similar_documents = [make_similar(document.pk, 0.5)]
 
-        result = build_taxonomy_candidates(nodes, user=None)
+        result = build_taxonomy_candidates(similar_documents, user=None)
 
         assert result["tags"][0]["name"] == "New Name"
 
@@ -102,9 +101,9 @@ class TestBuildTaxonomyCandidates:
         document = DocumentFactory.create()
         document.tags.add(tag)
         tag.delete()
-        nodes = [make_node(document.pk, 0.5)]
+        similar_documents = [make_similar(document.pk, 0.5)]
 
-        result = build_taxonomy_candidates(nodes, user=None)
+        result = build_taxonomy_candidates(similar_documents, user=None)
 
         assert result["tags"] == []
 
@@ -123,9 +122,12 @@ class TestBuildTaxonomyCandidates:
         strong_doc.tags.add(strong_tag)
         weak_doc = DocumentFactory.create()
         weak_doc.tags.add(weak_tag)
-        nodes = [make_node(strong_doc.pk, 0.9), make_node(weak_doc.pk, 0.1)]
+        similar_documents = [
+            make_similar(strong_doc.pk, 0.9),
+            make_similar(weak_doc.pk, 0.1),
+        ]
 
-        result = build_taxonomy_candidates(nodes, user=None)
+        result = build_taxonomy_candidates(similar_documents, user=None)
 
         assert [c["name"] for c in result["tags"]] == ["Strong", "Weak"]
 
@@ -141,9 +143,9 @@ class TestBuildTaxonomyCandidates:
         document = DocumentFactory.create()
         for i in range(15):
             document.tags.add(TagFactory.create(name=f"Tag{i}"))
-        nodes = [make_node(document.pk, 0.5)]
+        similar_documents = [make_similar(document.pk, 0.5)]
 
-        result = build_taxonomy_candidates(nodes, user=None)
+        result = build_taxonomy_candidates(similar_documents, user=None)
 
         assert len(result["tags"]) == 10
 
@@ -157,12 +159,12 @@ class TestBuildTaxonomyCandidates:
             - Only 5 correspondents are returned
         """
         correspondents = CorrespondentFactory.create_batch(7)
-        nodes = [
+        similar_documents = [
-            make_node(DocumentFactory.create(correspondent=c).pk, 0.5)
+            make_similar(DocumentFactory.create(correspondent=c).pk, 0.5)
             for c in correspondents
         ]
 
-        result = build_taxonomy_candidates(nodes, user=None)
+        result = build_taxonomy_candidates(similar_documents, user=None)
 
         assert len(result["correspondents"]) == 5
 
@@ -177,9 +179,9 @@ class TestBuildTaxonomyCandidates:
         """
         document_type = DocumentTypeFactory.create(name="Invoice")
         document = DocumentFactory.create(document_type=document_type)
-        nodes = [make_node(document.pk, 0.5)]
+        similar_documents = [make_similar(document.pk, 0.5)]
 
-        result = build_taxonomy_candidates(nodes, user=None)
+        result = build_taxonomy_candidates(similar_documents, user=None)
 
         assert len(result["document_types"]) == 1
         assert result["document_types"][0]["id"] == document_type.pk
@@ -195,12 +197,12 @@ class TestBuildTaxonomyCandidates:
             - Only 5 document_types are returned
         """
         document_types = DocumentTypeFactory.create_batch(7)
-        nodes = [
+        similar_documents = [
-            make_node(DocumentFactory.create(document_type=dt).pk, 0.5)
+            make_similar(DocumentFactory.create(document_type=dt).pk, 0.5)
             for dt in document_types
         ]
 
-        result = build_taxonomy_candidates(nodes, user=None)
+        result = build_taxonomy_candidates(similar_documents, user=None)
 
         assert len(result["document_types"]) == 5
 
@@ -215,9 +217,9 @@ class TestBuildTaxonomyCandidates:
         """
         storage_path = StoragePathFactory.create(name="Invoices")
         document = DocumentFactory.create(storage_path=storage_path)
-        nodes = [make_node(document.pk, 0.5)]
+        similar_documents = [make_similar(document.pk, 0.5)]
 
-        result = build_taxonomy_candidates(nodes, user=None)
+        result = build_taxonomy_candidates(similar_documents, user=None)
 
         assert len(result["storage_paths"]) == 1
         assert result["storage_paths"][0]["id"] == storage_path.pk
@@ -233,12 +235,12 @@ class TestBuildTaxonomyCandidates:
             - Only 5 storage_paths are returned
         """
         storage_paths = StoragePathFactory.create_batch(7)
-        nodes = [
+        similar_documents = [
-            make_node(DocumentFactory.create(storage_path=sp).pk, 0.5)
+            make_similar(DocumentFactory.create(storage_path=sp).pk, 0.5)
             for sp in storage_paths
         ]
 
-        result = build_taxonomy_candidates(nodes, user=None)
+        result = build_taxonomy_candidates(similar_documents, user=None)
 
         assert len(result["storage_paths"]) == 5
 
@@ -258,14 +260,14 @@ class TestBuildTaxonomyCandidates:
         tag = TagFactory.create(name="Restricted")
         document = DocumentFactory.create()
         document.tags.add(tag)
-        nodes = [make_node(document.pk, 0.5)]
+        similar_documents = [make_similar(document.pk, 0.5)]
         user = UserFactory.create()
         mocker.patch(
             "documents.permissions.permitted_object_ids",
             return_value=[],  # user cannot see this tag
         )
 
-        result = build_taxonomy_candidates(nodes, user=user)
+        result = build_taxonomy_candidates(similar_documents, user=user)
 
         assert result["tags"] == []
 
@@ -295,10 +297,10 @@ class TestBuildTaxonomyCandidates:
         tag.save()
         document = DocumentFactory.create()
         document.tags.add(tag)
-        nodes = [make_node(document.pk, 0.5)]
+        similar_documents = [make_similar(document.pk, 0.5)]
         spy = mocker.patch("documents.permissions.permitted_object_ids")
 
-        result = build_taxonomy_candidates(nodes, user=None)
+        result = build_taxonomy_candidates(similar_documents, user=None)
 
         assert result["tags"][0]["name"] == "Owned"
         spy.assert_not_called()