sort similar documents by weight, use unrestricted-user check consistently (existed before, but good to be consistent)

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

src-ui/src/app/components/document-list/filter-editor/filter-editor.component.html

@@ -10,7 +10,7 @@
             }
           </div>
         </div>
-        @if (textFilterTarget === 'asn' || textFilterTarget === 'duplicates') {
+        @if (textFilterTarget === 'asn') {
           <select class="form-select flex-grow-0 w-auto" [(ngModel)]="textFilterModifier" (change)="textFilterModifierChange()">
             @for (m of textFilterModifiers; track m) {
               <option ngbDropdownItem [value]="m.id">{{m.label}}</option>
@@ -23,7 +23,7 @@
           </button>
         }
         <input #textFilterInput class="form-control form-control-sm" type="text"
-        [disabled]="textFilterInputDisabled"
+        [disabled]="textFilterModifierIsNull"
         [(ngModel)]="textFilter"
         (keydown)="textFilterKeydown($event)"
         [ngbTypeahead]="searchAutoComplete"

src-ui/src/app/components/document-list/filter-editor/filter-editor.component.spec.ts

@@ -53,7 +53,6 @@ import {
   FILTER_HAS_CUSTOM_FIELDS_ALL,
   FILTER_HAS_CUSTOM_FIELDS_ANY,
   FILTER_HAS_DOCUMENT_TYPE_ANY,
-  FILTER_HAS_DUPLICATES,
   FILTER_HAS_STORAGE_PATH_ANY,
   FILTER_HAS_TAGS_ALL,
   FILTER_HAS_TAGS_ANY,
@@ -428,38 +427,6 @@ describe('FilterEditorComponent', () => {
     expect(component.textFilterTarget).toEqual('mime-type') // TEXT_FILTER_TARGET_MIME_TYPE
   })
 
-  it('should ingest filter rules for documents with duplicates', () => {
-    component.filterRules = [
-      {
-        rule_type: FILTER_HAS_DUPLICATES,
-        value: 'true',
-      },
-    ]
-    fixture.detectChanges()
-
-    expect(component.textFilterTarget).toEqual('duplicates')
-    expect(component.textFilterModifier).toEqual('has-duplicates')
-    expect(component.textFilterInputDisabled).toBeTruthy()
-  })
-
-  it('should ingest filter rules for documents without duplicates', () => {
-    component.filterRules = [
-      {
-        rule_type: FILTER_HAS_DUPLICATES,
-        value: 'false',
-      },
-    ]
-
-    expect(component.textFilterTarget).toEqual('duplicates')
-    expect(component.textFilterModifier).toEqual('does-not-have-duplicates')
-    expect(component.filterRules).toEqual([
-      {
-        rule_type: FILTER_HAS_DUPLICATES,
-        value: 'false',
-      },
-    ])
-  })
-
   it('should ingest text filter rules for fulltext query', () => {
     expect(component.textFilter).toEqual(null)
     component.filterRules = [
@@ -1423,33 +1390,6 @@ describe('FilterEditorComponent', () => {
     ])
   })
 
-  it('should convert duplicate target input to the correct filter rule', () => {
-    const textFieldTargetDropdown = fixture.debugElement.queryAll(
-      By.directive(NgbDropdownItem)
-    )[5]
-    textFieldTargetDropdown.triggerEventHandler('click')
-    fixture.detectChanges()
-
-    expect(component.textFilterTarget).toEqual('duplicates')
-    expect(component.filterRules).toEqual([
-      {
-        rule_type: FILTER_HAS_DUPLICATES,
-        value: 'true',
-      },
-    ])
-
-    const textFieldModifierSelect = fixture.debugElement.query(By.css('select'))
-    textFieldModifierSelect.nativeElement.value = 'does-not-have-duplicates'
-    textFieldModifierSelect.nativeElement.dispatchEvent(new Event('change'))
-    fixture.detectChanges()
-    expect(component.filterRules).toEqual([
-      {
-        rule_type: FILTER_HAS_DUPLICATES,
-        value: 'false',
-      },
-    ])
-  })
-
   it('should convert user input to correct filter rules on full text query', () => {
     component.textFilterInput.nativeElement.value = 'foo'
     component.textFilterInput.nativeElement.dispatchEvent(new Event('input'))
@@ -2238,22 +2178,6 @@ describe('FilterEditorComponent', () => {
     ]
     expect(component.generateFilterName()).toEqual('Without any tag')
 
-    component.filterRules = [
-      {
-        rule_type: FILTER_HAS_DUPLICATES,
-        value: 'true',
-      },
-    ]
-    expect(component.generateFilterName()).toEqual('With duplicates')
-
-    component.filterRules = [
-      {
-        rule_type: FILTER_HAS_DUPLICATES,
-        value: 'false',
-      },
-    ]
-    expect(component.generateFilterName()).toEqual('Without duplicates')
-
     component.filterRules = [
       {
         rule_type: FILTER_CUSTOM_FIELDS_QUERY,

src-ui/src/app/components/document-list/filter-editor/filter-editor.component.ts

@@ -65,7 +65,6 @@ import {
   FILTER_HAS_CUSTOM_FIELDS_ALL,
   FILTER_HAS_CUSTOM_FIELDS_ANY,
   FILTER_HAS_DOCUMENT_TYPE_ANY,
-  FILTER_HAS_DUPLICATES,
   FILTER_HAS_STORAGE_PATH_ANY,
   FILTER_HAS_TAGS_ALL,
   FILTER_HAS_TAGS_ANY,
@@ -130,15 +129,12 @@ const TEXT_FILTER_TARGET_FULLTEXT_QUERY = 'fulltext-query'
 const TEXT_FILTER_TARGET_FULLTEXT_MORELIKE = 'fulltext-morelike'
 const TEXT_FILTER_TARGET_CUSTOM_FIELDS = 'custom-fields'
 const TEXT_FILTER_TARGET_MIME_TYPE = 'mime-type'
-const TEXT_FILTER_TARGET_DUPLICATES = 'duplicates'
 
 const TEXT_FILTER_MODIFIER_EQUALS = 'equals'
 const TEXT_FILTER_MODIFIER_NULL = 'is null'
 const TEXT_FILTER_MODIFIER_NOTNULL = 'not null'
 const TEXT_FILTER_MODIFIER_GT = 'greater'
 const TEXT_FILTER_MODIFIER_LT = 'less'
-const TEXT_FILTER_MODIFIER_HAS_DUPLICATES = 'has-duplicates'
-const TEXT_FILTER_MODIFIER_DOES_NOT_HAVE_DUPLICATES = 'does-not-have-duplicates'
 
 const RELATIVE_DATE_QUERY_REGEXP_CREATED = /created:[\["]([^\]]+)[\]"]/g
 const RELATIVE_DATE_QUERY_REGEXP_ADDED = /added:[\["]([^\]]+)[\]"]/g
@@ -209,7 +205,6 @@ const DEFAULT_TEXT_FILTER_TARGET_OPTIONS = [
     id: TEXT_FILTER_TARGET_FULLTEXT_QUERY,
     name: $localize`Advanced search`,
   },
-  { id: TEXT_FILTER_TARGET_DUPLICATES, name: $localize`Duplicates` },
 ]
 
 const DEPRECATED_CUSTOM_FIELDS_TEXT_FILTER_TARGET_OPTION = {
@@ -246,17 +241,6 @@ const DEFAULT_TEXT_FILTER_MODIFIER_OPTIONS = [
   },
 ]
 
-const DUPLICATES_FILTER_MODIFIER_OPTIONS = [
-  {
-    id: TEXT_FILTER_MODIFIER_HAS_DUPLICATES,
-    label: $localize`exist`,
-  },
-  {
-    id: TEXT_FILTER_MODIFIER_DOES_NOT_HAVE_DUPLICATES,
-    label: $localize`do not exist`,
-  },
-]
-
 @Component({
   selector: 'pngx-filter-editor',
   templateUrl: './filter-editor.component.html',
@@ -336,12 +320,6 @@ export class FilterEditorComponent
           if (rule.value == 'false') {
             return $localize`Without any tag`
           }
-          break
-
-        case FILTER_HAS_DUPLICATES:
-          return rule.value == 'false'
-            ? $localize`Without duplicates`
-            : $localize`With duplicates`
 
         case FILTER_CUSTOM_FIELDS_QUERY:
           return $localize`Custom fields query`
@@ -412,9 +390,7 @@ export class FilterEditorComponent
   public textFilterModifier: string
 
   get textFilterModifiers() {
-    return this.textFilterTarget === TEXT_FILTER_TARGET_DUPLICATES
-      ? DUPLICATES_FILTER_MODIFIER_OPTIONS
-      : DEFAULT_TEXT_FILTER_MODIFIER_OPTIONS
+    return DEFAULT_TEXT_FILTER_MODIFIER_OPTIONS
   }
 
   get textFilterModifierIsNull(): boolean {
@@ -423,13 +399,6 @@ export class FilterEditorComponent
     )
   }
 
-  get textFilterInputDisabled(): boolean {
-    return (
-      this.textFilterModifierIsNull ||
-      this.textFilterTarget === TEXT_FILTER_TARGET_DUPLICATES
-    )
-  }
-
   tagSelectionModel = new FilterableDropdownSelectionModel(true)
   correspondentSelectionModel = new FilterableDropdownSelectionModel()
   documentTypeSelectionModel = new FilterableDropdownSelectionModel()
@@ -475,7 +444,6 @@ export class FilterEditorComponent
     this.customFieldQueriesModel.clear(false)
     this._textFilter = null
     this._moreLikeId = null
-    this.textFilterTarget = TEXT_FILTER_TARGET_TITLE_CONTENT
     this.dateAddedTo = null
     this.dateAddedFrom = null
     this.dateCreatedTo = null
@@ -509,13 +477,6 @@ export class FilterEditorComponent
           this.textFilterTarget = TEXT_FILTER_TARGET_MIME_TYPE
           this._textFilter = rule.value
           break
-        case FILTER_HAS_DUPLICATES:
-          this.textFilterTarget = TEXT_FILTER_TARGET_DUPLICATES
-          this.textFilterModifier =
-            rule.value == 'false' || rule.value == '0'
-              ? TEXT_FILTER_MODIFIER_DOES_NOT_HAVE_DUPLICATES
-              : TEXT_FILTER_MODIFIER_HAS_DUPLICATES
-          break
         case FILTER_FULLTEXT_QUERY:
           let allQueryArgs = rule.value.split(',')
           let textQueryArgs = []
@@ -839,14 +800,6 @@ export class FilterEditorComponent
         value: this._textFilter.trim(),
       })
     }
-    if (this.textFilterTarget == TEXT_FILTER_TARGET_DUPLICATES) {
-      filterRules.push({
-        rule_type: FILTER_HAS_DUPLICATES,
-        value: (
-          this.textFilterModifier == TEXT_FILTER_MODIFIER_HAS_DUPLICATES
-        ).toString(),
-      })
-    }
     if (this._textFilter && this.textFilterTarget == TEXT_FILTER_TARGET_TITLE) {
       filterRules.push({
         rule_type: FILTER_SIMPLE_TITLE,
@@ -1210,7 +1163,7 @@ export class FilterEditorComponent
   }
 
   get textFilter() {
-    return this.textFilterInputDisabled ? '' : this._textFilter
+    return this.textFilterModifierIsNull ? '' : this._textFilter
   }
 
   set textFilter(value) {
@@ -1410,24 +1363,12 @@ export class FilterEditorComponent
       this._textFilter = ''
     }
     this.textFilterTarget = target
-    if (target == TEXT_FILTER_TARGET_DUPLICATES) {
-      this._textFilter = ''
-      this.textFilterModifier = TEXT_FILTER_MODIFIER_HAS_DUPLICATES
-    } else if (
-      [
-        TEXT_FILTER_MODIFIER_HAS_DUPLICATES,
-        TEXT_FILTER_MODIFIER_DOES_NOT_HAVE_DUPLICATES,
-      ].includes(this.textFilterModifier)
-    ) {
-      this.textFilterModifier = TEXT_FILTER_MODIFIER_EQUALS
-    }
     this.textFilterInput.nativeElement.focus()
     this.updateRules()
   }
 
   textFilterModifierChange() {
     if (
-      this.textFilterTarget == TEXT_FILTER_TARGET_DUPLICATES ||
       this.textFilterModifierIsNull ||
       ([
         TEXT_FILTER_MODIFIER_EQUALS,

src-ui/src/app/data/filter-rule-type.ts

@@ -49,7 +49,6 @@ export const FILTER_MODIFIED_AFTER = 16
 export const FILTER_TITLE_CONTENT = 19 // Deprecated in favor of Tantivy-backed `text` filtervar. Keep for now for existing saved views
 export const FILTER_SIMPLE_TITLE = 48
 export const FILTER_SIMPLE_TEXT = 49
-export const FILTER_HAS_DUPLICATES = 50
 export const FILTER_FULLTEXT_QUERY = 20
 export const FILTER_FULLTEXT_MORELIKE = 21
 
@@ -383,13 +382,6 @@ export const FILTER_RULE_TYPES: FilterRuleType[] = [
     datatype: 'string',
     multi: false,
   },
-  {
-    id: FILTER_HAS_DUPLICATES,
-    filtervar: 'has_duplicates',
-    datatype: 'boolean',
-    multi: false,
-    default: true,
-  },
 ]
 
 export interface FilterRuleType {

src-ui/src/app/utils/query-params.spec.ts

@@ -7,7 +7,6 @@ import {
   FILTER_HAS_ANY_TAG,
   FILTER_HAS_CUSTOM_FIELDS_ALL,
   FILTER_HAS_CUSTOM_FIELDS_ANY,
-  FILTER_HAS_DUPLICATES,
   FILTER_HAS_TAGS_ALL,
   FILTER_SIMPLE_TEXT,
   FILTER_SIMPLE_TITLE,
@@ -133,16 +132,6 @@ describe('QueryParams Utils', () => {
       is_tagged: 0,
     })
 
-    params = queryParamsFromFilterRules([
-      {
-        rule_type: FILTER_HAS_DUPLICATES,
-        value: 'false',
-      },
-    ])
-    expect(params).toEqual({
-      has_duplicates: 0,
-    })
-
     params = queryParamsFromFilterRules([
       {
         rule_type: FILTER_TITLE_CONTENT,
@@ -258,18 +247,6 @@ describe('QueryParams Utils', () => {
       },
     ])
 
-    rules = filterRulesFromQueryParams(
-      convertToParamMap({
-        has_duplicates: 'true',
-      })
-    )
-    expect(rules).toEqual([
-      {
-        rule_type: FILTER_HAS_DUPLICATES,
-        value: 'true',
-      },
-    ])
-
     rules = filterRulesFromQueryParams(
       convertToParamMap({
         correspondent__isnull: '1',

src/documents/filters.py

@@ -25,7 +25,6 @@ from django.db.models import Sum
 from django.db.models import Value
 from django.db.models import When
 from django.db.models.functions import Cast
-from django.db.models.functions import NullIf
 from django.utils.translation import gettext_lazy as _
 from django_filters import DateFilter
 from django_filters.rest_framework import BooleanFilter
@@ -51,7 +50,6 @@ from documents.models import ShareLink
 from documents.models import ShareLinkBundle
 from documents.models import StoragePath
 from documents.models import Tag
-from documents.permissions import permitted_document_ids
 from documents.permissions import permitted_object_ids
 
 if TYPE_CHECKING:
@@ -795,12 +793,6 @@ class CustomFieldQueryFilter(Filter):
 
 
 class DocumentFilterSet(FilterSet):
-    has_duplicates = BooleanFilter(method="filter_has_duplicates")
-
-    def __init__(self, *args: Any, user: Any = None, **kwargs: Any) -> None:
-        super().__init__(*args, **kwargs)
-        self._user = user
-
     is_tagged = BooleanFilter(
         label="Is tagged",
         field_name="tags",
@@ -860,38 +852,6 @@ class DocumentFilterSet(FilterSet):
 
     mime_type = MimeTypeFilter()
 
-    def filter_has_duplicates(self, queryset, name, value):
-        if value is None:
-            return queryset
-
-        user = (
-            self._user
-            if self._user is not None
-            else getattr(self.request, "user", None)
-        )
-        queryset = queryset.alias(
-            nonempty_archive_checksum=NullIf("archive_checksum", Value("")),
-        )
-
-        visible_root_documents = Document.global_objects.filter(
-            root_document__isnull=True,
-            pk__in=permitted_document_ids(
-                user,
-                include_deleted=True,
-            ),
-        ).exclude(pk=OuterRef("pk"))
-        # see serialisers._get_viewable_duplicates().
-        matching_duplicates = visible_root_documents.filter(
-            Q(checksum=OuterRef("checksum"))
-            | Q(checksum=OuterRef("nonempty_archive_checksum"))
-            | Q(archive_checksum=OuterRef("checksum"))
-            | Q(archive_checksum=OuterRef("nonempty_archive_checksum")),
-        )
-
-        return queryset.alias(
-            has_visible_duplicates=Exists(matching_duplicates),
-        ).filter(has_visible_duplicates=value)
-
     # Backwards compatibility
     created__date__gt = DateFilter(field_name="created", lookup_expr="gt")
     created__date__gte = DateFilter(field_name="created", lookup_expr="gte")

src/documents/migrations/0026_alter_document_archive_checksum_and_more.py

@@ -1,86 +0,0 @@
-# Generated by Django 5.2.16 on 2026-09-05 16:29
-
-from django.db import migrations
-from django.db import models
-
-
-class Migration(migrations.Migration):
-    dependencies = [
-        ("documents", "0025_workflowaction_apply_ai_suggestions"),
-    ]
-
-    operations = [
-        migrations.AlterField(
-            model_name="document",
-            name="archive_checksum",
-            field=models.CharField(
-                blank=True,
-                db_index=True,
-                editable=False,
-                help_text="The checksum of the archived document.",
-                max_length=64,
-                null=True,
-                verbose_name="archive checksum",
-            ),
-        ),
-        migrations.AlterField(
-            model_name="savedviewfilterrule",
-            name="rule_type",
-            field=models.PositiveSmallIntegerField(
-                choices=[
-                    (0, "title contains"),
-                    (1, "content contains"),
-                    (2, "ASN is"),
-                    (3, "correspondent is"),
-                    (4, "document type is"),
-                    (5, "is in inbox"),
-                    (6, "has tag"),
-                    (7, "has any tag"),
-                    (8, "created before"),
-                    (9, "created after"),
-                    (10, "created year is"),
-                    (11, "created month is"),
-                    (12, "created day is"),
-                    (13, "added before"),
-                    (14, "added after"),
-                    (15, "modified before"),
-                    (16, "modified after"),
-                    (17, "does not have tag"),
-                    (18, "does not have ASN"),
-                    (19, "title or content contains"),
-                    (20, "fulltext query"),
-                    (21, "more like this"),
-                    (22, "has tags in"),
-                    (23, "ASN greater than"),
-                    (24, "ASN less than"),
-                    (25, "storage path is"),
-                    (26, "has correspondent in"),
-                    (27, "does not have correspondent in"),
-                    (28, "has document type in"),
-                    (29, "does not have document type in"),
-                    (30, "has storage path in"),
-                    (31, "does not have storage path in"),
-                    (32, "owner is"),
-                    (33, "has owner in"),
-                    (34, "does not have owner"),
-                    (35, "does not have owner in"),
-                    (36, "has custom field value"),
-                    (37, "is shared by me"),
-                    (38, "has custom fields"),
-                    (39, "has custom field in"),
-                    (40, "does not have custom field in"),
-                    (41, "does not have custom field"),
-                    (42, "custom fields query"),
-                    (43, "created to"),
-                    (44, "created from"),
-                    (45, "added to"),
-                    (46, "added from"),
-                    (47, "mime type is"),
-                    (48, "simple title search"),
-                    (49, "simple text search"),
-                    (50, "has duplicates"),
-                ],
-                verbose_name="rule type",
-            ),
-        ),
-    ]

src/documents/models.py

@@ -227,7 +227,6 @@ class Document(SoftDeleteModel, ModelWithOwner):  # type: ignore[django-manager-
         editable=False,
         blank=True,
         null=True,
-        db_index=True,
         help_text=_("The checksum of the archived document."),
     )
 
@@ -707,7 +706,6 @@ class SavedViewFilterRule(models.Model):
         (47, _("mime type is")),
         (48, _("simple title search")),
         (49, _("simple text search")),
-        (50, _("has duplicates")),
     ]
 
     saved_view = models.ForeignKey(

src/documents/tests/test_api_bulk_edit.py

@@ -717,44 +717,6 @@ class TestBulkEditAPI(DirectoriesMixin, APITestCase):
         self.assertEqual(args[0], [self.doc2.id])
         self.assertEqual(kwargs["storage_path"], self.sp1.id)
 
-    @mock.patch("documents.serialisers.bulk_edit.set_storage_path")
-    def test_api_bulk_edit_with_all_true_resolves_owned_duplicates(self, m) -> None:
-        self.setup_mock(m, "set_storage_path")
-        user = User.objects.create_user(username="duplicate-owner")
-        user.user_permissions.add(
-            Permission.objects.get(codename="change_document"),
-        )
-        first_duplicate = Document.objects.create(
-            checksum="owned-duplicate",
-            title="First duplicate",
-            owner=user,
-        )
-        second_duplicate = Document.objects.create(
-            checksum="owned-duplicate",
-            title="Second duplicate",
-            owner=user,
-        )
-        self.client.force_authenticate(user=user)
-
-        response = self.client.post(
-            "/api/documents/bulk_edit/",
-            json.dumps(
-                {
-                    "all": True,
-                    "filters": {"has_duplicates": True},
-                    "method": "set_storage_path",
-                    "parameters": {"storage_path": self.sp1.id},
-                },
-            ),
-            content_type="application/json",
-        )
-
-        self.assertEqual(response.status_code, status.HTTP_200_OK)
-        m.assert_called_once()
-        args, kwargs = m.call_args
-        self.assertCountEqual(args[0], [first_duplicate.id, second_duplicate.id])
-        self.assertEqual(kwargs["storage_path"], self.sp1.id)
-
     @mock.patch("documents.search.get_backend")
     @mock.patch("documents.serialisers.bulk_edit.set_storage_path")
     def test_api_bulk_edit_with_all_true_resolves_documents_from_search_filters(

src/documents/tests/test_api_documents.py

@@ -981,128 +981,6 @@ class TestDocumentApi(DirectoriesMixin, ConsumeTaskMixin, APITestCase):
         self.assertEqual(len(results), 1)
         self.assertEqual(results[0]["id"], doc.id)
 
-    def test_has_duplicates_filter(self) -> None:
-        original_match = Document.objects.create(
-            title="original match",
-            checksum="same-original",
-        )
-        second_original_match = Document.objects.create(
-            title="second original match",
-            checksum="same-original",
-        )
-        archive_match = Document.objects.create(
-            title="archive match",
-            checksum="archive-source",
-            archive_checksum="same-archive",
-        )
-        original_to_archive_match = Document.objects.create(
-            title="original to archive match",
-            checksum="same-archive",
-        )
-        first_archive_match = Document.objects.create(
-            title="first archive match",
-            checksum="first-archive-source",
-            archive_checksum="same-archive-only",
-        )
-        second_archive_match = Document.objects.create(
-            title="second archive match",
-            checksum="second-archive-source",
-            archive_checksum="same-archive-only",
-        )
-        first_empty_archive = Document.objects.create(
-            title="first empty archive",
-            checksum="first-empty-archive",
-            archive_checksum="",
-        )
-        second_empty_archive = Document.objects.create(
-            title="second empty archive",
-            checksum="second-empty-archive",
-            archive_checksum="",
-        )
-        unique = Document.objects.create(title="unique", checksum="unique")
-        version_root = Document.objects.create(
-            title="version root",
-            checksum="version-root",
-        )
-        Document.objects.create(
-            title="version",
-            checksum=unique.checksum,
-            root_document=version_root,
-            version_index=1,
-        )
-        trash_match = Document.objects.create(
-            title="trash match",
-            checksum="trash-match",
-        )
-        trashed_duplicate = Document.objects.create(
-            title="trashed duplicate",
-            checksum="trash-match",
-        )
-        trashed_duplicate.delete()
-
-        response = self.client.get("/api/documents/?has_duplicates=true")
-
-        self.assertEqual(response.status_code, status.HTTP_200_OK)
-        self.assertCountEqual(
-            [document["id"] for document in response.data["results"]],
-            [
-                original_match.id,
-                second_original_match.id,
-                archive_match.id,
-                original_to_archive_match.id,
-                first_archive_match.id,
-                second_archive_match.id,
-                trash_match.id,
-            ],
-        )
-
-        response = self.client.get("/api/documents/?has_duplicates=false")
-
-        self.assertEqual(response.status_code, status.HTTP_200_OK)
-        self.assertCountEqual(
-            [document["id"] for document in response.data["results"]],
-            [
-                unique.id,
-                version_root.id,
-                first_empty_archive.id,
-                second_empty_archive.id,
-            ],
-        )
-
-        response = self.client.get(f"/api/documents/{first_empty_archive.id}/")
-        self.assertEqual(response.data["duplicate_documents"], [])
-
-    def test_has_duplicates_filter_respects_document_permissions(self) -> None:
-        owner = User.objects.create_user(username="duplicate-owner")
-        requester = User.objects.create_user(username="duplicate-requester")
-        requester.user_permissions.add(
-            Permission.objects.get(codename="view_document"),
-        )
-        visible_document = Document.objects.create(
-            title="visible document",
-            checksum="permission-match",
-            owner=requester,
-        )
-        hidden_duplicate = Document.objects.create(
-            title="hidden duplicate",
-            checksum="permission-match",
-            owner=owner,
-        )
-        self.client.force_authenticate(user=requester)
-
-        response = self.client.get("/api/documents/?has_duplicates=true")
-        self.assertNotIn(
-            visible_document.id,
-            [document["id"] for document in response.data["results"]],
-        )
-
-        assign_perm("view_document", requester, hidden_duplicate)
-        response = self.client.get("/api/documents/?has_duplicates=true")
-        self.assertIn(
-            visible_document.id,
-            [document["id"] for document in response.data["results"]],
-        )
-
     def test_custom_fields_icontains_filter_no_duplicates(self) -> None:
         """
         GIVEN:

src/documents/views.py

@@ -2815,7 +2815,6 @@ class DocumentSelectionMixin:
         filtered_documents = DocumentFilterSet(
             data=orm_filters,
             queryset=permitted_documents,
-            user=user,
         ).qs.distinct()
         # tantivy-filtered docs (if search params provided)
         search_filtered_ids = self._get_search_document_ids(

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.
-    On any retrieval failure, degrades to empty candidates/context rather than
-    propagating the exception - a vector-store outage should not block
-    classification, only its RAG-assisted enrichment.
+    """One retrieval feeds both taxonomy candidates and RAG text context. Uses
+    vector similarity when an embedding backend is configured, otherwise
+    falls back to Tantivy full-text "More Like This" similarity - see
+    _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
-        # (like no user at all) can see every document, so skip materializing
-        # every visible pk into a Python list and passing it through as an IN
-        # filter: for a large library that is a wasted quadratic scan in the
-        # vector store at best, and past ~32,763 documents a hard
-        # sqlite3.OperationalError (SQLite's bound-parameter limit) at worst.
-        # get_objects_for_user_owner_aware() would return every Document for a
-        # superuser anyway (guardian's own with_superuser shortcut), 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
-            else list(
-                get_objects_for_user_owner_aware(
-                    user,
-                    "view_document",
-                    Document,
-                ).values_list("pk", flat=True),
+        if ai_config.llm_embedding_backend:
+            # None means "no restriction" to retrieve_similar_nodes. An
+            # unrestricted user (no user at all, or an active superuser -- see
+            # user_is_unrestricted) can see every document, so skip
+            # materializing every visible pk into a Python list and passing it
+            # through as an IN filter: for a large library that is a wasted
+            # quadratic scan in the vector store at best, and past ~32,763
+            # documents a hard sqlite3.OperationalError (SQLite's
+            # bound-parameter limit) at worst.
+            # 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_unrestricted(user)
+                else list(permitted_object_ids(user, Document, "view_document"))
+            )
+            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,
             )
-        )
-        nodes = retrieve_similar_nodes(
-            document,
-            top_k=TAXONOMY_CANDIDATE_TOP_K,
-            document_ids=visible_document_ids,
-        )
 
-        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_document_ids = list(dict.fromkeys(_node_document_ids(nodes)))
+        # similar_documents is already ordered by descending weight; don't lose it.
+        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 "
-            "without taxonomy candidates or similar-document context.",
+            "Failed to retrieve similar-document context for document %s; "
+            "continuing without taxonomy candidates or similar-document context.",
             document.pk,
         )
         return empty_taxonomy_candidates(), ""
@@ -241,17 +295,13 @@ 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,
-            ai_config,
-            candidates=candidates,
-            context=context,
-        )
-    else:
-        candidates = empty_taxonomy_candidates()
-        prompt = build_prompt_without_rag(document, ai_config, candidates=candidates)
+    candidates, context = get_taxonomy_context(document, user)
+    prompt = build_prompt_with_rag(
+        document,
+        ai_config,
+        candidates=candidates,
+        context=context,
+    )
 
     client = AIClient()
     # Hand the pooled DB connection back while the (slow) LLM query runs so it

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]:
-    """document_id -> that node's similarity score, summed if a document_id
-    appears more than once across the retrieved nodes (e.g. multiple chunks
-    of the same source document)."""
+def _node_document_weights(nodes: list["NodeWithScore"]) -> list[SimilarDocument]:
+    """Sum each node's similarity score into its document_id (a document can
+    appear via multiple chunks/nodes) and return one SimilarDocument per
+    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
-    *current* tags/type/correspondent/storage_path via the ORM (never the
-    possibly-stale names cached in vector-index node metadata), weight each
-    distinct taxonomy object by aggregate neighbour similarity, permission-filter
+    """Resolve each similar document's id to a live Document, read its
+    *current* tags/type/correspondent/storage_path via the ORM (never any
+    possibly-stale names an adapter's source might have cached), weight each
+    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.
     """
-
-    document_weights = _node_document_weights(nodes)
-    if not document_weights:
+    if not similar_documents:
         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
@@ -8,16 +9,20 @@ import pytest_mock
 from django.test import override_settings
 
 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 +225,10 @@ 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.AIConfig")
+@patch("paperless_ai.ai_classifier.build_prompt_with_rag")
 @override_settings(LLM_BACKEND="ollama", LLM_MODEL="some_model")
-def test_use_without_rag_if_not_configured(
-    mock_ai_config,
-    mock_build_prompt_without_rag,
+def test_use_rag_prompt_even_without_embedding_backend(
+    mock_build_prompt_with_rag,
     mock_run_llm_query,
     mock_document,
 ):
@@ -235,13 +238,13 @@ 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_without_rag.return_value = "Prompt without RAG"
+    mock_build_prompt_with_rag.return_value = "Prompt with 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 +323,7 @@ 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 +358,7 @@ 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 +429,7 @@ 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 +453,67 @@ 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 +526,7 @@ 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 +545,18 @@ class TestGetTaxonomyContextVisibility:
             "paperless_ai.ai_classifier.retrieve_similar_nodes",
             return_value=[],
         )
-        mock_get_objects = mocker.patch(
-            "paperless_ai.ai_classifier.get_objects_for_user_owner_aware",
+        mock_permitted = mocker.patch(
+            "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 +575,17 @@ class TestGetTaxonomyContextVisibility:
             "paperless_ai.ai_classifier.retrieve_similar_nodes",
             return_value=[],
         )
-        mock_get_objects = mocker.patch(
-            "paperless_ai.ai_classifier.get_objects_for_user_owner_aware",
+        mock_permitted = mocker.patch(
+            "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 +596,7 @@ 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 +604,186 @@ class TestGetTaxonomyContextVisibility:
             "paperless_ai.ai_classifier.retrieve_similar_nodes",
             return_value=[],
         )
-        mock_queryset = mocker.MagicMock()
-        mock_queryset.values_list.return_value = [1, 2, 3]
-        mock_get_objects = mocker.patch(
-            "paperless_ai.ai_classifier.get_objects_for_user_owner_aware",
-            return_value=mock_queryset,
+        mock_permitted = mocker.patch(
+            "paperless_ai.ai_classifier.permitted_object_ids",
+            return_value=[1, 2, 3],
         )
         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]
+
+
+@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 +810,7 @@ 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_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:
-    """A stand-in for NodeWithScore: only ``.metadata``/``.score`` are read."""
-    return SimpleNamespace(metadata={"document_id": str(document_id)}, score=score)
+def make_similar(document_id: int, weight: float) -> SimilarDocument:
+    return SimilarDocument(document_id=document_id, weight=weight)
 
 
 @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 = [
-            make_node(DocumentFactory.create(correspondent=c).pk, 0.5)
+        similar_documents = [
+            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 = [
-            make_node(DocumentFactory.create(document_type=dt).pk, 0.5)
+        similar_documents = [
+            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 = [
-            make_node(DocumentFactory.create(storage_path=sp).pk, 0.5)
+        similar_documents = [
+            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()