Use SVG, get rid of all the manual redraw stuff

[skip ci]

docker/rootfs/etc/s6-overlay/s6-rc.d/init-compile-bytecode/run

@@ -1,58 +0,0 @@
-#!/command/with-contenv /usr/bin/bash
-# shellcheck shell=bash
-declare -r log_prefix="[init-compile-bytecode]"
-
-# PYTHONDONTWRITEBYTECODE=1 is set for the whole container. This unit compiles a
-# scoped set of libraries anyway, to speed up startup without bloating image size.
-
-# Handle the people using a read only file system
-if [[ "${S6_READ_ONLY_ROOT}" == "1" ]]; then
-	echo "${log_prefix} S6_READ_ONLY_ROOT=1, skipping (nothing to write bytecode to)"
-	exit 0
-fi
-
-# When running as a non-root user, site-packages is still root-owned and unwritable,
-# so this step would just fail loudly on every container start. Skip it.
-if [[ -n "${USER_IS_NON_ROOT}" ]]; then
-	echo "${log_prefix} USER_IS_NON_ROOT is set, skipping (site-packages is not writable)"
-	exit 0
-fi
-
-declare -r site_packages="$(python3 -c 'import site; print(site.getsitepackages()[0])')"
-
-# Deliberately scoped to packages that paperless.settings/paperless/__init__.py import
-# unconditionally on every manage.py invocation (Django itself, the always-loaded
-# INSTALLED_APPS, and celery). This is NOT "compile everything" - the optional AI stack
-# (torch, llama-index, sentence-transformers, ...) is intentionally excluded since it is
-# lazy-imported and large.
-declare -a scope=(
-	"${PAPERLESS_SRC_DIR}"
-	"${site_packages}/django"
-	"${site_packages}/celery"
-	"${site_packages}/kombu"
-	"${site_packages}/rest_framework"
-	"${site_packages}/django_filters"
-	"${site_packages}/whitenoise"
-	"${site_packages}/corsheaders"
-	"${site_packages}/django_extensions"
-	"${site_packages}/guardian"
-	"${site_packages}/allauth"
-	"${site_packages}/drf_spectacular"
-	"${site_packages}/drf_spectacular_sidecar"
-	"${site_packages}/treenode"
-	"${site_packages}/compression_middleware"
-)
-
-declare -a existing_scope=()
-for path in "${scope[@]}"; do
-	[[ -d "${path}" ]] && existing_scope+=("${path}")
-done
-
-echo "${log_prefix} Compiling bytecode for: ${existing_scope[*]}"
-declare -r start_seconds=${SECONDS}
-
-if ! PYTHONDONTWRITEBYTECODE= python3 -m compileall -q "${existing_scope[@]}"; then
-	echo "${log_prefix} WARNING: compileall reported errors (read-only filesystem or unwritable site-packages?); continuing without a bytecode cache"
-fi
-
-echo "${log_prefix} Done in $((SECONDS - start_seconds))s"

docker/rootfs/etc/s6-overlay/s6-rc.d/init-compile-bytecode/type

@@ -1 +0,0 @@
-oneshot

docker/rootfs/etc/s6-overlay/s6-rc.d/init-compile-bytecode/up

@@ -1 +0,0 @@
-/etc/s6-overlay/s6-rc.d/init-compile-bytecode/run

docs/configuration.md

@@ -1200,15 +1200,6 @@ still perform some basic text pre-processing before matching.
 
     Defaults to true, enabling the feature.
 
-#### [`PAPERLESS_CLASSIFIER_MATCH_THRESHOLD=<float>`](#PAPERLESS_CLASSIFIER_MATCH_THRESHOLD) {#PAPERLESS_CLASSIFIER_MATCH_THRESHOLD}
-
-: Sets the minimum confidence score (0.0-1.0) required for the automatic
-classifier to assign a correspondent, document type, or storage path to a
-document. Predictions below this threshold are discarded and the field is
-left unassigned, preventing low-confidence guesses from being applied.
-
-    Defaults to 0.6.
-
 #### [`PAPERLESS_DATE_PARSER_LANGUAGES=<lang>`](#PAPERLESS_DATE_PARSER_LANGUAGES) {#PAPERLESS_DATE_PARSER_LANGUAGES}
 
 : Specifies which language Paperless should use when parsing dates from documents.

src-ui/src/app/app-routing.module.ts

@@ -13,6 +13,8 @@ import { DocumentDetailComponent } from './components/document-detail/document-d
 import { DocumentListComponent } from './components/document-list/document-list.component'
 import { DocumentAttributesComponent } from './components/manage/document-attributes/document-attributes.component'
 import { MailComponent } from './components/manage/mail/mail.component'
+import { OcrTemplateEditorComponent } from './components/manage/ocr-templates/ocr-template-editor/ocr-template-editor.component'
+import { OcrTemplatesComponent } from './components/manage/ocr-templates/ocr-templates.component'
 import { SavedViewsComponent } from './components/manage/saved-views/saved-views.component'
 import { WorkflowsComponent } from './components/manage/workflows/workflows.component'
 import { NotFoundComponent } from './components/not-found/not-found.component'
@@ -274,6 +276,42 @@ export const routes: Routes = [
           componentName: 'WorkflowsComponent',
         },
       },
+      {
+        path: 'ocr-templates',
+        component: OcrTemplatesComponent,
+        canActivate: [PermissionsGuard],
+        data: {
+          requiredPermission: {
+            action: PermissionAction.View,
+            type: PermissionType.OcrTemplate,
+          },
+          componentName: 'OcrTemplatesComponent',
+        },
+      },
+      {
+        path: 'ocr-templates/new',
+        component: OcrTemplateEditorComponent,
+        canActivate: [PermissionsGuard],
+        data: {
+          requiredPermission: {
+            action: PermissionAction.Add,
+            type: PermissionType.OcrTemplate,
+          },
+          componentName: 'OcrTemplateEditorComponent',
+        },
+      },
+      {
+        path: 'ocr-templates/:id',
+        component: OcrTemplateEditorComponent,
+        canActivate: [PermissionsGuard],
+        data: {
+          requiredPermission: {
+            action: PermissionAction.Change,
+            type: PermissionType.OcrTemplate,
+          },
+          componentName: 'OcrTemplateEditorComponent',
+        },
+      },
       {
         path: 'mail',
         component: MailComponent,

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

@@ -112,22 +112,6 @@
 
                 <pngx-input-check i18n-title title="Use 'slim' sidebar (icons only)" formControlName="slimSidebarEnabled"></pngx-input-check>
 
-                <p class="mb-2 mt-3" i18n>Sidebar items to show:</p>
-                @for (option of sidebarItemOptions; track option.id) {
-                  <div class="form-check">
-                    <input
-                      class="form-check-input"
-                      type="checkbox"
-                      [id]="'sidebar-item-setting-' + option.id"
-                      [checked]="isSidebarItemShown(option.id)"
-                      (change)="toggleSidebarItem(option.id, $event.target.checked)"
-                    />
-                    <label class="form-check-label" [for]="'sidebar-item-setting-' + option.id">
-                      {{ option.label }}
-                    </label>
-                  </div>
-                }
-
               </div>
             </div>
 

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

@@ -24,7 +24,7 @@ import {
   SystemStatus,
   SystemStatusItemStatus,
 } from 'src/app/data/system-status'
-import { HideableSidebarItemID, SETTINGS_KEYS } from 'src/app/data/ui-settings'
+import { SETTINGS_KEYS } from 'src/app/data/ui-settings'
 import { IfOwnerDirective } from 'src/app/directives/if-owner.directive'
 import { IfPermissionsDirective } from 'src/app/directives/if-permissions.directive'
 import { PermissionsGuard } from 'src/app/guards/permissions.guard'
@@ -209,45 +209,6 @@ describe('SettingsComponent', () => {
     fixture.detectChanges()
   }
 
-  it('supports configuring sidebar items and canceling changes', () => {
-    completeSetup()
-
-    component.toggleSidebarItem(HideableSidebarItemID.Workflows, false)
-    fixture.detectChanges()
-
-    expect(component.settingsForm.value.sidebarHiddenItems).toContain(
-      HideableSidebarItemID.Workflows
-    )
-
-    settingsService.updateSidebarItemVisibility(
-      HideableSidebarItemID.Mail,
-      false
-    )
-
-    expect(component.settingsForm.value.sidebarHiddenItems).toContain(
-      HideableSidebarItemID.Mail
-    )
-
-    component.reset()
-
-    expect(component.settingsForm.value.sidebarHiddenItems).not.toContain(
-      HideableSidebarItemID.Workflows
-    )
-    expect(component.settingsForm.value.sidebarHiddenItems).not.toContain(
-      HideableSidebarItemID.Mail
-    )
-  })
-
-  it('enables sidebar item controls on general settings until destroyed', () => {
-    completeSetup()
-
-    expect(settingsService.organizingSidebarItems()).toBe(true)
-
-    component.ngOnDestroy()
-
-    expect(settingsService.organizingSidebarItems()).toBe(false)
-  })
-
   it('should support tabbed settings & change URL, prevent navigation if dirty confirmation rejected', async () => {
     completeSetup()
     const navigateSpy = jest.spyOn(router, 'navigate')
@@ -288,7 +249,6 @@ describe('SettingsComponent', () => {
 
   it('should support save local settings updating appearance settings and calling API, show error', () => {
     completeSetup()
-    component.toggleSidebarItem(HideableSidebarItemID.Workflows, false)
     const toastErrorSpy = jest.spyOn(toastService, 'showError')
     const toastSpy = jest.spyOn(toastService, 'show')
     const storeSpy = jest.spyOn(settingsService, 'storeSettings')
@@ -307,10 +267,7 @@ describe('SettingsComponent', () => {
     expect(toastErrorSpy).toHaveBeenCalled()
     expect(storeSpy).toHaveBeenCalled()
     expect(appearanceSettingsSpy).not.toHaveBeenCalled()
-    expect(setSpy).toHaveBeenCalledTimes(34)
-    expect(setSpy).toHaveBeenCalledWith(SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS, [
-      HideableSidebarItemID.Workflows,
-    ])
+    expect(setSpy).toHaveBeenCalledTimes(33)
 
     // succeed
     storeSpy.mockReturnValueOnce(of(true))

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

@@ -39,12 +39,7 @@ import {
   SystemStatus,
   SystemStatusItemStatus,
 } from 'src/app/data/system-status'
-import {
-  GlobalSearchType,
-  HIDEABLE_SIDEBAR_ITEM_IDS,
-  HideableSidebarItemID,
-  SETTINGS_KEYS,
-} from 'src/app/data/ui-settings'
+import { GlobalSearchType, SETTINGS_KEYS } from 'src/app/data/ui-settings'
 import { User } from 'src/app/data/user'
 import { IfPermissionsDirective } from 'src/app/directives/if-permissions.directive'
 import { CustomDatePipe } from 'src/app/pipes/custom-date.pipe'
@@ -107,14 +102,6 @@ const documentDetailFieldOptions = [
   { id: DocumentDetailFieldID.Tags, label: $localize`Tags` },
 ]
 
-const sidebarItemLabels: Record<HideableSidebarItemID, string> = {
-  [HideableSidebarItemID.Dashboard]: $localize`Dashboard`,
-  [HideableSidebarItemID.SavedViews]: $localize`Saved Views`,
-  [HideableSidebarItemID.Workflows]: $localize`Workflows`,
-  [HideableSidebarItemID.Mail]: $localize`Mail`,
-  [HideableSidebarItemID.Documentation]: $localize`Documentation`,
-}
-
 @Component({
   selector: 'pngx-settings',
   templateUrl: './settings.component.html',
@@ -162,7 +149,6 @@ export class SettingsComponent
     bulkEditApplyOnClose: new FormControl(null),
     documentListItemPerPage: new FormControl(null),
     slimSidebarEnabled: new FormControl(null),
-    sidebarHiddenItems: new FormControl<HideableSidebarItemID[]>([]),
     darkModeUseSystem: new FormControl(null),
     darkModeEnabled: new FormControl(null),
     darkModeInvertThumbs: new FormControl(null),
@@ -200,7 +186,6 @@ export class SettingsComponent
 
   store: BehaviorSubject<any>
   storeSub: Subscription
-  sidebarItemsSub: Subscription
   isDirty$: Observable<boolean>
   isDirty: boolean = false
   unsubscribeNotifier: Subject<any> = new Subject()
@@ -218,10 +203,6 @@ export class SettingsComponent
   public readonly PdfEditorEditMode = PdfEditorEditMode
 
   public readonly documentDetailFieldOptions = documentDetailFieldOptions
-  public readonly sidebarItemOptions = HIDEABLE_SIDEBAR_ITEM_IDS.map((id) => ({
-    id,
-    label: sidebarItemLabels[id],
-  }))
 
   get systemStatusHasErrors(): boolean {
     const status = this.systemStatus()
@@ -249,10 +230,6 @@ export class SettingsComponent
 
   constructor() {
     super()
-    this.sidebarItemsSub =
-      this.settings.sidebarHiddenItemsEditingChanged.subscribe((hiddenItems) =>
-        this.settingsForm.controls.sidebarHiddenItems.setValue(hiddenItems)
-      )
     this.settings.settingsSaved.subscribe(() => {
       if (!this.savePending) this.initialize()
       this.savedViewsService.maybeRefreshDocumentCounts()
@@ -302,21 +279,14 @@ export class SettingsComponent
 
     this.activatedRoute.paramMap.subscribe((paramMap) => {
       const section = paramMap.get('section')
-      let navID = SettingsNavIDs.General
       if (section) {
         const navIDKey: string = Object.keys(SettingsNavIDs).find(
           (navID) => navID.toLowerCase() == section
         )
         if (navIDKey) {
-          navID = SettingsNavIDs[navIDKey]
+          this.activeNavID.set(SettingsNavIDs[navIDKey])
         }
       }
-      this.activeNavID.set(navID)
-      this.settings.sidebarHiddenItemsEditing.set(
-        navID === SettingsNavIDs.General
-          ? [...this.settingsForm.controls.sidebarHiddenItems.value]
-          : null
-      )
     })
   }
 
@@ -340,7 +310,6 @@ export class SettingsComponent
         SETTINGS_KEYS.DOCUMENT_LIST_SIZE
       ),
       slimSidebarEnabled: this.settings.get(SETTINGS_KEYS.SLIM_SIDEBAR),
-      sidebarHiddenItems: this.settings.get(SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS),
       darkModeUseSystem: this.settings.get(SETTINGS_KEYS.DARK_MODE_USE_SYSTEM),
       darkModeEnabled: this.settings.get(SETTINGS_KEYS.DARK_MODE_ENABLED),
       darkModeInvertThumbs: this.settings.get(
@@ -467,12 +436,6 @@ export class SettingsComponent
       this.settingsForm.patchValue(currentFormValue)
     }
 
-    if (this.settings.organizingSidebarItems()) {
-      this.settings.sidebarHiddenItemsEditing.set([
-        ...this.settingsForm.controls.sidebarHiddenItems.value,
-      ])
-    }
-
     if (this.canViewSystemStatus) {
       this.systemStatusService.get().subscribe((status) => {
         this.systemStatus.set(status)
@@ -481,18 +444,8 @@ export class SettingsComponent
   }
 
   ngOnDestroy() {
-    this.settings.sidebarHiddenItemsEditing.set(null)
     if (this.isDirty) this.settings.updateAppearanceSettings() // in case user changed appearance but didn't save
     this.storeSub && this.storeSub.unsubscribe()
-    this.sidebarItemsSub.unsubscribe()
-  }
-
-  isSidebarItemShown(item: HideableSidebarItemID): boolean {
-    return !(this.settingsForm.value.sidebarHiddenItems || []).includes(item)
-  }
-
-  toggleSidebarItem(item: HideableSidebarItemID, checked: boolean): void {
-    this.settings.updateSidebarItemVisibility(item, checked)
   }
 
   public saveSettings() {
@@ -520,10 +473,6 @@ export class SettingsComponent
       SETTINGS_KEYS.SLIM_SIDEBAR,
       this.settingsForm.value.slimSidebarEnabled
     )
-    this.settings.set(
-      SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS,
-      this.settingsForm.value.sidebarHiddenItems
-    )
     this.settings.set(
       SETTINGS_KEYS.DARK_MODE_USE_SYSTEM,
       this.settingsForm.value.darkModeUseSystem
@@ -683,11 +632,6 @@ export class SettingsComponent
 
   reset() {
     this.settingsForm.patchValue(this.store.getValue())
-    if (this.settings.organizingSidebarItems()) {
-      this.settings.sidebarHiddenItemsEditing.set([
-        ...this.settingsForm.controls.sidebarHiddenItems.value,
-      ])
-    }
   }
 
   clearThemeColor() {

src-ui/src/app/components/app-frame/app-frame.component.html

@@ -86,15 +86,12 @@
       }
       <div class="sidebar-sticky pt-3 pb-1 d-flex flex-column justify-space-around">
         <ul class="nav flex-column">
-          <li class="nav-item app-link position-relative" [class.d-none]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Dashboard) && !settingsService.organizingSidebarItems()">
-            <a class="nav-link" [class.opacity-50]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Dashboard)" [class.pe-5]="settingsService.organizingSidebarItems() && !slimSidebarEnabled && !slimSidebarAnimating()" routerLink="dashboard" routerLinkActive="active" (click)="closeMenu()"
+          <li class="nav-item app-link">
+            <a class="nav-link" routerLink="dashboard" routerLinkActive="active" (click)="closeMenu()"
               ngbPopover="Dashboard" i18n-ngbPopover [disablePopover]="!slimSidebarPopoversEnabled" placement="end"
               container="body" triggers="mouseenter:mouseleave" popoverClass="popover-slim">
               <i-bs class="me-2" name="house"></i-bs><span class="nav-link-label"><ng-container i18n>Dashboard</ng-container></span>
             </a>
-            @if (settingsService.organizingSidebarItems()) {
-              <pngx-input-switch class="position-absolute top-50 end-0 translate-middle-y me-1" [class.d-none]="slimSidebarEnabled || slimSidebarAnimating()" [compact]="true" title="Dashboard" i18n-title [ngModel]="!settingsService.sidebarItemIsHidden(HideableSidebarItemID.Dashboard)" (ngModelChange)="toggleSidebarItem(HideableSidebarItemID.Dashboard, $event)"></pngx-input-switch>
-            }
           </li>
           <li class="nav-item app-link" *pngxIfPermissions="{ action: PermissionAction.View, type: PermissionType.Document }">
             <a class="nav-link" routerLink="documents" routerLinkActive="active"
@@ -240,38 +237,37 @@
                 </div>
               </li>
             }
-            <li class="nav-item app-link position-relative" [class.d-none]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.SavedViews) && !settingsService.organizingSidebarItems()" *pngxIfPermissions="{ action: PermissionAction.View, type: PermissionType.SavedView }">
-              <a class="nav-link" [class.opacity-50]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.SavedViews)" [class.pe-5]="settingsService.organizingSidebarItems() && !slimSidebarEnabled && !slimSidebarAnimating()" routerLink="savedviews" routerLinkActive="active" (click)="closeMenu()"
+            <li class="nav-item app-link" *pngxIfPermissions="{ action: PermissionAction.View, type: PermissionType.SavedView }">
+              <a class="nav-link" routerLink="savedviews" routerLinkActive="active" (click)="closeMenu()"
                 ngbPopover="Saved Views" i18n-ngbPopover [disablePopover]="!slimSidebarPopoversEnabled" placement="end"
                 container="body" triggers="mouseenter:mouseleave" popoverClass="popover-slim">
                 <i-bs class="me-2" name="window-stack"></i-bs><span class="nav-link-label"><ng-container i18n>Saved Views</ng-container></span>
               </a>
-              @if (settingsService.organizingSidebarItems()) {
-                <pngx-input-switch class="position-absolute top-50 end-0 translate-middle-y me-1" [class.d-none]="slimSidebarEnabled || slimSidebarAnimating()" [compact]="true" title="Saved Views" i18n-title [ngModel]="!settingsService.sidebarItemIsHidden(HideableSidebarItemID.SavedViews)" (ngModelChange)="toggleSidebarItem(HideableSidebarItemID.SavedViews, $event)"></pngx-input-switch>
-              }
             </li>
-            <li class="nav-item app-link position-relative" [class.d-none]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Workflows) && !settingsService.organizingSidebarItems()"
+            <li class="nav-item app-link"
               *pngxIfPermissions="{ action: PermissionAction.View, type: PermissionType.Workflow }"
               tourAnchor="tour.workflows">
-              <a class="nav-link" [class.opacity-50]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Workflows)" [class.pe-5]="settingsService.organizingSidebarItems() && !slimSidebarEnabled && !slimSidebarAnimating()" routerLink="workflows" routerLinkActive="active" (click)="closeMenu()"
+              <a class="nav-link" routerLink="workflows" routerLinkActive="active" (click)="closeMenu()"
                 ngbPopover="Workflows" i18n-ngbPopover [disablePopover]="!slimSidebarPopoversEnabled" placement="end"
                 container="body" triggers="mouseenter:mouseleave" popoverClass="popover-slim">
                 <i-bs class="me-2" name="boxes"></i-bs><span class="nav-link-label"><ng-container i18n>Workflows</ng-container></span>
               </a>
-              @if (settingsService.organizingSidebarItems()) {
-                <pngx-input-switch class="position-absolute top-50 end-0 translate-middle-y me-1" [class.d-none]="slimSidebarEnabled || slimSidebarAnimating()" [compact]="true" title="Workflows" i18n-title [ngModel]="!settingsService.sidebarItemIsHidden(HideableSidebarItemID.Workflows)" (ngModelChange)="toggleSidebarItem(HideableSidebarItemID.Workflows, $event)"></pngx-input-switch>
-              }
             </li>
-            <li class="nav-item app-link position-relative" [class.d-none]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Mail) && !settingsService.organizingSidebarItems()" *pngxIfPermissions="{ action: PermissionAction.View, type: PermissionType.MailAccount }"
+            <li class="nav-item app-link"
+              *pngxIfPermissions="{ action: PermissionAction.View, type: PermissionType.OcrTemplate }">
+              <a class="nav-link" routerLink="ocr-templates" routerLinkActive="active" (click)="closeMenu()"
+                ngbPopover="OCR Templates" i18n-ngbPopover [disablePopover]="!slimSidebarEnabled" placement="end"
+                container="body" triggers="mouseenter:mouseleave" popoverClass="popover-slim">
+                <i-bs class="me-2" name="file-earmark-break"></i-bs><span><ng-container i18n>OCR Templates</ng-container></span>
+              </a>
+            </li>
+            <li class="nav-item app-link" *pngxIfPermissions="{ action: PermissionAction.View, type: PermissionType.MailAccount }"
               tourAnchor="tour.mail">
-              <a class="nav-link" [class.opacity-50]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Mail)" [class.pe-5]="settingsService.organizingSidebarItems() && !slimSidebarEnabled && !slimSidebarAnimating()" routerLink="mail" routerLinkActive="active" (click)="closeMenu()" ngbPopover="Mail"
+              <a class="nav-link" routerLink="mail" routerLinkActive="active" (click)="closeMenu()" ngbPopover="Mail"
                 i18n-ngbPopover [disablePopover]="!slimSidebarPopoversEnabled" placement="end" container="body"
                 triggers="mouseenter:mouseleave" popoverClass="popover-slim">
                 <i-bs class="me-2" name="envelope"></i-bs><span class="nav-link-label"><ng-container i18n>Mail</ng-container></span>
               </a>
-              @if (settingsService.organizingSidebarItems()) {
-                <pngx-input-switch class="position-absolute top-50 end-0 translate-middle-y me-1" [class.d-none]="slimSidebarEnabled || slimSidebarAnimating()" [compact]="true" title="Mail" i18n-title [ngModel]="!settingsService.sidebarItemIsHidden(HideableSidebarItemID.Mail)" (ngModelChange)="toggleSidebarItem(HideableSidebarItemID.Mail, $event)"></pngx-input-switch>
-              }
             </li>
             <li class="nav-item app-link" *pngxIfPermissions="{ action: PermissionAction.Delete, type: PermissionType.Document }">
               <a class="nav-link" routerLink="trash" routerLinkActive="active" (click)="closeMenu()" ngbPopover="Trash"
@@ -334,16 +330,13 @@
                 </a>
               </li>
             }
-            <li class="nav-item mt-2 position-relative" [class.d-none]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Documentation) && !settingsService.organizingSidebarItems()" tourAnchor="tour.outro">
-              <a class="text-muted small d-flex align-items-center flex-wrap text-decoration-none nav-anchor" [class.opacity-50]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Documentation)" [class.pe-5]="settingsService.organizingSidebarItems() && !slimSidebarEnabled && !slimSidebarAnimating()"
+            <li class="nav-item mt-2" tourAnchor="tour.outro">
+              <a class="text-muted small d-flex align-items-center flex-wrap text-decoration-none nav-anchor"
                 target="_blank" rel="noopener noreferrer" href="https://docs.paperless-ngx.com" ngbPopover="Documentation"
                 i18n-ngbPopover [disablePopover]="!slimSidebarPopoversEnabled" placement="end" container="body"
                 triggers="mouseenter:mouseleave" popoverClass="popover-slim">
                 <i-bs class="d-flex me-2" name="question-circle"></i-bs><span><ng-container i18n>Documentation</ng-container></span>
               </a>
-              @if (settingsService.organizingSidebarItems()) {
-                <pngx-input-switch class="position-absolute top-50 end-0 translate-middle-y me-1" [class.d-none]="slimSidebarEnabled || slimSidebarAnimating()" [compact]="true" title="Documentation" i18n-title [ngModel]="!settingsService.sidebarItemIsHidden(HideableSidebarItemID.Documentation)" (ngModelChange)="toggleSidebarItem(HideableSidebarItemID.Documentation, $event)"></pngx-input-switch>
-              }
             </li>
             <li class="nav-item" [class.visually-hidden]="slimSidebarEnabled">
               <div class="text-muted small d-flex align-items-center flex-wrap nav-label">

src-ui/src/app/components/app-frame/app-frame.component.spec.ts

@@ -15,7 +15,7 @@ import { provideUiTour } from 'ngx-ui-tour-ng-bootstrap'
 import { of, throwError } from 'rxjs'
 import { routes } from 'src/app/app-routing.module'
 import { SavedView } from 'src/app/data/saved-view'
-import { HideableSidebarItemID, SETTINGS_KEYS } from 'src/app/data/ui-settings'
+import { SETTINGS_KEYS } from 'src/app/data/ui-settings'
 import { IfPermissionsDirective } from 'src/app/directives/if-permissions.directive'
 import { PermissionsGuard } from 'src/app/guards/permissions.guard'
 import {
@@ -287,82 +287,6 @@ describe('AppFrameComponent', () => {
     jest.useRealTimers()
   })
 
-  it('should hide configured sidebar items', () => {
-    settingsService.set(SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS, [
-      HideableSidebarItemID.Dashboard,
-      HideableSidebarItemID.Workflows,
-    ])
-    fixture.detectChanges()
-
-    expect(
-      fixture.nativeElement.querySelector('[routerLink="dashboard"]')
-        .parentElement.classList
-    ).toContain('d-none')
-    expect(
-      fixture.nativeElement.querySelector('[routerLink="workflows"]')
-        .parentElement.classList
-    ).toContain('d-none')
-    expect(
-      fixture.nativeElement.querySelector('[routerLink="mail"]').parentElement
-        .classList
-    ).not.toContain('d-none')
-  })
-
-  it('should show hidden items and visibility switches while customizing', () => {
-    settingsService.set(SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS, [
-      HideableSidebarItemID.Dashboard,
-    ])
-    settingsService.sidebarHiddenItemsEditing.set([
-      HideableSidebarItemID.Dashboard,
-    ])
-    fixture.detectChanges()
-
-    expect(
-      fixture.nativeElement.querySelectorAll('pngx-input-switch').length
-    ).toBe(5)
-    expect(
-      fixture.nativeElement.querySelector('[routerLink="dashboard"]')
-        .parentElement.classList
-    ).not.toContain('d-none')
-    expect(
-      fixture.nativeElement.querySelector('[routerLink="dashboard"]').classList
-    ).toContain('opacity-50')
-
-    settingsService.set(SETTINGS_KEYS.SLIM_SIDEBAR, true)
-    fixture.detectChanges()
-
-    expect(
-      Array.from(
-        fixture.nativeElement.querySelectorAll('pngx-input-switch')
-      ).every((toggle: HTMLElement) => toggle.classList.contains('d-none'))
-    ).toBe(true)
-    expect(
-      fixture.nativeElement.querySelector('[routerLink="dashboard"]').classList
-    ).not.toContain('pe-5')
-
-    settingsService.set(SETTINGS_KEYS.SLIM_SIDEBAR, false)
-    component.slimSidebarAnimating.set(true)
-    fixture.detectChanges()
-
-    expect(
-      Array.from(
-        fixture.nativeElement.querySelectorAll('pngx-input-switch')
-      ).every((toggle: HTMLElement) => toggle.classList.contains('d-none'))
-    ).toBe(true)
-
-    component.slimSidebarAnimating.set(false)
-    fixture.detectChanges()
-
-    expect(
-      Array.from(
-        fixture.nativeElement.querySelectorAll('pngx-input-switch')
-      ).every((toggle: HTMLElement) => !toggle.classList.contains('d-none'))
-    ).toBe(true)
-    expect(
-      fixture.nativeElement.querySelector('[routerLink="dashboard"]').classList
-    ).toContain('pe-5')
-  })
-
   it('should show error on toggle slim sidebar if store settings fails', () => {
     jest.spyOn(console, 'warn').mockImplementation(() => {})
     const toastSpy = jest.spyOn(toastService, 'showError')

src-ui/src/app/components/app-frame/app-frame.component.ts

@@ -7,7 +7,6 @@ import {
 } from '@angular/cdk/drag-drop'
 import { NgClass } from '@angular/common'
 import { Component, HostListener, inject, OnInit, signal } from '@angular/core'
-import { FormsModule } from '@angular/forms'
 import { ActivatedRoute, Router, RouterModule } from '@angular/router'
 import {
   NgbCollapseModule,
@@ -22,11 +21,7 @@ import { Observable } from 'rxjs'
 import { first } from 'rxjs/operators'
 import { Document } from 'src/app/data/document'
 import { SavedView } from 'src/app/data/saved-view'
-import {
-  CollapsibleSection,
-  HideableSidebarItemID,
-  SETTINGS_KEYS,
-} from 'src/app/data/ui-settings'
+import { CollapsibleSection, SETTINGS_KEYS } from 'src/app/data/ui-settings'
 import { IfPermissionsDirective } from 'src/app/directives/if-permissions.directive'
 import { ComponentCanDeactivate } from 'src/app/guards/dirty-doc.guard'
 import { DocumentTitlePipe } from 'src/app/pipes/document-title.pipe'
@@ -53,7 +48,6 @@ import { ChatComponent } from '../chat/chat/chat.component'
 import { BrandMarkComponent } from '../common/logo/brand-mark/brand-mark.component'
 import { LogoComponent } from '../common/logo/logo.component'
 import { ProfileEditDialogComponent } from '../common/profile-edit-dialog/profile-edit-dialog.component'
-import { SwitchComponent } from '../common/input/switch/switch.component'
 import { DocumentDetailComponent } from '../document-detail/document-detail.component'
 import { ComponentWithPermissions } from '../with-permissions/with-permissions.component'
 import { GlobalSearchComponent } from './global-search/global-search.component'
@@ -82,8 +76,6 @@ const SCROLL_THRESHOLD = 16
     NgxBootstrapIconsModule,
     DragDropModule,
     TourNgBootstrap,
-    FormsModule,
-    SwitchComponent,
   ],
 })
 export class AppFrameComponent
@@ -106,7 +98,6 @@ export class AppFrameComponent
   readonly isMenuCollapsed = signal(true)
   readonly slimSidebarAnimating = signal(false)
   readonly mobileSearchHidden = signal(false)
-  readonly HideableSidebarItemID = HideableSidebarItemID
   private readonly versionSetting = this.settingsService.getSignal<string>(
     SETTINGS_KEYS.VERSION
   )
@@ -204,10 +195,6 @@ export class AppFrameComponent
     }, 200) // slightly longer than css animation for slim sidebar
   }
 
-  toggleSidebarItem(item: HideableSidebarItemID, visible: boolean): void {
-    this.settingsService.updateSidebarItemVisibility(item, visible)
-  }
-
   toggleAttributesSections(event?: Event): void {
     event?.preventDefault()
     event?.stopPropagation()

src-ui/src/app/components/common/input/switch/switch.component.html

@@ -1,6 +1,6 @@
-<div [class.mb-3]="!compact">
-  <div [class.row]="!compact">
-    @if (!horizontal && !compact) {
+<div class="mb-3">
+  <div class="row">
+    @if (!horizontal) {
       <div class="d-flex align-items-center position-relative hidden-button-container col-md-3">
         <label class="form-label" [for]="inputId" [ngbTooltip]="showUnsetNote && isUnset ? tipContent: null" placement="end">
           {{title}}
@@ -17,8 +17,8 @@
     }
     <div [ngClass]="{'align-items-center': horizontal, 'd-flex': horizontal}">
       <div class="form-check form-switch">
-        <input #inputField type="checkbox" class="form-check-input" [id]="inputId" [(ngModel)]="value" [ngModelOptions]="{standalone: true}" (change)="onChange(value)" (blur)="onTouched()" [disabled]="disabled" [attr.aria-label]="compact ? title : null">
-        @if (horizontal && !compact) {
+        <input #inputField type="checkbox" class="form-check-input" [id]="inputId" [(ngModel)]="value" [ngModelOptions]="{standalone: true}" (change)="onChange(value)" (blur)="onTouched()" [disabled]="disabled">
+        @if (horizontal) {
           <label class="form-check-label" [class.text-muted]="showUnsetNote && isUnset" [for]="inputId" [ngbTooltip]="showUnsetNote && isUnset ? tipContent: null" placement="end">
             {{title}}
             @if (showUnsetNote && isUnset) {

src-ui/src/app/components/common/input/switch/switch.component.spec.ts

@@ -48,14 +48,4 @@ describe('SwitchComponent', () => {
     component.value = undefined
     expect(component.isUnset).toBeTruthy()
   })
-
-  it('should support a compact layout', () => {
-    component.compact = true
-    component.title = 'Test switch'
-    fixture.detectChanges()
-
-    expect(fixture.nativeElement.querySelector('.mb-3')).toBeNull()
-    expect(fixture.nativeElement.querySelector('.row')).toBeNull()
-    expect(input.getAttribute('aria-label')).toEqual('Test switch')
-  })
 })

src-ui/src/app/components/common/input/switch/switch.component.ts

@@ -25,9 +25,6 @@ export class SwitchComponent extends AbstractInputComponent<boolean> {
   @Input()
   showUnsetNote: boolean = false
 
-  @Input()
-  compact: boolean = false
-
   constructor() {
     super()
   }

src-ui/src/app/components/document-detail/document-detail.component.html

@@ -82,6 +82,23 @@
         <i-bs name="pencil" class="me-1"></i-bs><ng-container i18n>PDF Editor</ng-container>
       </button>
 
+      <button
+        ngbDropdownItem
+        (click)="runZoneOcr()"
+        [disabled]="!userCanEdit || !document?.document_type"
+        *pngxIfPermissions="{ action: PermissionAction.Change, type: PermissionType.OcrTemplate }"
+      >
+        <i-bs width="1em" height="1em" name="file-earmark-ruled" class="me-1"></i-bs><span i18n>Run Zone OCR</span>
+      </button>
+
+      <button
+        ngbDropdownItem
+        (click)="createOcrTemplate()"
+        *pngxIfPermissions="{ action: PermissionAction.Add, type: PermissionType.OcrTemplate }"
+      >
+        <i-bs width="1em" height="1em" name="file-earmark-medical" class="me-1"></i-bs><span i18n>Create OCR Template</span>
+      </button>
+
       @if (userIsOwner && (requiresPassword || password)) {
         <button ngbDropdownItem (click)="removePassword()" [disabled]="!password">
           <i-bs name="unlock" class="me-1"></i-bs><ng-container i18n>Remove Password</ng-container>

src-ui/src/app/components/document-detail/document-detail.component.ts

@@ -1449,6 +1449,48 @@ export class DocumentDetailComponent
     })
   }
 
+  runZoneOcr() {
+    this.documentsService.runZoneOcr(this.document.id).subscribe({
+      next: (res) => {
+        const results = res.results ?? []
+        if (results.length) {
+          const failed = results.filter(
+            (r) =>
+              r.value === null ||
+              r.value === undefined ||
+              `${r.value}`.trim() === ''
+          )
+          const filled = results.length - failed.length
+          let msg = $localize`Filled ${filled} of ${results.length} fields`
+          if (failed.length) {
+            const names = failed.map((r) => r.zone).join(', ')
+            msg = `${msg}. ${$localize`Failed to match zones: ${names}`}`
+          }
+          this.toastService.showInfo(msg)
+        } else {
+          this.toastService.showInfo(
+            $localize`Zone OCR ran but no results extracted.`
+          )
+        }
+        this.documentsService
+          .get(this.documentId)
+          .subscribe((doc) => this.updateComponent(doc))
+      },
+      error: (error) => {
+        this.toastService.showError($localize`Zone OCR failed`, error)
+      },
+    })
+  }
+
+  createOcrTemplate() {
+    this.router.navigate(['/ocr-templates', 'new'], {
+      queryParams: {
+        document_type: this.document.document_type,
+        sample_document: this.document.id,
+      },
+    })
+  }
+
   private getSelectedNonLatestVersionId(): number | null {
     const versions = this.document()?.versions ?? []
     if (!versions.length || !this.selectedVersionId()) {

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

@@ -98,6 +98,9 @@
           <button ngbDropdownItem (click)="mergeSelectedAsVersions()" [disabled]="!userOwnsAll || !userCanEditAll || !userCanDelete || list.allSelected || list.selectedCount < 2">
             <i-bs name="journal-bookmark-fill" class="me-1"></i-bs><ng-container i18n>Merge as versions</ng-container>
           </button>
+          <button ngbDropdownItem (click)="runZoneOcrSelected()" [disabled]="!userCanEditAll || list.allSelected">
+            <i-bs name="file-earmark-ruled" class="me-1"></i-bs><ng-container i18n>Run Zone OCR</ng-container>
+          </button>
         </div>
       </div>
     </div>

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

@@ -19,7 +19,15 @@ import {
 } from '@ng-bootstrap/ng-bootstrap'
 import { saveAs } from 'file-saver'
 import { NgxBootstrapIconsModule } from 'ngx-bootstrap-icons'
-import { first, map, Observable, Subject, switchMap, takeUntil } from 'rxjs'
+import {
+  first,
+  forkJoin,
+  map,
+  Observable,
+  Subject,
+  switchMap,
+  takeUntil,
+} from 'rxjs'
 import { ConfirmDialogComponent } from 'src/app/components/common/confirm-dialog/confirm-dialog.component'
 import { CustomField } from 'src/app/data/custom-field'
 import { MatchingModel } from 'src/app/data/matching-model'
@@ -939,6 +947,27 @@ export class BulkEditorComponent
       })
   }
 
+  runZoneOcrSelected() {
+    const ids = Array.from(this.list.selected)
+    if (!ids.length) return
+    const modal = this.modalService.open(ConfirmDialogComponent, {
+      backdrop: 'static',
+    })
+    modal.componentInstance.title = $localize`Run Zone OCR`
+    modal.componentInstance.messageBold = $localize`Run zone OCR on ${this.getSelectionSize()} selected document(s)?`
+    modal.componentInstance.message = $localize`Each document's type template (if it has one) is applied, overwriting the mapped fields.`
+    modal.componentInstance.btnCaption = $localize`Proceed`
+    modal.componentInstance.confirmClicked
+      .pipe(takeUntil(this.unsubscribeNotifier))
+      .subscribe(() => {
+        modal.componentInstance.buttonsEnabled = false
+        this.executeDocumentAction(
+          modal,
+          forkJoin(ids.map((id) => this.documentService.runZoneOcr(id)))
+        )
+      })
+  }
+
   setPermissions() {
     let modal = this.modalService.open(PermissionsDialogComponent, {
       backdrop: 'static',

src-ui/src/app/components/manage/ocr-templates/ocr-template-editor/ocr-template-editor-zone-list/ocr-template-editor-zone-list.component.html

@@ -0,0 +1,34 @@
+@if (zones.length === 0) {
+  <p class="text-muted" i18n>
+    No zones defined. Load a document preview and draw rectangles to add zones.
+  </p>
+}
+
+<div class="list-group">
+  @for (zone of zones; track $index; let i = $index) {
+    <div
+      class="list-group-item list-group-item-action d-flex justify-content-between align-items-center"
+      [style.box-shadow]="selectedZoneIndex === i ? 'inset 3px 0 0 0 var(--bs-primary)' : null"
+    >
+      <div class="flex-grow-1" role="button" style="cursor: pointer;" (click)="zoneSelected.emit(i)">
+        <div>
+          <strong [class.text-primary]="selectedZoneIndex === i">
+            {{ zone.name }}
+          </strong>
+        </div>
+        <div class="small text-muted">
+          {{ getZoneTargetName(zone) }} - {{ zone.width }}x{{ zone.height }}px
+          <ng-container i18n>p.</ng-container>{{ zonePage(zone) }}
+        </div>
+      </div>
+      <div class="btn-group">
+        <button class="btn btn-sm btn-outline-secondary" type="button" (click)="zoneSelected.emit(i)" title="Edit" i18n-title>
+          <i-bs name="pencil"></i-bs>
+        </button>
+        <button class="btn btn-sm btn-outline-danger" type="button" (click)="zoneRemoved.emit(i)" title="Delete" i18n-title>
+          <i-bs name="trash"></i-bs>
+        </button>
+      </div>
+    </div>
+  }
+</div>

src-ui/src/app/components/manage/ocr-templates/ocr-template-editor/ocr-template-editor-zone-list/ocr-template-editor-zone-list.component.spec.ts

@@ -0,0 +1,72 @@
+import { ComponentFixture, TestBed } from '@angular/core/testing'
+import { NgxBootstrapIconsModule, allIcons } from 'ngx-bootstrap-icons'
+import { CustomField } from 'src/app/data/custom-field'
+import { OcrTemplateZone } from 'src/app/data/ocr-template'
+import { OcrTemplateEditorZoneListComponent } from './ocr-template-editor-zone-list.component'
+
+function zone(overrides: Partial<OcrTemplateZone> = {}): OcrTemplateZone {
+  return {
+    name: 'Zone 1',
+    target: 'custom_field',
+    custom_field: 7,
+    x: 10,
+    y: 20,
+    width: 30,
+    height: 40,
+    page: 1,
+    ocr_language: 'eng',
+    transform: 'strip',
+    validation_regex: '',
+    order: 0,
+    ...overrides,
+  }
+}
+
+describe('OcrTemplateEditorZoneListComponent', () => {
+  let fixture: ComponentFixture<OcrTemplateEditorZoneListComponent>
+  let component: OcrTemplateEditorZoneListComponent
+
+  beforeEach(async () => {
+    await TestBed.configureTestingModule({
+      imports: [
+        OcrTemplateEditorZoneListComponent,
+        NgxBootstrapIconsModule.pick(allIcons),
+      ],
+    }).compileComponents()
+
+    fixture = TestBed.createComponent(OcrTemplateEditorZoneListComponent)
+    component = fixture.componentInstance
+  })
+
+  it('shows empty state when no zones are defined', () => {
+    fixture.detectChanges()
+
+    expect(fixture.nativeElement.textContent).toContain('No zones defined')
+  })
+
+  it('renders zone target, size, and page', () => {
+    component.zones = [zone()]
+    component.customFields = [{ id: 7, name: 'Invoice Number' } as CustomField]
+    fixture.detectChanges()
+
+    const text = fixture.nativeElement.textContent
+    expect(text).toContain('Zone 1')
+    expect(text).toContain('Invoice Number')
+    expect(text).toContain('30x40px')
+    expect(text).toContain('p.1')
+  })
+
+  it('emits select and remove events', () => {
+    component.zones = [zone()]
+    const selectSpy = jest.spyOn(component.zoneSelected, 'emit')
+    const removeSpy = jest.spyOn(component.zoneRemoved, 'emit')
+    fixture.detectChanges()
+
+    const buttons = fixture.nativeElement.querySelectorAll('button')
+    buttons[0].click()
+    buttons[1].click()
+
+    expect(selectSpy).toHaveBeenCalledWith(0)
+    expect(removeSpy).toHaveBeenCalledWith(0)
+  })
+})

src-ui/src/app/components/manage/ocr-templates/ocr-template-editor/ocr-template-editor-zone-list/ocr-template-editor-zone-list.component.ts

@@ -0,0 +1,41 @@
+import { Component, EventEmitter, Input, Output } from '@angular/core'
+import { NgxBootstrapIconsModule } from 'ngx-bootstrap-icons'
+import { CustomField } from 'src/app/data/custom-field'
+import { OCR_BUILTIN_TARGETS, OcrTemplateZone } from 'src/app/data/ocr-template'
+import { getZonePage } from '../zone-geometry'
+
+@Component({
+  selector: 'pngx-ocr-template-zone-list',
+  imports: [NgxBootstrapIconsModule],
+  templateUrl: './ocr-template-editor-zone-list.component.html',
+})
+export class OcrTemplateEditorZoneListComponent {
+  @Input() zones: OcrTemplateZone[] = []
+  @Input() selectedZoneIndex: number | null = null
+  @Input() previewPage = 0
+  @Input() previewPageCount: number | null = null
+  @Input() customFields: CustomField[] = []
+
+  @Output() zoneSelected = new EventEmitter<number>()
+  @Output() zoneRemoved = new EventEmitter<number>()
+
+  zonePage(zone: OcrTemplateZone): number {
+    return getZonePage(zone, this.previewPage, this.previewPageCount)
+  }
+
+  getZoneTargetName(zone: OcrTemplateZone): string {
+    const target = zone.target || 'custom_field'
+    if (target === 'custom_field') {
+      return zone.custom_field
+        ? this.getCustomFieldName(zone.custom_field)
+        : $localize`(no field)`
+    }
+    return OCR_BUILTIN_TARGETS.find((t) => t.id === target)?.name ?? target
+  }
+
+  private getCustomFieldName(id: number): string {
+    return (
+      this.customFields.find((field) => field.id === id)?.name ?? `Field #${id}`
+    )
+  }
+}

src-ui/src/app/components/manage/ocr-templates/ocr-template-editor/ocr-template-editor.component.html

@@ -0,0 +1,442 @@
+<pngx-page-header [title]="pageTitle" [id]="template.id">
+    <div class="input-group input-group-sm me-5 align-items-center">
+      <div class="input-group-text">
+        <i-bs name="file-text"></i-bs>
+      </div>
+        <input
+        type="text"
+        class="form-control"
+        [(ngModel)]="previewDocModel"
+        [ngbTypeahead]="searchDocuments"
+        [inputFormatter]="documentFormatter"
+        [resultFormatter]="documentFormatter"
+        (selectItem)="onPreviewDocSelected($event)"
+        [editable]="false"
+        placeholder="Search documents by title..."
+        i18n-placeholder
+      />
+  </div>
+
+  <div class="d-flex align-items-center flex-wrap gap-2">
+    <div class="input-group input-group-sm ms-2 d-none d-md-flex">
+      <div class="input-group-text" i18n>Page</div>
+        <input class="form-control flex-grow-0 w-auto" type="number" min="1" [max]="previewPageCount" [(ngModel)]="previewPageDisplay" />
+      <div class="input-group-text" i18n>of {{previewPageCount}}</div>
+    </div>
+    <button type="button" class="btn btn-sm btn-outline-secondary" i18n-title title="Previous" (click)="prevPage()" [disabled]="!pageImageUrl || previewPage <= 0">
+      <i-bs width="1.2em" height="1.2em" name="arrow-left"></i-bs>
+    </button>
+    <button type="button" class="btn btn-sm btn-outline-secondary"  i18n-title title="Next" (click)="nextPage()" [disabled]="!pageImageUrl || previewPage >= (previewPageCount ?? 1) - 1">
+      <i-bs width="1.2em" height="1.2em" name="arrow-right"></i-bs>
+    </button>
+
+    <div class="input-group input-group-sm">
+      <button class="btn btn-outline-secondary" (click)="zoomOut()" i18n>-</button>
+      <span class="input-group-text">{{ zoom * 100 | number: '1.0-0' }}%</span>
+      <button class="btn btn-outline-secondary" (click)="zoomIn()" i18n>+</button>
+    </div>
+  </div>
+</pngx-page-header>
+
+<div class="row">
+  <div class="col-md-4">
+    <div class="btn-toolbar mb-1 border-bottom">
+        <div class="btn-group pb-3">
+          <a routerLink="/ocr-templates" class="btn btn-sm btn-outline-secondary">
+            <i-bs width="1.2em" height="1.2em" name="x"></i-bs>
+            <span class="ms-1" i18n>Close</span>
+          </a>
+        </div>
+        <div class="btn-group ms-auto pb-3">
+          <button class="btn btn-sm btn-primary" (click)="save()" [disabled]="saving">
+          @if (saving) {
+            <span class="spinner-border spinner-border-sm me-1"></span>
+          }
+          <span i18n>Save</span>
+          </button>
+      </div>
+    </div>
+
+    <ul ngbNav #nav="ngbNav" [(activeId)]="activeTab" class="nav-underline flex-nowrap flex-md-wrap overflow-auto">
+        <li ngbNavItem="settings">
+          <a ngbNavLink i18n>Settings</a>
+          <ng-template ngbNavContent>
+            <div class="row mb-3">
+              <div class="col-9">
+                <pngx-input-text [(ngModel)]="template.name" title="Template name" i18n-title></pngx-input-text>
+              </div>
+              <div class="col-3">
+                <pngx-input-switch [(ngModel)]="template.enabled" title="Enabled" i18n-title></pngx-input-switch>
+              </div>
+            </div>
+
+            <pngx-input-select [(ngModel)]="template.document_type" [items]="documentTypes" bindLabel="name" bindValue="id" title="Document type" i18n-title></pngx-input-select>
+
+            <small class="text-muted" i18n>
+              Draw rectangles on the preview to define extraction zones. Use the
+              page controls above the preview to add zones on different pages.
+            </small>
+          </ng-template>
+        </li>
+
+        <li ngbNavItem="zones">
+          <a ngbNavLink><ng-container i18n>Zones</ng-container> <span class="badge bg-primary ms-2">{{ template.zones.length }}</span></a>
+          <ng-template ngbNavContent>
+            <pngx-ocr-template-zone-list
+              [zones]="template.zones"
+              [selectedZoneIndex]="selectedZoneIndex"
+              [previewPage]="previewPage"
+              [previewPageCount]="previewPageCount"
+              [customFields]="customFields"
+              (zoneSelected)="selectZone($event)"
+              (zoneRemoved)="removeZone($event)"
+            ></pngx-ocr-template-zone-list>
+          </ng-template>
+        </li>
+
+        <li ngbNavItem="zone">
+          <a ngbNavLink i18n>Zone</a>
+          <ng-template ngbNavContent>
+            @if (selectedZone; as zone) {
+              <div class="d-flex justify-content-between align-items-center mb-3">
+                <strong>{{ zone.name }}</strong>
+                <div class="d-flex gap-2">
+                  <button class="btn btn-sm btn-primary" (click)="save()" [disabled]="saving">
+                    @if (saving) {
+                      <span class="spinner-border spinner-border-sm me-1"></span>
+                    }
+                    <span i18n>Save</span>
+                  </button>
+                  <button class="btn btn-sm btn-outline-danger" (click)="deleteSelectedZone()">
+                    <i-bs name="trash" class="me-1"></i-bs><ng-container i18n>Delete zone</ng-container>
+                  </button>
+                </div>
+              </div>
+
+              <div class="mb-3">
+                <label class="form-label" i18n>Zone Name</label>
+                <input
+                  type="text"
+                  class="form-control"
+                  [(ngModel)]="zone.name"
+                />
+              </div>
+
+              <div class="mb-3">
+                <label class="form-label" i18n>Page</label>
+                <input
+                  type="number"
+                  class="form-control"
+                  [(ngModel)]="zone.page"
+                  min="-1"
+                />
+                <small class="text-muted" i18n>Page this zone is on. Use -1 for the last page. Set automatically when you draw it.</small>
+              </div>
+
+              <div class="mb-3">
+                <label class="form-label" i18n>Field</label>
+                <div class="input-group">
+                  <select class="form-select" [ngModel]="zoneFieldValue(zone)" (ngModelChange)="setZoneField(zone, $event)">
+                    <optgroup label="Built-in fields" i18n-label>
+                      @for (t of builtinTargets; track t.id) {
+                        <option [ngValue]="t.id">{{ t.name }}</option>
+                      }
+                    </optgroup>
+                    <optgroup label="Custom fields" i18n-label>
+                      @for (cf of customFields; track cf.id) {
+                        <option [ngValue]="cf.id">{{ cf.name }} ({{ cf.data_type }})</option>
+                      }
+                    </optgroup>
+                  </select>
+                  <button
+                    class="btn btn-outline-secondary"
+                    type="button"
+                    (click)="openQuickCreate(selectedZoneIndex)"
+                    title="Create new custom field"
+                    i18n-title
+                  >
+                    <i-bs name="plus"></i-bs>
+                  </button>
+                </div>
+                <small class="text-muted" i18n>Write the extracted value to a custom field, or to a built-in field (Title, ASN, Date created).</small>
+              </div>
+
+              @if (isFieldShared(zone)) {
+                <div class="card mb-3 border-info">
+                  <div class="card-body">
+                    <h6 class="card-title d-flex align-items-center gap-2">
+                      <i-bs name="braces"></i-bs>
+                      <span i18n>Combine zones into this field</span>
+                    </h6>
+                    <p class="small text-muted mb-2" i18n>
+                      More than one zone writes to this field. Build the combined
+                      value below: click a zone to insert its token, and type any
+                      separators or literal text between tokens.
+                    </p>
+                    <div class="d-flex flex-wrap gap-1 mb-2">
+                      @for (z of zonesForField(zone); track $index) {
+                        <button
+                          type="button"
+                          class="btn btn-sm btn-outline-info"
+                          (click)="insertCombineToken(zone, z)"
+                          title="Insert token"
+                          i18n-title
+                        >
+                          + {{ z.name || 'Zone' }}
+                        </button>
+                      }
+                    </div>
+                    <input
+                      type="text"
+                      class="form-control font-monospace"
+                      [ngModel]="getCombineFormat(zone)"
+                      (ngModelChange)="setCombineFormat(zone, $event)"
+                      placeholder="{Zone 1} - {Zone 2}"
+                    />
+                    <small class="text-muted" i18n>
+                      Tokens are matched by zone name. An empty zone leaves its
+                      token blank and the stray separator is trimmed. Leave empty
+                      to just join the zones in order with a space.
+                    </small>
+                  </div>
+                </div>
+              }
+
+              @if (showQuickCreate) {
+                <div class="card mb-3 border-primary">
+                  <div class="card-body">
+                    <h6 class="card-title" i18n>Create Custom Field</h6>
+                    <div class="mb-2">
+                      <label class="form-label small" i18n>Field Name</label>
+                      <input type="text" class="form-control form-control-sm"
+                        [(ngModel)]="quickCreateName" placeholder="e.g. Invoice Number" />
+                    </div>
+                    <div class="mb-2">
+                      <label class="form-label small" i18n>Field Type</label>
+                      <select class="form-select form-select-sm" [(ngModel)]="quickCreateType">
+                        @for (t of quickCreateTypes; track t.id) {
+                          <option [ngValue]="t.id">{{ t.name }}</option>
+                        }
+                      </select>
+                    </div>
+                    <div class="d-flex gap-2">
+                      <button class="btn btn-primary btn-sm" (click)="submitQuickCreate()"
+                        [disabled]="!quickCreateName.trim()" i18n>
+                        Create & Assign
+                      </button>
+                      <button class="btn btn-outline-secondary btn-sm" (click)="cancelQuickCreate()" i18n>
+                        Cancel
+                      </button>
+                    </div>
+                  </div>
+                </div>
+              }
+
+              <div class="mb-3">
+                <label class="form-label" i18n>OCR Language</label>
+                <ng-select
+                  [items]="ocrLanguageOptions"
+                  bindLabel="name"
+                  bindValue="id"
+                  [multiple]="true"
+                  [closeOnSelect]="false"
+                  [ngModel]="ocrLanguageArray(zone)"
+                  (ngModelChange)="setOcrLanguages(zone, $event)"
+                  placeholder="Select languages"
+                  i18n-placeholder
+                ></ng-select>
+              </div>
+
+              <div class="mb-3">
+                <label class="form-label" i18n>Transform</label>
+                <select class="form-select" [(ngModel)]="zone.transform">
+                  @for (opt of transformOptions; track opt.id) {
+                    <option [ngValue]="opt.id">{{ opt.name }}</option>
+                  }
+                </select>
+              </div>
+
+              @if (zone.transform === dateTransform) {
+                <div class="mb-3">
+                  <label class="form-label" i18n>Date format</label>
+                  <select class="form-select" [ngModel]="dateFormatChoice(zone)" (ngModelChange)="setDateFormatChoice(zone, $event)">
+                    @for (opt of dateFormatOptions; track opt.id) {
+                      <option [ngValue]="opt.id">{{ opt.name }}</option>
+                    }
+                    <option [ngValue]="customDateFormatChoice" i18n>Custom...</option>
+                  </select>
+                  @if (usesCustomDateFormat(zone)) {
+                    <div class="input-group mt-2">
+                      <input type="text" class="form-control font-monospace" [(ngModel)]="zone.date_format" placeholder="%d.%m.%Y" />
+                      <button class="btn btn-outline-secondary" type="button" [ngbPopover]="dateFmtHelp" [autoClose]="true" title="Date format help" i18n-title>
+                        <i-bs name="question-circle"></i-bs>
+                      </button>
+                    </div>
+                    <ng-template #dateFmtHelp>
+                      <p class="mb-1" i18n>Python date codes:</p>
+                      <ul class="mb-1 ps-3">
+                        <li><code>%d</code> <ng-container i18n>day (01-31)</ng-container></li>
+                        <li><code>%m</code> <ng-container i18n>month (01-12)</ng-container></li>
+                        <li><code>%Y</code> <ng-container i18n>year, 4-digit</ng-container></li>
+                        <li><code>%y</code> <ng-container i18n>year, 2-digit</ng-container></li>
+                        <li><code>%b</code> <ng-container i18n>month name (Jan)</ng-container></li>
+                      </ul>
+                      <span i18n>Example:</span> <code>%d.%m.%Y</code> -> 03.03.2026
+                    </ng-template>
+                  }
+                </div>
+              }
+
+              <div class="mb-3">
+                <label class="form-label" i18n>Validation Regex</label>
+                <input
+                  type="text"
+                  class="form-control font-monospace"
+                  [(ngModel)]="zone.validation_regex"
+                  placeholder="e.g. \d{2}\.\d{2}\.\d{4}"
+                >
+              </div>
+
+              <div class="text-muted small">
+                {{ zone.x }}, {{ zone.y }} - {{ zone.width }}x{{ zone.height }}px
+              </div>
+
+              <hr class="my-3" />
+              <h6 i18n>Test</h6>
+              @if (!previewDocId) {
+                <p class="text-muted small mb-0" i18n>
+                  Load a document in the Settings tab to test this zone.
+                </p>
+              } @else {
+                <button class="btn btn-sm btn-outline-secondary" (click)="testZone()" [disabled]="zoneTesting">
+                  @if (zoneTesting) {
+                    <span class="spinner-border spinner-border-sm me-1"></span>
+                  }
+                  <span i18n>Test this zone</span>
+                </button>
+                @if (zoneTestResult) {
+                  @if (zoneTestResult.error) {
+                    <div class="alert alert-warning py-2 mt-2 mb-0 small">{{ zoneTestResult.error }}</div>
+                  } @else {
+                    <dl class="row small mt-2 mb-0">
+                      <dt class="col-sm-4" i18n>OCR text</dt>
+                      <dd class="col-sm-8"><code>{{ zoneTestResult.raw_text || '(nothing detected)' }}</code></dd>
+                      <dt class="col-sm-4" i18n>Value</dt>
+                      <dd class="col-sm-8"><code>{{ zoneTestResult.value || '(empty)' }}</code></dd>
+                      @if (zoneTestResult.regex) {
+                        <dt class="col-sm-4" i18n>Validation</dt>
+                        <dd class="col-sm-8">
+                          @if (zoneTestResult.regex_match) {
+                            <span class="badge bg-success" i18n>Regex matches</span>
+                          } @else {
+                            <span class="badge bg-danger" i18n>Regex does not match</span>
+                          }
+                        </dd>
+                      }
+                    </dl>
+                  }
+                }
+              }
+            } @else {
+              <p class="text-muted" i18n>
+                Select a zone from the Zones tab, or draw a rectangle on the document to create one.
+              </p>
+            }
+          </ng-template>
+        </li>
+      </ul>
+
+      <div [ngbNavOutlet]="nav" class="mt-3"></div>
+    </div>
+
+    <!-- Right column: Document preview with zone overlay -->
+    <div class="col-md-8">
+      @if (pageImageUrl) {
+        <div class="zone-preview-scroll border">
+        <div class="zone-preview-stage" [style.width.%]="zoom * 100">
+          <img
+            #pageImage
+            [src]="pageImageUrl"
+            (load)="onImageLoad()"
+            class="zone-preview-image"
+            [style.visibility]="imageLoaded ? 'visible' : 'hidden'"
+            crossorigin="use-credentials"
+          />
+          @if (imageLoaded) {
+            <svg
+              #zoneOverlay
+              class="zone-overlay"
+              [attr.viewBox]="overlayViewBox()"
+              preserveAspectRatio="none"
+              [style.cursor]="overlayCursor"
+              (mousedown)="onOverlayMouseDown($event)"
+              (mousemove)="onOverlayMouseMove($event)"
+              (mouseup)="onOverlayMouseUp($event)"
+            >
+              @for (zone of template.zones; track $index; let i = $index) {
+                @if (zoneDisplayRect(i); as rect) {
+                  <g>
+                    <rect
+                      class="zone-rect"
+                      [class.zone-rect-selected]="selectedZoneIndex === i"
+                      [attr.x]="rect.x"
+                      [attr.y]="rect.y"
+                      [attr.width]="rect.w"
+                      [attr.height]="rect.h"
+                      [attr.stroke]="zoneColor(i)"
+                      [attr.fill]="zoneFill(i)"
+                    ></rect>
+                    <text
+                      class="zone-label"
+                      [attr.x]="rect.x + overlayUnitSize(6)"
+                      [attr.y]="zoneLabelY(rect)"
+                      [attr.font-size]="overlayFontSize()"
+                      [attr.fill]="zoneColor(i)"
+                    >{{ zoneLabel(zone, i) }}</text>
+
+                    @if (selectedZoneIndex === i) {
+                      @for (handle of resizeHandles(rect); track handle.handle) {
+                        <rect
+                          class="zone-resize-handle"
+                          [attr.x]="handle.x - overlayHandleSize() / 2"
+                          [attr.y]="handle.y - overlayHandleSize() / 2"
+                          [attr.width]="overlayHandleSize()"
+                          [attr.height]="overlayHandleSize()"
+                          [attr.fill]="zoneColor(i)"
+                        ></rect>
+                      }
+                    }
+                  </g>
+                }
+              }
+
+              @if (drawingRect(); as rect) {
+                <rect
+                  class="zone-drawing-rect"
+                  [attr.x]="rect.x"
+                  [attr.y]="rect.y"
+                  [attr.width]="rect.w"
+                  [attr.height]="rect.h"
+                ></rect>
+              }
+            </svg>
+          }
+          @if (!imageLoaded) {
+            <div class="d-flex justify-content-center p-5">
+              <div class="spinner-border" role="status">
+                <span class="visually-hidden" i18n>Loading page...</span>
+              </div>
+            </div>
+          }
+        </div>
+        </div>
+      } @else {
+        <div class="border rounded p-5 text-center text-muted">
+          <i-bs name="file-earmark-image" width="48" height="48"></i-bs>
+          <p class="mt-3" i18n>
+            Enter a document ID and click "Load" to preview a page and draw extraction zones.
+          </p>
+        </div>
+      }
+    </div>
+  </div>

src-ui/src/app/components/manage/ocr-templates/ocr-template-editor/ocr-template-editor.component.scss

@@ -0,0 +1,63 @@
+:host {
+  display: block;
+}
+
+.zone-preview-scroll {
+  max-height: 78vh;
+  overflow: auto;
+}
+
+.zone-preview-stage {
+  display: inline-block;
+  position: relative;
+}
+
+.zone-preview-image {
+  display: block;
+  width: 100%;
+}
+
+.zone-overlay {
+  height: 100%;
+  inset: 0;
+  position: absolute;
+  touch-action: none;
+  width: 100%;
+}
+
+.zone-rect,
+.zone-drawing-rect {
+  vector-effect: non-scaling-stroke;
+}
+
+.zone-rect {
+  stroke-width: 2;
+}
+
+.zone-rect-selected {
+  stroke-width: 3;
+}
+
+.zone-label {
+  font-family: var(--bs-font-sans-serif);
+  font-weight: 600;
+  paint-order: stroke;
+  pointer-events: none;
+  stroke: #fff;
+  stroke-linejoin: round;
+  stroke-width: 4px;
+  vector-effect: non-scaling-stroke;
+}
+
+.zone-resize-handle {
+  stroke: #fff;
+  stroke-width: 1;
+  vector-effect: non-scaling-stroke;
+}
+
+.zone-drawing-rect {
+  fill: rgba(105, 219, 124, 0.25);
+  stroke: #69db7c;
+  stroke-dasharray: 5 5;
+  stroke-width: 2;
+}

src-ui/src/app/components/manage/ocr-templates/ocr-template-editor/ocr-template-editor.component.ts

@@ -0,0 +1,962 @@
+import { CommonModule } from '@angular/common'
+import {
+  Component,
+  ElementRef,
+  HostListener,
+  inject,
+  OnDestroy,
+  OnInit,
+  ViewChild,
+} from '@angular/core'
+import { FormsModule } from '@angular/forms'
+import { ActivatedRoute, Router, RouterModule } from '@angular/router'
+import {
+  NgbNavModule,
+  NgbPopoverModule,
+  NgbTypeaheadModule,
+  NgbTypeaheadSelectItemEvent,
+} from '@ng-bootstrap/ng-bootstrap'
+import { NgSelectModule } from '@ng-select/ng-select'
+import { NgxBootstrapIconsModule } from 'ngx-bootstrap-icons'
+import {
+  catchError,
+  debounceTime,
+  distinctUntilChanged,
+  map,
+  Observable,
+  of,
+  Subject,
+  switchMap,
+  takeUntil,
+} from 'rxjs'
+import { SelectComponent } from 'src/app/components/common/input/select/select.component'
+import { SwitchComponent } from 'src/app/components/common/input/switch/switch.component'
+import { TextComponent } from 'src/app/components/common/input/text/text.component'
+import { PageHeaderComponent } from 'src/app/components/common/page-header/page-header.component'
+import { CustomField, CustomFieldDataType } from 'src/app/data/custom-field'
+import { Document } from 'src/app/data/document'
+import { DocumentType } from 'src/app/data/document-type'
+import {
+  DATE_FORMAT_OPTIONS,
+  DEFAULT_OCR_ZONE_LANGUAGE,
+  DEFAULT_OCR_ZONE_TARGET,
+  DEFAULT_OCR_ZONE_TRANSFORM,
+  isOcrBuiltinTarget,
+  OCR_BUILTIN_TARGETS,
+  OCR_LANGUAGE_OPTIONS,
+  OCR_ZONE_TARGET,
+  OCR_ZONE_TRANSFORM,
+  OcrBuiltinTarget,
+  OcrTemplate,
+  OcrTemplateZone,
+  OcrZoneTestResult,
+  TRANSFORM_OPTIONS,
+  ZoneTestRequest,
+} from 'src/app/data/ocr-template'
+import { CorrespondentService } from 'src/app/services/rest/correspondent.service'
+import { CustomFieldsService } from 'src/app/services/rest/custom-fields.service'
+import { DocumentTypeService } from 'src/app/services/rest/document-type.service'
+import { DocumentService } from 'src/app/services/rest/document.service'
+import { OcrTemplateService } from 'src/app/services/rest/ocr-template.service'
+import { ToastService } from 'src/app/services/toast.service'
+import { OcrTemplateEditorZoneListComponent } from './ocr-template-editor-zone-list/ocr-template-editor-zone-list.component'
+import {
+  DisplayRect,
+  DrawingRect,
+  findHandleAt,
+  findZoneAt,
+  getZoneDisplayRect,
+  getZonePage,
+  HANDLE_SIZE,
+  isZoneOnPage,
+  MoveStart,
+  moveZone,
+  Point,
+  ResizeHandle,
+  resizeZone,
+} from './zone-geometry'
+
+type ActiveTab = 'settings' | 'zones' | 'zone'
+type ZoneFieldSelection = OcrBuiltinTarget | number | null
+type OverlayInteraction =
+  | { kind: 'idle' }
+  | { kind: 'drawing'; rect: DrawingRect }
+  | { kind: 'moving'; zoneIndex: number; start: MoveStart }
+  | { kind: 'resizing'; zoneIndex: number; handle: ResizeHandle }
+interface ResizeHandleMarker extends Point {
+  handle: ResizeHandle
+}
+
+const CUSTOM_DATE_FORMAT_CHOICE = 'custom'
+const MIN_DRAWN_ZONE_SIZE = 10
+const NO_OVERLAY_INTERACTION: OverlayInteraction = { kind: 'idle' }
+const ZONE_COLORS = [
+  '#4f8ff7',
+  '#ff6b6b',
+  '#51cf66',
+  '#ffd43b',
+  '#cc5de8',
+  '#ff922b',
+  '#20c997',
+  '#e599f7',
+]
+const RESIZE_CURSOR: Record<ResizeHandle, string> = {
+  nw: 'nw-resize',
+  ne: 'ne-resize',
+  sw: 'sw-resize',
+  se: 'se-resize',
+  n: 'n-resize',
+  s: 's-resize',
+  w: 'w-resize',
+  e: 'e-resize',
+}
+
+@Component({
+  selector: 'pngx-ocr-template-editor',
+  standalone: true,
+  imports: [
+    PageHeaderComponent,
+    TextComponent,
+    SelectComponent,
+    SwitchComponent,
+    CommonModule,
+    FormsModule,
+    RouterModule,
+    NgbNavModule,
+    NgbPopoverModule,
+    NgbTypeaheadModule,
+    NgSelectModule,
+    NgxBootstrapIconsModule,
+    OcrTemplateEditorZoneListComponent,
+  ],
+  templateUrl: './ocr-template-editor.component.html',
+  styleUrls: ['./ocr-template-editor.component.scss'],
+})
+export class OcrTemplateEditorComponent implements OnInit, OnDestroy {
+  private readonly route = inject(ActivatedRoute)
+  private readonly router = inject(Router)
+  private readonly templateService = inject(OcrTemplateService)
+  private readonly customFieldsService = inject(CustomFieldsService)
+  private readonly documentTypeService = inject(DocumentTypeService)
+  private readonly correspondentService = inject(CorrespondentService)
+  private readonly documentService = inject(DocumentService)
+  private readonly toastService = inject(ToastService)
+  private readonly destroy$ = new Subject<void>()
+  private readonly customDateFormatZones = new WeakSet<OcrTemplateZone>()
+
+  @ViewChild('zoneOverlay') overlayRef: ElementRef<SVGSVGElement>
+  @ViewChild('pageImage') imageRef: ElementRef<HTMLImageElement>
+
+  template: OcrTemplate = {
+    id: null,
+    name: '',
+    document_type: null,
+    sample_document: null,
+    source_width: 0,
+    source_height: 0,
+    enabled: true,
+    combine_formats: {},
+    zones: [],
+  }
+
+  customFields: CustomField[] = []
+  documentTypes: DocumentType[] = []
+  transformOptions = TRANSFORM_OPTIONS
+  builtinTargets = OCR_BUILTIN_TARGETS
+  dateFormatOptions = DATE_FORMAT_OPTIONS
+  ocrLanguageOptions = OCR_LANGUAGE_OPTIONS
+  dateTransform = OCR_ZONE_TRANSFORM.Date
+  customDateFormatChoice = CUSTOM_DATE_FORMAT_CHOICE
+  isNew = true
+  saving = false
+
+  previewDocId: number | null = null
+  previewPage = 0
+  previewPageCount: number | null = null
+  private pageCountForDoc: number | null = null
+  pageImageUrl: string | null = null
+  imageLoaded = false
+  zoom = 1
+  previewDocModel: Document | string = ''
+  private correspondentNames = new Map<number, string>()
+
+  public get previewPageDisplay(): number {
+    return this.previewPage + 1
+  }
+
+  public set previewPageDisplay(value: number) {
+    this.goToPage(value - 1)
+  }
+
+  activeTab: ActiveTab = 'settings'
+
+  selectedZoneIndex: number | null = null
+  private overlayInteraction: OverlayInteraction = NO_OVERLAY_INTERACTION
+  overlayCursor = 'crosshair'
+
+  zoneTestResult: OcrZoneTestResult | null = null
+  zoneTesting = false
+
+  showQuickCreate = false
+  quickCreateName = ''
+  quickCreateType = CustomFieldDataType.String
+  quickCreateForZoneIndex: number | null = null
+  quickCreateTypes = [
+    { id: CustomFieldDataType.String, name: $localize`String` },
+    { id: CustomFieldDataType.Integer, name: $localize`Integer` },
+    { id: CustomFieldDataType.Float, name: $localize`Float` },
+    { id: CustomFieldDataType.Date, name: $localize`Date` },
+    { id: CustomFieldDataType.Monetary, name: $localize`Monetary` },
+    { id: CustomFieldDataType.Boolean, name: $localize`Boolean` },
+    { id: CustomFieldDataType.Url, name: $localize`URL` },
+    { id: CustomFieldDataType.LongText, name: $localize`Long Text` },
+  ]
+
+  get selectedZone(): OcrTemplateZone | null {
+    return this.selectedZoneIndex !== null
+      ? (this.template.zones[this.selectedZoneIndex] ?? null)
+      : null
+  }
+
+  get pageTitle(): string {
+    return this.isNew
+      ? $localize`New OCR Template`
+      : $localize`Edit OCR Template`
+  }
+
+  ngOnInit() {
+    this.customFieldsService
+      .listAll()
+      .pipe(takeUntil(this.destroy$))
+      .subscribe((r) => (this.customFields = r.results))
+
+    this.documentTypeService
+      .listAll()
+      .pipe(takeUntil(this.destroy$))
+      .subscribe((r) => (this.documentTypes = r.results))
+
+    this.correspondentService
+      .listAll()
+      .pipe(takeUntil(this.destroy$))
+      .subscribe((r) => {
+        this.correspondentNames = new Map(r.results.map((c) => [c.id, c.name]))
+      })
+
+    const id = this.route.snapshot.paramMap.get('id')
+    if (id && id !== 'new') {
+      this.isNew = false
+      this.templateService
+        .get(parseInt(id))
+        .pipe(takeUntil(this.destroy$))
+        .subscribe((t) => {
+          this.template = t
+          this.template.combine_formats ??= {}
+          if (t.sample_document) {
+            this.previewDocId = t.sample_document
+            this.loadPreview()
+          }
+        })
+    } else {
+      const qp = this.route.snapshot.queryParams
+      if (qp['document_type']) {
+        this.template.document_type = parseInt(qp['document_type'])
+      }
+      if (qp['sample_document']) {
+        const docId = parseInt(qp['sample_document'])
+        this.template.sample_document = docId
+        this.previewDocId = docId
+        this.loadPreview()
+      }
+    }
+  }
+
+  searchDocuments = (text$: Observable<string>): Observable<Document[]> =>
+    text$.pipe(
+      debounceTime(250),
+      distinctUntilChanged(),
+      switchMap((term) => {
+        if (!term || term.trim().length < 2) return of([])
+        const params: { title__icontains: string; document_type__id?: number } =
+          { title__icontains: term.trim() }
+        if (this.template.document_type) {
+          params['document_type__id'] = this.template.document_type
+        }
+        return this.documentService.list(1, 10, 'created', true, params).pipe(
+          map((r) => r.results),
+          catchError(() => of([]))
+        )
+      })
+    )
+
+  documentFormatter = (doc: Document | string): string => {
+    if (typeof doc === 'string') return doc
+    const corr = doc.correspondent
+      ? this.correspondentNames.get(doc.correspondent)
+      : null
+    return corr
+      ? `#${doc.id} ${doc.title} (${corr})`
+      : `#${doc.id} ${doc.title}`
+  }
+
+  onPreviewDocSelected(event: NgbTypeaheadSelectItemEvent<Document>) {
+    event.preventDefault()
+    const doc: Document = event.item
+    this.previewDocModel = doc
+    this.previewDocId = doc.id
+    if (!this.template.document_type && doc.document_type) {
+      this.template.document_type = doc.document_type
+    }
+    this.previewPage = 0
+    this.loadPreview()
+  }
+
+  clearPreviewDoc() {
+    this.previewDocModel = ''
+    this.previewDocId = null
+    this.previewPageCount = null
+    this.pageCountForDoc = null
+    this.previewPage = 0
+    this.pageImageUrl = null
+    this.imageLoaded = false
+  }
+
+  loadPreview() {
+    if (!this.previewDocId) return
+    if (this.pageCountForDoc !== this.previewDocId) {
+      this.pageCountForDoc = this.previewDocId
+      this.previewPageCount = null
+      this.documentService
+        .get(this.previewDocId)
+        .pipe(takeUntil(this.destroy$))
+        .subscribe({
+          next: (doc) => {
+            this.previewPageCount = doc?.page_count ?? null
+            if (doc && !this.previewDocModel) this.previewDocModel = doc
+          },
+          error: () => (this.previewPageCount = null),
+        })
+    }
+    this.pageImageUrl = this.templateService.getPageImageUrl(
+      this.previewDocId,
+      this.previewPage
+    )
+    this.imageLoaded = false
+  }
+
+  goToPage(page: number) {
+    if (!Number.isFinite(page)) return
+    const max = this.previewPageCount ? this.previewPageCount - 1 : page
+    const clamped = Math.max(0, Math.min(page, max))
+    if (clamped === this.previewPage) return
+    this.previewPage = clamped
+    this.loadPreview()
+  }
+
+  prevPage() {
+    this.goToPage(this.previewPage - 1)
+  }
+
+  nextPage() {
+    this.goToPage(this.previewPage + 1)
+  }
+
+  zoomIn() {
+    this.zoom = Math.min(4, Math.round((this.zoom + 0.25) * 100) / 100)
+  }
+
+  zoomOut() {
+    this.zoom = Math.max(0.5, Math.round((this.zoom - 0.25) * 100) / 100)
+  }
+
+  resetZoom() {
+    this.zoom = 1
+  }
+
+  zonePage(zone: OcrTemplateZone): number {
+    return getZonePage(zone, this.previewPage, this.previewPageCount)
+  }
+
+  private isOnCurrentPage(zone: OcrTemplateZone): boolean {
+    return isZoneOnPage(zone, this.previewPage, this.previewPageCount)
+  }
+
+  onImageLoad() {
+    this.imageLoaded = true
+    const img = this.imageRef.nativeElement
+    this.template.source_width = img.naturalWidth
+    this.template.source_height = img.naturalHeight
+  }
+
+  onOverlayMouseDown(event: MouseEvent) {
+    const point = this.svgPointFromEvent(event)
+    if (!point) return
+    event.preventDefault()
+
+    if (this.selectedZoneIndex !== null) {
+      const handle = this.findHandleAt(point, this.selectedZoneIndex)
+      if (handle) {
+        this.overlayInteraction = {
+          kind: 'resizing',
+          zoneIndex: this.selectedZoneIndex,
+          handle,
+        }
+        return
+      }
+    }
+
+    const clickedIdx = this.findZoneAt(point)
+    if (clickedIdx !== null && !event.shiftKey) {
+      this.selectZone(clickedIdx)
+      const zone = this.template.zones[clickedIdx]
+      this.overlayInteraction = {
+        kind: 'moving',
+        zoneIndex: clickedIdx,
+        start: {
+          mouseX: point.x,
+          mouseY: point.y,
+          zoneX: zone.x,
+          zoneY: zone.y,
+        },
+      }
+      return
+    }
+
+    // Shift+click or click on empty area starts a new zone.
+    this.overlayInteraction = {
+      kind: 'drawing',
+      rect: {
+        startX: point.x,
+        startY: point.y,
+        endX: point.x,
+        endY: point.y,
+      },
+    }
+    this.selectedZoneIndex = null
+  }
+
+  onOverlayMouseMove(event: MouseEvent) {
+    const point = this.svgPointFromEvent(event)
+    if (!point) return
+
+    if (this.overlayInteraction.kind === 'resizing') {
+      this.applyResize(
+        this.overlayInteraction.zoneIndex,
+        this.overlayInteraction.handle,
+        point
+      )
+      return
+    }
+
+    if (this.overlayInteraction.kind === 'moving') {
+      moveZone(
+        this.template.zones[this.overlayInteraction.zoneIndex],
+        point,
+        this.overlayInteraction.start,
+        this.imageNaturalSize(),
+        this.imageNaturalSize()
+      )
+      return
+    }
+
+    if (this.overlayInteraction.kind === 'drawing') {
+      this.overlayInteraction.rect.endX = point.x
+      this.overlayInteraction.rect.endY = point.y
+      return
+    }
+
+    this.updateOverlayCursor(point)
+  }
+
+  private updateOverlayCursor(point: Point) {
+    if (this.selectedZoneIndex !== null) {
+      const handle = this.findHandleAt(point, this.selectedZoneIndex)
+      if (handle) {
+        this.overlayCursor = RESIZE_CURSOR[handle] || 'crosshair'
+        return
+      }
+    }
+    this.overlayCursor = this.findZoneAt(point) !== null ? 'move' : 'crosshair'
+  }
+
+  onOverlayMouseUp(_event: MouseEvent) {
+    if (
+      this.overlayInteraction.kind === 'moving' ||
+      this.overlayInteraction.kind === 'resizing'
+    ) {
+      this.stopOverlayInteraction()
+      return
+    }
+
+    if (this.overlayInteraction.kind !== 'drawing') return
+    const drawingRect = this.overlayInteraction.rect
+    this.stopOverlayInteraction()
+
+    const rect = this.sourceRectFromDrawing(drawingRect)
+
+    // Ignore tiny accidental clicks.
+    if (rect.w < MIN_DRAWN_ZONE_SIZE || rect.h < MIN_DRAWN_ZONE_SIZE) {
+      return
+    }
+
+    this.template.zones.push(this.createZoneFromRect(rect))
+    this.selectZone(this.template.zones.length - 1)
+  }
+
+  private createZoneFromRect(rect: DisplayRect): OcrTemplateZone {
+    const imageSize = this.imageNaturalSize()
+    return {
+      name: `Zone ${this.template.zones.length + 1}`,
+      target: DEFAULT_OCR_ZONE_TARGET,
+      custom_field: this.defaultCustomFieldId(),
+      x: rect.x,
+      y: rect.y,
+      width: rect.w,
+      height: rect.h,
+      page: this.previewPageDisplay,
+      ocr_language: DEFAULT_OCR_ZONE_LANGUAGE,
+      transform: DEFAULT_OCR_ZONE_TRANSFORM,
+      date_format: '',
+      validation_regex: '',
+      order: this.template.zones.length,
+      zone_source_width: imageSize.width,
+      zone_source_height: imageSize.height,
+    }
+  }
+
+  private defaultCustomFieldId(): number | null {
+    return this.customFields[0]?.id ?? null
+  }
+
+  @HostListener('document:mouseup')
+  onDocumentMouseUp() {
+    if (this.overlayInteraction.kind === 'idle') return
+    this.stopOverlayInteraction()
+  }
+
+  private stopOverlayInteraction() {
+    this.overlayInteraction = NO_OVERLAY_INTERACTION
+    this.overlayCursor = 'crosshair'
+  }
+
+  drawingRect(): DisplayRect | null {
+    return this.overlayInteraction.kind === 'drawing'
+      ? this.displayRectFromDrawing(this.overlayInteraction.rect)
+      : null
+  }
+
+  zoneDisplayRect(zoneIdx: number): DisplayRect | null {
+    const img = this.imageRef?.nativeElement
+    if (!img || !img.naturalWidth) return null
+    const zone = this.template.zones[zoneIdx]
+    if (!zone) return null
+    if (!this.isOnCurrentPage(zone)) return null
+    return getZoneDisplayRect(
+      zone,
+      this.imageNaturalSize(),
+      this.imageNaturalSize()
+    )
+  }
+
+  private findHandleAt(point: Point, zoneIdx: number): ResizeHandle | null {
+    const r = this.zoneDisplayRect(zoneIdx)
+    if (!r) return null
+    return findHandleAt(point, r, this.overlayHandleSize())
+  }
+
+  private applyResize(zoneIndex: number, handle: ResizeHandle, point: Point) {
+    const zone = this.template.zones[zoneIndex]
+    if (!zone) return
+    resizeZone(
+      zone,
+      handle,
+      point,
+      this.imageNaturalSize(),
+      this.imageNaturalSize()
+    )
+  }
+
+  private findZoneAt(point: Point): number | null {
+    const img = this.imageRef.nativeElement
+    if (!img.naturalWidth) return null
+
+    return findZoneAt(
+      point,
+      this.template.zones,
+      this.previewPage,
+      this.previewPageCount,
+      this.imageNaturalSize(),
+      this.imageNaturalSize()
+    )
+  }
+
+  overlayViewBox(): string {
+    const imageSize = this.imageNaturalSize()
+    return `0 0 ${imageSize.width} ${imageSize.height}`
+  }
+
+  zoneColor(index: number): string {
+    return ZONE_COLORS[index % ZONE_COLORS.length]
+  }
+
+  zoneFill(index: number): string {
+    return `${this.zoneColor(index)}33`
+  }
+
+  zoneLabel(zone: OcrTemplateZone, index: number): string {
+    return zone.name || `Zone ${index + 1}`
+  }
+
+  zoneLabelY(rect: DisplayRect): number {
+    return Math.max(this.overlayUnitSize(14), rect.y - this.overlayUnitSize(4))
+  }
+
+  resizeHandles(rect: DisplayRect): ResizeHandleMarker[] {
+    return [
+      { handle: 'nw', x: rect.x, y: rect.y },
+      { handle: 'n', x: rect.x + rect.w / 2, y: rect.y },
+      { handle: 'ne', x: rect.x + rect.w, y: rect.y },
+      { handle: 'w', x: rect.x, y: rect.y + rect.h / 2 },
+      { handle: 'e', x: rect.x + rect.w, y: rect.y + rect.h / 2 },
+      { handle: 'sw', x: rect.x, y: rect.y + rect.h },
+      { handle: 's', x: rect.x + rect.w / 2, y: rect.y + rect.h },
+      { handle: 'se', x: rect.x + rect.w, y: rect.y + rect.h },
+    ]
+  }
+
+  overlayHandleSize(): number {
+    return this.overlayUnitSize(HANDLE_SIZE)
+  }
+
+  overlayFontSize(): number {
+    return this.overlayUnitSize(12)
+  }
+
+  overlayUnitSize(screenPixels: number): number {
+    const img = this.imageRef?.nativeElement
+    if (!img?.naturalWidth || !img.clientWidth) return screenPixels
+    return (screenPixels * img.naturalWidth) / img.clientWidth
+  }
+
+  private svgPointFromEvent(event: MouseEvent): Point | null {
+    const svg = this.overlayRef?.nativeElement
+    const matrix = svg?.getScreenCTM()
+    if (!svg || !matrix) return null
+
+    const point = svg.createSVGPoint()
+    point.x = event.clientX
+    point.y = event.clientY
+
+    const svgPoint = point.matrixTransform(matrix.inverse())
+    return { x: svgPoint.x, y: svgPoint.y }
+  }
+
+  private displayRectFromDrawing(rect: DrawingRect): DisplayRect {
+    return {
+      x: Math.min(rect.startX, rect.endX),
+      y: Math.min(rect.startY, rect.endY),
+      w: Math.abs(rect.endX - rect.startX),
+      h: Math.abs(rect.endY - rect.startY),
+    }
+  }
+
+  private sourceRectFromDrawing(rect: DrawingRect): DisplayRect {
+    const displayRect = this.displayRectFromDrawing(rect)
+    return {
+      x: Math.round(displayRect.x),
+      y: Math.round(displayRect.y),
+      w: Math.round(displayRect.w),
+      h: Math.round(displayRect.h),
+    }
+  }
+
+  private imageNaturalSize() {
+    const img = this.imageRef.nativeElement
+    return { width: img.naturalWidth, height: img.naturalHeight }
+  }
+
+  removeZone(index: number) {
+    this.template.zones.splice(index, 1)
+    if (this.selectedZoneIndex === index) {
+      this.selectedZoneIndex = null
+    } else if (this.selectedZoneIndex > index) {
+      this.selectedZoneIndex--
+    }
+  }
+
+  selectZone(index: number) {
+    this.selectedZoneIndex = index
+    this.activeTab = 'zone'
+    this.zoneTestResult = null
+    const zone = this.template.zones[index]
+    if (zone) {
+      this.seedCombineDefault(zone)
+      this.goToPage(this.zonePage(zone) - 1)
+    }
+  }
+
+  testZone() {
+    const zone = this.selectedZone
+    if (!zone || !this.previewDocId) return
+    this.zoneTesting = true
+    this.zoneTestResult = null
+    this.templateService
+      .testZone(this.previewDocId, this.zoneTestRequest(zone))
+      .pipe(takeUntil(this.destroy$))
+      .subscribe({
+        next: (res) => {
+          this.zoneTestResult = res
+          this.zoneTesting = false
+        },
+        error: (err) => {
+          this.zoneTestResult = {
+            error: err.error?.error || $localize`Test failed`,
+          }
+          this.zoneTesting = false
+        },
+      })
+  }
+
+  private zoneTestRequest(zone: OcrTemplateZone): ZoneTestRequest {
+    return {
+      name: zone.name,
+      x: zone.x,
+      y: zone.y,
+      width: zone.width,
+      height: zone.height,
+      page: zone.page ?? 1,
+      ocr_language: zone.ocr_language,
+      transform: zone.transform,
+      date_format: zone.date_format,
+      validation_regex: zone.validation_regex,
+      zone_source_width: zone.zone_source_width,
+      zone_source_height: zone.zone_source_height,
+    }
+  }
+
+  deleteSelectedZone() {
+    if (this.selectedZoneIndex === null) return
+    this.removeZone(this.selectedZoneIndex)
+    this.activeTab = 'zones'
+  }
+
+  save() {
+    this.saving = true
+    this.pruneCombineFormats()
+    this.template.sample_document = this.previewDocId
+    const obs = this.isNew
+      ? this.templateService.create(this.template)
+      : this.templateService.update(this.template)
+
+    obs.pipe(takeUntil(this.destroy$)).subscribe({
+      next: (saved) => {
+        const idx = this.selectedZoneIndex
+        this.template = saved
+        this.isNew = false
+        this.selectedZoneIndex = idx
+        this.saving = false
+        this.toastService.showInfo($localize`OCR template saved.`)
+      },
+      error: (e) => {
+        this.saving = false
+        this.toastService.showError($localize`Error saving OCR template.`, e)
+      },
+    })
+  }
+
+  private ocrLangCache = new WeakMap<
+    OcrTemplateZone,
+    { src: string; arr: string[] }
+  >()
+
+  ocrLanguageArray(zone: OcrTemplateZone): string[] {
+    const src = zone.ocr_language || ''
+    const cached = this.ocrLangCache.get(zone)
+    if (cached && cached.src === src) return cached.arr
+    const arr = src ? src.split('+').filter(Boolean) : []
+    this.ocrLangCache.set(zone, { src, arr })
+    return arr
+  }
+
+  setOcrLanguages(zone: OcrTemplateZone, langs: string[]) {
+    zone.ocr_language = (langs || []).join('+')
+    this.ocrLangCache.set(zone, {
+      src: zone.ocr_language,
+      arr: langs ? [...langs] : [],
+    })
+  }
+
+  getCustomFieldName(id: number): string {
+    const cf = this.customFields.find((f) => f.id === id)
+    return cf ? cf.name : `Field #${id}`
+  }
+
+  /** Value bound to the field select: a built-in id string or a custom-field id. */
+  zoneFieldValue(zone: OcrTemplateZone): ZoneFieldSelection {
+    const target = zone.target || DEFAULT_OCR_ZONE_TARGET
+    return target === OCR_ZONE_TARGET.CustomField ? zone.custom_field : target
+  }
+
+  setZoneField(zone: OcrTemplateZone, value: ZoneFieldSelection) {
+    if (isOcrBuiltinTarget(value)) {
+      zone.target = value
+      zone.custom_field = null
+    } else {
+      zone.target = OCR_ZONE_TARGET.CustomField
+      zone.custom_field = typeof value === 'number' ? value : null
+    }
+    this.seedCombineDefault(zone)
+  }
+
+  fieldKeyFor(zone: OcrTemplateZone): string | null {
+    const v = this.zoneFieldValue(zone)
+    return v === null || v === undefined ? null : String(v)
+  }
+
+  zonesForField(zone: OcrTemplateZone): OcrTemplateZone[] {
+    const key = this.fieldKeyFor(zone)
+    if (!key) return []
+    return this.template.zones.filter((z) => this.fieldKeyFor(z) === key)
+  }
+
+  isFieldShared(zone: OcrTemplateZone): boolean {
+    return this.zonesForField(zone).length > 1
+  }
+
+  getCombineFormat(zone: OcrTemplateZone): string {
+    const key = this.fieldKeyFor(zone)
+    return (key && this.template.combine_formats?.[key]) || ''
+  }
+
+  setCombineFormat(zone: OcrTemplateZone, value: string) {
+    const key = this.fieldKeyFor(zone)
+    if (!key) return
+    this.template.combine_formats ??= {}
+    this.template.combine_formats[key] = value
+  }
+
+  insertCombineToken(zone: OcrTemplateZone, tokenZone: OcrTemplateZone) {
+    const token = `{${tokenZone.name}}`
+    const current = this.getCombineFormat(zone)
+    const sep = current && !current.endsWith(' ') ? ' ' : ''
+    this.setCombineFormat(zone, `${current}${sep}${token}`)
+  }
+
+  private seedCombineDefault(zone: OcrTemplateZone) {
+    const key = this.fieldKeyFor(zone)
+    if (!key) return
+    const shared = this.zonesForField(zone)
+    if (shared.length <= 1) return
+    this.template.combine_formats ??= {}
+    if (!this.template.combine_formats[key]) {
+      this.template.combine_formats[key] = shared
+        .map((z) => `{${z.name}}`)
+        .join(' ')
+    }
+  }
+
+  private pruneCombineFormats() {
+    const formats = this.template.combine_formats
+    if (!formats) return
+    const counts = new Map<string, number>()
+    for (const z of this.template.zones) {
+      const key = this.fieldKeyFor(z)
+      if (key) counts.set(key, (counts.get(key) ?? 0) + 1)
+    }
+    for (const key of Object.keys(formats)) {
+      if ((counts.get(key) ?? 0) <= 1) delete formats[key]
+    }
+  }
+
+  /** Value bound to the date-format select: a preset, '' (auto), or 'custom'. */
+  dateFormatChoice(zone: OcrTemplateZone): string {
+    return this.usesCustomDateFormat(zone)
+      ? CUSTOM_DATE_FORMAT_CHOICE
+      : zone.date_format || ''
+  }
+
+  setDateFormatChoice(zone: OcrTemplateZone, value: string) {
+    if (value === CUSTOM_DATE_FORMAT_CHOICE) {
+      this.customDateFormatZones.add(zone)
+      zone.date_format ||= ''
+    } else {
+      this.customDateFormatZones.delete(zone)
+      zone.date_format = value
+    }
+  }
+
+  usesCustomDateFormat(zone: OcrTemplateZone): boolean {
+    return (
+      this.customDateFormatZones.has(zone) ||
+      (!!zone.date_format &&
+        !this.dateFormatOptions.some(
+          (option) => option.id === zone.date_format
+        ))
+    )
+  }
+
+  getZoneTargetName(zone: OcrTemplateZone): string {
+    const target = zone.target || DEFAULT_OCR_ZONE_TARGET
+    if (target === OCR_ZONE_TARGET.CustomField) {
+      return zone.custom_field
+        ? this.getCustomFieldName(zone.custom_field)
+        : $localize`(no field)`
+    }
+    return this.builtinTargets.find((t) => t.id === target)?.name ?? target
+  }
+
+  getDocumentTypeName(id: number): string {
+    const dt = this.documentTypes.find((d) => d.id === id)
+    return dt ? dt.name : `Type #${id}`
+  }
+
+  openQuickCreate(zoneIndex: number | null) {
+    if (zoneIndex === null) return
+    this.quickCreateForZoneIndex = zoneIndex
+    this.quickCreateName = this.template.zones[zoneIndex]?.name || ''
+    this.quickCreateType = CustomFieldDataType.String
+    this.showQuickCreate = true
+  }
+
+  cancelQuickCreate() {
+    this.showQuickCreate = false
+    this.quickCreateForZoneIndex = null
+  }
+
+  submitQuickCreate() {
+    if (!this.quickCreateName.trim()) return
+
+    this.templateService
+      .quickCreateField(this.quickCreateName.trim(), this.quickCreateType)
+      .pipe(takeUntil(this.destroy$))
+      .subscribe({
+        next: (result) => {
+          this.customFieldsService.clearCache()
+          this.customFieldsService
+            .listAll()
+            .pipe(takeUntil(this.destroy$))
+            .subscribe((r) => {
+              this.customFields = r.results
+              if (this.quickCreateForZoneIndex !== null) {
+                this.template.zones[this.quickCreateForZoneIndex].custom_field =
+                  result.id
+                this.template.zones[this.quickCreateForZoneIndex].target =
+                  OCR_ZONE_TARGET.CustomField
+              }
+              this.showQuickCreate = false
+              this.quickCreateForZoneIndex = null
+            })
+        },
+        error: (err) => {
+          this.toastService.showError(
+            $localize`Failed to create custom field.`,
+            err
+          )
+        },
+      })
+  }
+
+  ngOnDestroy() {
+    this.destroy$.next()
+    this.destroy$.complete()
+  }
+}

src-ui/src/app/components/manage/ocr-templates/ocr-template-editor/zone-geometry.spec.ts

@@ -0,0 +1,140 @@
+import { OcrTemplateZone } from 'src/app/data/ocr-template'
+import {
+  findHandleAt,
+  findZoneAt,
+  getZoneDisplayRect,
+  getZonePage,
+  isZoneOnPage,
+  moveZone,
+  resizeZone,
+  sourceRectFromDrawing,
+} from './zone-geometry'
+
+function zone(overrides: Partial<OcrTemplateZone> = {}): OcrTemplateZone {
+  return {
+    name: 'Zone',
+    target: 'custom_field',
+    custom_field: 1,
+    x: 100,
+    y: 200,
+    width: 300,
+    height: 400,
+    page: 1,
+    ocr_language: 'eng',
+    transform: 'strip',
+    validation_regex: '',
+    order: 0,
+    ...overrides,
+  }
+}
+
+describe('OCR template editor geometry', () => {
+  it('normalizes zone pages', () => {
+    expect(getZonePage(zone({ page: 2 }), 0, 5)).toBe(2)
+    expect(getZonePage(zone({ page: -1 }), 0, 5)).toBe(5)
+    expect(getZonePage(zone({ page: -1 }), 2, null)).toBe(3)
+    expect(getZonePage(zone({ page: 0 }), 0, 5)).toBe(1)
+    expect(getZonePage(zone({ page: undefined }), 0, 5)).toBe(1)
+  })
+
+  it('checks whether a zone is on the current preview page', () => {
+    expect(isZoneOnPage(zone({ page: 2 }), 1, 5)).toBe(true)
+    expect(isZoneOnPage(zone({ page: 2 }), 0, 5)).toBe(false)
+    expect(isZoneOnPage(zone({ page: -1 }), 4, 5)).toBe(true)
+  })
+
+  it('scales source coordinates to canvas display coordinates', () => {
+    expect(
+      getZoneDisplayRect(
+        zone({ x: 100, y: 200, width: 300, height: 400 }),
+        { width: 500, height: 1000 },
+        { width: 1000, height: 2000 }
+      )
+    ).toEqual({ x: 50, y: 100, w: 150, h: 200 })
+  })
+
+  it('uses per-zone source dimensions when present', () => {
+    expect(
+      getZoneDisplayRect(
+        zone({
+          x: 100,
+          y: 100,
+          width: 100,
+          height: 100,
+          zone_source_width: 1000,
+          zone_source_height: 1000,
+        }),
+        { width: 500, height: 500 },
+        { width: 2000, height: 2000 }
+      )
+    ).toEqual({ x: 50, y: 50, w: 50, h: 50 })
+  })
+
+  it('finds zones from topmost to bottommost on the current page', () => {
+    const zones = [
+      zone({ name: 'first', x: 0, y: 0, width: 100, height: 100, page: 1 }),
+      zone({ name: 'second', x: 0, y: 0, width: 50, height: 50, page: 1 }),
+      zone({ name: 'third', x: 0, y: 0, width: 50, height: 50, page: 2 }),
+    ]
+
+    expect(
+      findZoneAt(
+        { x: 25, y: 25 },
+        zones,
+        0,
+        2,
+        { width: 100, height: 100 },
+        { width: 100, height: 100 }
+      )
+    ).toBe(1)
+  })
+
+  it('finds resize handles around a display rect', () => {
+    const rect = { x: 10, y: 20, w: 100, h: 200 }
+
+    expect(findHandleAt({ x: 10, y: 20 }, rect)).toBe('nw')
+    expect(findHandleAt({ x: 110, y: 220 }, rect)).toBe('se')
+    expect(findHandleAt({ x: 60, y: 20 }, rect)).toBe('n')
+    expect(findHandleAt({ x: 90, y: 160 }, rect)).toBeNull()
+  })
+
+  it('moves zones without leaving source image bounds', () => {
+    const z = zone({ x: 50, y: 50, width: 100, height: 100 })
+
+    moveZone(
+      z,
+      { x: 500, y: 500 },
+      { mouseX: 50, mouseY: 50, zoneX: 50, zoneY: 50 },
+      { width: 500, height: 500 },
+      { width: 500, height: 500 }
+    )
+
+    expect(z.x).toBe(400)
+    expect(z.y).toBe(400)
+  })
+
+  it('resizes zones without leaving source image bounds', () => {
+    const z = zone({ x: 50, y: 50, width: 100, height: 100 })
+
+    resizeZone(
+      z,
+      'se',
+      { x: 500, y: 500 },
+      { width: 500, height: 500 },
+      { width: 200, height: 200 }
+    )
+
+    expect(z.width).toBe(150)
+    expect(z.height).toBe(150)
+  })
+
+  it('converts drawn canvas rectangles to source rectangles', () => {
+    expect(
+      sourceRectFromDrawing(
+        { startX: 100, startY: 200, endX: 50, endY: 100 },
+        { width: 500, height: 1000 },
+        { width: 1000, height: 2000 }
+      )
+    ).toEqual({ x: 100, y: 200, w: 100, h: 200 })
+  })
+})

src-ui/src/app/components/manage/ocr-templates/ocr-template-editor/zone-geometry.ts

@@ -0,0 +1,201 @@
+import { OcrTemplateZone } from 'src/app/data/ocr-template'
+
+export interface DrawingRect {
+  startX: number
+  startY: number
+  endX: number
+  endY: number
+}
+
+export interface Dimensions {
+  width: number
+  height: number
+}
+
+export interface Point {
+  x: number
+  y: number
+}
+
+export interface DisplayRect {
+  x: number
+  y: number
+  w: number
+  h: number
+}
+
+export interface MoveStart {
+  mouseX: number
+  mouseY: number
+  zoneX: number
+  zoneY: number
+}
+
+export type ResizeHandle = 'n' | 's' | 'e' | 'w' | 'ne' | 'nw' | 'se' | 'sw'
+
+export const HANDLE_SIZE = 8
+export const MIN_ZONE_SIZE = 10
+
+export function getZonePage(
+  zone: OcrTemplateZone,
+  previewPage: number,
+  previewPageCount: number | null
+): number {
+  const page = zone.page ?? 1
+  if (page === -1) return previewPageCount ?? previewPage + 1
+  return page >= 1 ? page : 1
+}
+
+export function isZoneOnPage(
+  zone: OcrTemplateZone,
+  previewPage: number,
+  previewPageCount: number | null
+): boolean {
+  return getZonePage(zone, previewPage, previewPageCount) === previewPage + 1
+}
+
+export function getZoneSourceSize(
+  zone: OcrTemplateZone,
+  imageSize: Dimensions
+): Dimensions {
+  return {
+    width: zone.zone_source_width || imageSize.width,
+    height: zone.zone_source_height || imageSize.height,
+  }
+}
+
+export function getZoneDisplayRect(
+  zone: OcrTemplateZone,
+  canvasSize: Dimensions,
+  imageSize: Dimensions
+): DisplayRect {
+  const sourceSize = getZoneSourceSize(zone, imageSize)
+  const scaleX = canvasSize.width / sourceSize.width
+  const scaleY = canvasSize.height / sourceSize.height
+
+  return {
+    x: zone.x * scaleX,
+    y: zone.y * scaleY,
+    w: zone.width * scaleX,
+    h: zone.height * scaleY,
+  }
+}
+
+export function findHandleAt(
+  point: Point,
+  rect: DisplayRect,
+  handleSize = HANDLE_SIZE
+): ResizeHandle | null {
+  const handles: [ResizeHandle, number, number][] = [
+    ['nw', rect.x, rect.y],
+    ['n', rect.x + rect.w / 2, rect.y],
+    ['ne', rect.x + rect.w, rect.y],
+    ['w', rect.x, rect.y + rect.h / 2],
+    ['e', rect.x + rect.w, rect.y + rect.h / 2],
+    ['sw', rect.x, rect.y + rect.h],
+    ['s', rect.x + rect.w / 2, rect.y + rect.h],
+    ['se', rect.x + rect.w, rect.y + rect.h],
+  ]
+
+  return (
+    handles.find(
+      ([, x, y]) =>
+        Math.abs(point.x - x) <= handleSize &&
+        Math.abs(point.y - y) <= handleSize
+    )?.[0] ?? null
+  )
+}
+
+export function findZoneAt(
+  point: Point,
+  zones: OcrTemplateZone[],
+  previewPage: number,
+  previewPageCount: number | null,
+  canvasSize: Dimensions,
+  imageSize: Dimensions
+): number | null {
+  for (let i = zones.length - 1; i >= 0; i--) {
+    const zone = zones[i]
+    if (!isZoneOnPage(zone, previewPage, previewPageCount)) continue
+    const rect = getZoneDisplayRect(zone, canvasSize, imageSize)
+
+    if (
+      point.x >= rect.x &&
+      point.x <= rect.x + rect.w &&
+      point.y >= rect.y &&
+      point.y <= rect.y + rect.h
+    ) {
+      return i
+    }
+  }
+
+  return null
+}
+
+export function moveZone(
+  zone: OcrTemplateZone,
+  point: Point,
+  moveStart: MoveStart,
+  canvasSize: Dimensions,
+  imageSize: Dimensions
+) {
+  const sourceSize = getZoneSourceSize(zone, imageSize)
+  const scaleX = sourceSize.width / canvasSize.width
+  const scaleY = sourceSize.height / canvasSize.height
+  const dx = Math.round((point.x - moveStart.mouseX) * scaleX)
+  const dy = Math.round((point.y - moveStart.mouseY) * scaleY)
+
+  zone.x = clamp(moveStart.zoneX + dx, 0, sourceSize.width - zone.width)
+  zone.y = clamp(moveStart.zoneY + dy, 0, sourceSize.height - zone.height)
+}
+
+export function resizeZone(
+  zone: OcrTemplateZone,
+  handle: ResizeHandle,
+  point: Point,
+  canvasSize: Dimensions,
+  imageSize: Dimensions
+) {
+  const sourceSize = getZoneSourceSize(zone, imageSize)
+  const scaleX = sourceSize.width / canvasSize.width
+  const scaleY = sourceSize.height / canvasSize.height
+  const imageX = clamp(Math.round(point.x * scaleX), 0, sourceSize.width)
+  const imageY = clamp(Math.round(point.y * scaleY), 0, sourceSize.height)
+
+  if (handle.includes('w')) {
+    const right = Math.min(zone.x + zone.width, sourceSize.width)
+    zone.x = clamp(imageX, 0, right - MIN_ZONE_SIZE)
+    zone.width = right - zone.x
+  }
+  if (handle.includes('e')) {
+    zone.width = Math.max(MIN_ZONE_SIZE, imageX - zone.x)
+  }
+  if (handle.includes('n')) {
+    const bottom = Math.min(zone.y + zone.height, sourceSize.height)
+    zone.y = clamp(imageY, 0, bottom - MIN_ZONE_SIZE)
+    zone.height = bottom - zone.y
+  }
+  if (handle.includes('s')) {
+    zone.height = Math.max(MIN_ZONE_SIZE, imageY - zone.y)
+  }
+}
+
+export function sourceRectFromDrawing(
+  rect: DrawingRect,
+  canvasSize: Dimensions,
+  imageSize: Dimensions
+): DisplayRect {
+  const scaleX = imageSize.width / canvasSize.width
+  const scaleY = imageSize.height / canvasSize.height
+
+  return {
+    x: Math.round(Math.min(rect.startX, rect.endX) * scaleX),
+    y: Math.round(Math.min(rect.startY, rect.endY) * scaleY),
+    w: Math.round(Math.abs(rect.endX - rect.startX) * scaleX),
+    h: Math.round(Math.abs(rect.endY - rect.startY) * scaleY),
+  }
+}
+
+function clamp(value: number, min: number, max: number): number {
+  return Math.max(min, Math.min(value, max))
+}

src-ui/src/app/components/manage/ocr-templates/ocr-templates.component.html

@@ -0,0 +1,75 @@
+<pngx-page-header
+  title="OCR Templates"
+  i18n-title
+  info="Define extraction zones on document types to automatically populate custom fields via OCR."
+  i18n-info
+  >
+  <button type="button" class="btn btn-sm btn-outline-primary" (click)="createTemplate()" *pngxIfPermissions="{ action: PermissionAction.Add, type: PermissionType.OcrTemplate }">
+    <i-bs name="plus-circle" class="me-1"></i-bs><ng-container i18n>Create Template</ng-container>
+  </button>
+</pngx-page-header>
+
+<ul class="list-group">
+
+  <li class="list-group-item">
+    <div class="row">
+      <div class="col" i18n>Name</div>
+      <div class="col d-none d-sm-flex" i18n>Document Type</div>
+      <div class="col d-none d-sm-flex" i18n>Zones</div>
+      <div class="col" i18n>Status</div>
+      <div class="col" i18n>Actions</div>
+    </div>
+  </li>
+
+  @if (loading && templates.length === 0) {
+    <li class="list-group-item">
+      <div class="spinner-border spinner-border-sm me-2" role="status"></div>
+      <ng-container i18n>Loading...</ng-container>
+    </li>
+  }
+
+  @for (t of templates; track t.id) {
+    <li class="list-group-item">
+      <div class="row fade" [class.show]="show">
+        <div class="col d-flex align-items-center"><button class="btn btn-link p-0 text-start" type="button" (click)="editTemplate(t)" [disabled]="!permissionsService.currentUserCan(PermissionAction.Change, PermissionType.OcrTemplate)">{{t.name}}</button></div>
+        <div class="col d-flex align-items-center d-none d-sm-flex">{{getDocumentTypeName(t)}}</div>
+        <div class="col d-flex align-items-center d-none d-sm-flex"><code>{{t.zones?.length || 0}}</code></div>
+        <div class="col d-flex align-items-center">
+          <div class="form-check form-switch mb-0">
+            <input type="checkbox" class="form-check-input cursor-pointer" [id]="t.id+'_enable'" [(ngModel)]="t.enabled" (change)="toggleTemplate(t)" *pngxIfPermissions="{ action: PermissionAction.Change, type: PermissionType.OcrTemplate }">
+            <label class="form-check-label cursor-pointer" [for]="t.id+'_enable'">
+              <code> @if(t.enabled) { <ng-container i18n>Enabled</ng-container> } @else { <span i18n class="text-muted">Disabled</span> }</code>
+            </label>
+          </div>
+        </div>
+        <div class="col">
+
+          <div class="btn-group d-block d-sm-none">
+            <div ngbDropdown container="body" class="d-inline-block">
+              <button type="button" class="btn btn-link" id="actionsMenuMobile{{t.id}}" (click)="$event.stopPropagation()" ngbDropdownToggle>
+                <i-bs name="three-dots-vertical"></i-bs>
+              </button>
+              <div ngbDropdownMenu aria-labelledby="actionsMenuMobile{{t.id}}">
+                <button (click)="editTemplate(t)" *pngxIfPermissions="{ action: PermissionAction.Change, type: PermissionType.OcrTemplate }" ngbDropdownItem i18n>Edit</button>
+                <button (click)="deleteTemplate(t)" *pngxIfPermissions="{ action: PermissionAction.Delete, type: PermissionType.OcrTemplate }" ngbDropdownItem i18n>Delete</button>
+              </div>
+            </div>
+          </div>
+          <div class="btn-toolbar d-none d-sm-flex gap-2" role="toolbar">
+            <div class="btn-group">
+              <button *pngxIfPermissions="{ action: PermissionAction.Change, type: PermissionType.OcrTemplate }" class="btn btn-sm btn-outline-secondary" type="button" (click)="editTemplate(t)">
+                <i-bs width="1em" height="1em" name="pencil" class="me-1"></i-bs><ng-container i18n>Edit</ng-container>
+              </button>
+              <button *pngxIfPermissions="{ action: PermissionAction.Delete, type: PermissionType.OcrTemplate }" class="btn btn-sm btn-outline-danger" type="button" (click)="deleteTemplate(t)">
+                <i-bs width="1em" height="1em" name="trash" class="me-1"></i-bs><ng-container i18n>Delete</ng-container>
+              </button>
+            </div>
+          </div>
+        </div>
+      </div>
+    </li>
+  }
+  @if (!loading && templates.length === 0) {
+    <li class="list-group-item" [class.show]="show" i18n>No OCR templates defined.</li>
+  }
+</ul>

src-ui/src/app/components/manage/ocr-templates/ocr-templates.component.ts

@@ -0,0 +1,109 @@
+import { Component, OnInit, inject } from '@angular/core'
+import { FormsModule } from '@angular/forms'
+import { Router } from '@angular/router'
+import { NgbDropdownModule, NgbModal } from '@ng-bootstrap/ng-bootstrap'
+import { NgxBootstrapIconsModule } from 'ngx-bootstrap-icons'
+import { delay, takeUntil, tap } from 'rxjs'
+import { OcrTemplate } from 'src/app/data/ocr-template'
+import { IfPermissionsDirective } from 'src/app/directives/if-permissions.directive'
+import { PermissionsService } from 'src/app/services/permissions.service'
+import { DocumentTypeService } from 'src/app/services/rest/document-type.service'
+import { OcrTemplateService } from 'src/app/services/rest/ocr-template.service'
+import { ToastService } from 'src/app/services/toast.service'
+import { ConfirmDialogComponent } from '../../common/confirm-dialog/confirm-dialog.component'
+import { PageHeaderComponent } from '../../common/page-header/page-header.component'
+import { LoadingComponentWithPermissions } from '../../loading-component/loading.component'
+
+@Component({
+  selector: 'pngx-ocr-templates',
+  templateUrl: './ocr-templates.component.html',
+  imports: [
+    PageHeaderComponent,
+    IfPermissionsDirective,
+    FormsModule,
+    NgbDropdownModule,
+    NgxBootstrapIconsModule,
+  ],
+})
+export class OcrTemplatesComponent
+  extends LoadingComponentWithPermissions
+  implements OnInit
+{
+  private readonly service = inject(OcrTemplateService)
+  private readonly documentTypeService = inject(DocumentTypeService)
+  private readonly router = inject(Router)
+  private readonly modalService = inject(NgbModal)
+  private readonly toastService = inject(ToastService)
+  permissionsService = inject(PermissionsService)
+
+  public templates: OcrTemplate[] = []
+  private documentTypeNames: Map<number, string> = new Map()
+
+  ngOnInit() {
+    this.documentTypeService
+      .listAll()
+      .pipe(takeUntil(this.unsubscribeNotifier))
+      .subscribe((r) => {
+        this.documentTypeNames = new Map(
+          r.results.map((dt) => [dt.id, dt.name])
+        )
+      })
+    this.reload()
+  }
+
+  reload() {
+    this.loading = true
+    this.service
+      .listAll()
+      .pipe(
+        takeUntil(this.unsubscribeNotifier),
+        tap((r) => (this.templates = r.results)),
+        delay(100)
+      )
+      .subscribe(() => {
+        this.show = true
+        this.loading = false
+      })
+  }
+
+  getDocumentTypeName(t: OcrTemplate): string {
+    return (
+      this.documentTypeNames.get(t.document_type) ?? `${t.document_type ?? ''}`
+    )
+  }
+
+  createTemplate() {
+    this.router.navigate(['/ocr-templates', 'new'])
+  }
+
+  editTemplate(t: OcrTemplate) {
+    this.router.navigate(['/ocr-templates', t.id])
+  }
+
+  toggleTemplate(t: OcrTemplate) {
+    // ngModel has already flipped t.enabled; restore it if persistence fails.
+    const enabled = t.enabled
+    this.service.patch(t).subscribe({
+      error: (error) => {
+        t.enabled = !enabled
+        this.toastService.showError(
+          $localize`Error updating OCR template.`,
+          error
+        )
+      },
+    })
+  }
+
+  deleteTemplate(t: OcrTemplate) {
+    const modal = this.modalService.open(ConfirmDialogComponent)
+    modal.componentInstance.title = $localize`Delete OCR Template`
+    modal.componentInstance.messageBoldPart = t.name
+    modal.componentInstance.message = $localize`Do you really want to delete this OCR template?`
+    modal.componentInstance.btnClass = 'btn-danger'
+    modal.componentInstance.btnCaption = $localize`Delete`
+    modal.componentInstance.confirmClicked.subscribe(() => {
+      modal.close()
+      this.service.delete(t).subscribe(() => this.reload())
+    })
+  }
+}

src-ui/src/app/data/ocr-template.ts

@@ -0,0 +1,142 @@
+import { ObjectWithId } from './object-with-id'
+
+export type OcrZoneTarget = 'custom_field' | 'title' | 'asn' | 'created'
+export type OcrBuiltinTarget = Exclude<OcrZoneTarget, 'custom_field'>
+export type OcrZoneTransform =
+  | 'none'
+  | 'strip'
+  | 'uppercase'
+  | 'lowercase'
+  | 'numeric'
+  | 'strip_punctuation'
+  | 'date'
+  | 'qr_code'
+
+export const OCR_ZONE_TARGET = {
+  CustomField: 'custom_field',
+  Title: 'title',
+  Asn: 'asn',
+  Created: 'created',
+} as const satisfies Record<string, OcrZoneTarget>
+
+export const OCR_ZONE_TRANSFORM = {
+  None: 'none',
+  Strip: 'strip',
+  Uppercase: 'uppercase',
+  Lowercase: 'lowercase',
+  Numeric: 'numeric',
+  StripPunctuation: 'strip_punctuation',
+  Date: 'date',
+  QrCode: 'qr_code',
+} as const satisfies Record<string, OcrZoneTransform>
+
+export const DEFAULT_OCR_ZONE_TARGET = OCR_ZONE_TARGET.CustomField
+export const DEFAULT_OCR_ZONE_TRANSFORM = OCR_ZONE_TRANSFORM.Strip
+export const DEFAULT_OCR_ZONE_LANGUAGE = 'deu+eng'
+
+export function isOcrBuiltinTarget(value: unknown): value is OcrBuiltinTarget {
+  return (
+    value === OCR_ZONE_TARGET.Title ||
+    value === OCR_ZONE_TARGET.Asn ||
+    value === OCR_ZONE_TARGET.Created
+  )
+}
+
+export const OCR_BUILTIN_TARGETS = [
+  { id: OCR_ZONE_TARGET.Title, name: $localize`Title` },
+  { id: OCR_ZONE_TARGET.Asn, name: $localize`Archive serial number` },
+  { id: OCR_ZONE_TARGET.Created, name: $localize`Date created` },
+]
+
+export interface OcrTemplateZone {
+  id?: number
+  name: string
+  target?: OcrZoneTarget
+  custom_field: number | null
+  page?: number
+  x: number
+  y: number
+  width: number
+  height: number
+  ocr_language: string
+  transform: OcrZoneTransform
+  date_format?: string
+  validation_regex: string
+  order: number
+  zone_source_width?: number
+  zone_source_height?: number
+}
+
+export const TRANSFORM_OPTIONS = [
+  { id: OCR_ZONE_TRANSFORM.None, name: $localize`None` },
+  { id: OCR_ZONE_TRANSFORM.Strip, name: $localize`Strip whitespace` },
+  { id: OCR_ZONE_TRANSFORM.Uppercase, name: $localize`Uppercase` },
+  { id: OCR_ZONE_TRANSFORM.Lowercase, name: $localize`Lowercase` },
+  { id: OCR_ZONE_TRANSFORM.Numeric, name: $localize`Numeric only` },
+  {
+    id: OCR_ZONE_TRANSFORM.StripPunctuation,
+    name: $localize`Remove leading/trailing punctuation`,
+  },
+  { id: OCR_ZONE_TRANSFORM.Date, name: $localize`Parse date` },
+  { id: OCR_ZONE_TRANSFORM.QrCode, name: $localize`Read QR/barcode` },
+]
+
+export const OCR_LANGUAGE_OPTIONS = [
+  { id: 'eng', name: $localize`English` },
+  { id: 'deu', name: $localize`German` },
+  { id: 'fra', name: $localize`French` },
+  { id: 'ita', name: $localize`Italian` },
+  { id: 'spa', name: $localize`Spanish` },
+  { id: 'por', name: $localize`Portuguese` },
+  { id: 'nld', name: $localize`Dutch` },
+]
+
+export const DATE_FORMAT_OPTIONS = [
+  { id: '', name: $localize`Auto-detect` },
+  { id: '%d.%m.%Y', name: 'DD.MM.YYYY' },
+  { id: '%Y/%m/%d', name: 'YYYY/MM/DD' },
+  { id: '%d/%m/%Y', name: 'DD/MM/YYYY' },
+]
+
+export interface OcrTemplate extends ObjectWithId {
+  name: string
+  document_type: number
+  sample_document: number | null
+  source_width: number
+  source_height: number
+  enabled: boolean
+  combine_formats?: Record<string, string>
+  created?: string
+  updated?: string
+  zones: OcrTemplateZone[]
+}
+
+export interface ZoneTestRequest {
+  name: string
+  x: number
+  y: number
+  width: number
+  height: number
+  page: number
+  ocr_language: string
+  transform: OcrZoneTransform
+  date_format?: string
+  validation_regex: string
+  zone_source_width?: number
+  zone_source_height?: number
+}
+
+export interface OcrZoneTestResult {
+  raw_text?: string | null
+  value?: string | null
+  regex?: string
+  regex_match?: boolean | null
+  error?: string
+}
+
+export interface OcrZoneRunResult {
+  template: string
+  zone: string
+  custom_field: string
+  value: string | number | null
+}

src-ui/src/app/data/ui-settings.ts

@@ -24,16 +24,6 @@ export enum CollapsibleSection {
   ATTRIBUTES = 'attributes',
 }
 
-export enum HideableSidebarItemID {
-  Dashboard = 'dashboard',
-  SavedViews = 'saved_views',
-  Workflows = 'workflows',
-  Mail = 'mail',
-  Documentation = 'documentation',
-}
-
-export const HIDEABLE_SIDEBAR_ITEM_IDS = Object.values(HideableSidebarItemID)
-
 export const PAPERLESS_GREEN_HEX = '#17541f'
 
 export const SETTINGS_KEYS = {
@@ -66,7 +56,6 @@ export const SETTINGS_KEYS = {
   NOTES_ENABLED: 'general-settings:notes-enabled',
   AUDITLOG_ENABLED: 'general-settings:auditlog-enabled',
   SLIM_SIDEBAR: 'general-settings:slim-sidebar',
-  SIDEBAR_HIDDEN_ITEMS: 'general-settings:sidebar:hidden-items',
   ATTRIBUTES_SECTIONS_COLLAPSED:
     'general-settings:attributes-sections-collapsed',
   UPDATE_CHECKING_ENABLED: 'general-settings:update-checking:enabled',
@@ -138,11 +127,6 @@ export const SETTINGS: UiSetting[] = [
     type: 'boolean',
     default: false,
   },
-  {
-    key: SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS,
-    type: 'array',
-    default: [],
-  },
   {
     key: SETTINGS_KEYS.ATTRIBUTES_SECTIONS_COLLAPSED,
     type: 'array',

src-ui/src/app/services/permissions.service.ts

@@ -29,6 +29,7 @@ export enum PermissionType {
   ShareLinkBundle = '%s_sharelinkbundle',
   CustomField = '%s_customfield',
   Workflow = '%s_workflow',
+  OcrTemplate = '%s_ocrtemplate',
   ProcessedMail = '%s_processedmail',
   GlobalStatistics = '%s_global_statistics',
   SystemMonitoring = '%s_system_monitoring',

src-ui/src/app/services/rest/document.service.ts

@@ -12,6 +12,7 @@ import {
 import { DocumentMetadata } from 'src/app/data/document-metadata'
 import { DocumentSuggestions } from 'src/app/data/document-suggestions'
 import { FilterRule } from 'src/app/data/filter-rule'
+import { OcrZoneRunResult } from 'src/app/data/ocr-template'
 import { Results, SelectionData } from 'src/app/data/results'
 import { SETTINGS_KEYS } from 'src/app/data/ui-settings'
 import { queryParamsFromFilterRules } from '../../utils/query-params'
@@ -359,6 +360,13 @@ export class DocumentService extends AbstractPaperlessService<Document> {
     })
   }
 
+  runZoneOcr(id: number): Observable<{ results: OcrZoneRunResult[] }> {
+    return this.http.post<{ results: OcrZoneRunResult[] }>(
+      this.getResourceUrl(id, 'run-zone-ocr'),
+      {}
+    )
+  }
+
   rotateDocuments(
     selection: DocumentSelectionQuery,
     degrees: number,

src-ui/src/app/services/rest/ocr-template.service.ts

@@ -0,0 +1,47 @@
+import { Injectable } from '@angular/core'
+import { Observable } from 'rxjs'
+import {
+  OcrTemplate,
+  OcrZoneTestResult,
+  ZoneTestRequest,
+} from '../../data/ocr-template'
+import { AbstractPaperlessService } from './abstract-paperless-service'
+
+export interface QuickCreateFieldResult {
+  id: number
+  name: string
+  data_type: string
+  created: boolean
+}
+
+@Injectable({ providedIn: 'root' })
+export class OcrTemplateService extends AbstractPaperlessService<OcrTemplate> {
+  constructor() {
+    super()
+    this.resourceName = 'ocr_templates'
+  }
+
+  getPageImageUrl(docId: number, page: number): string {
+    return `${this.baseUrl}${this.resourceName}/document-page-image/${docId}/${page}/`
+  }
+
+  testZone(
+    docId: number,
+    zone: ZoneTestRequest
+  ): Observable<OcrZoneTestResult> {
+    return this.http.post<OcrZoneTestResult>(
+      `${this.baseUrl}${this.resourceName}/test-zone/`,
+      { document: docId, zone }
+    )
+  }
+
+  quickCreateField(
+    name: string,
+    dataType: string
+  ): Observable<QuickCreateFieldResult> {
+    return this.http.post<QuickCreateFieldResult>(
+      `${this.baseUrl}${this.resourceName}/quick-create-field/`,
+      { name, data_type: dataType }
+    )
+  }
+}

src-ui/src/app/services/settings.service.spec.ts

@@ -14,11 +14,7 @@ import { CustomFieldDataType } from '../data/custom-field'
 import { DEFAULT_DISPLAY_FIELDS, DisplayField } from '../data/document'
 import { SavedView } from '../data/saved-view'
 import { RemoteOCRModeConfig } from '../data/paperless-config'
-import {
-  HideableSidebarItemID,
-  SETTINGS_KEYS,
-  UiSettings,
-} from '../data/ui-settings'
+import { SETTINGS_KEYS, UiSettings } from '../data/ui-settings'
 import { PermissionsService } from './permissions.service'
 import { CustomFieldsService } from './rest/custom-fields.service'
 import { SettingsService } from './settings.service'
@@ -234,35 +230,6 @@ describe('SettingsService', () => {
     expect(notesEnabled()).toBeFalsy()
   })
 
-  it('updates sidebar item visibility', () => {
-    httpTestingController
-      .expectOne(`${environment.apiBaseUrl}ui_settings/`)
-      .flush(ui_settings)
-
-    expect(
-      settingsService.sidebarItemIsHidden(HideableSidebarItemID.Workflows)
-    ).toBe(false)
-
-    settingsService.updateSidebarItemVisibility(
-      HideableSidebarItemID.Workflows,
-      false
-    )
-
-    expect(
-      settingsService.sidebarItemIsHidden(HideableSidebarItemID.Workflows)
-    ).toBe(true)
-    expect(settingsService.get(SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS)).toEqual([])
-
-    settingsService.updateSidebarItemVisibility(
-      HideableSidebarItemID.Workflows,
-      true
-    )
-
-    expect(
-      settingsService.sidebarItemIsHidden(HideableSidebarItemID.Workflows)
-    ).toBe(false)
-  })
-
   it('updates setting signals when settings are reinitialized', () => {
     let req = httpTestingController.expectOne(
       `${environment.apiBaseUrl}ui_settings/`

src-ui/src/app/services/settings.service.ts

@@ -24,7 +24,6 @@ import { DEFAULT_DISPLAY_FIELDS, DisplayField } from '../data/document'
 import { RemoteOCRModeConfig } from '../data/paperless-config'
 import { SavedView } from '../data/saved-view'
 import {
-  HideableSidebarItemID,
   PAPERLESS_GREEN_HEX,
   SETTINGS,
   SETTINGS_KEYS,
@@ -314,18 +313,6 @@ export class SettingsService {
   readonly globalDropzoneEnabled = signal(true)
   readonly globalDropzoneActive = signal(false)
   readonly organizingSidebarSavedViews = signal(false)
-  readonly sidebarHiddenItemsEditing = signal<HideableSidebarItemID[] | null>(
-    null
-  )
-  readonly organizingSidebarItems = computed(
-    () => this.sidebarHiddenItemsEditing() !== null
-  )
-  readonly sidebarHiddenItemsEditingChanged = new EventEmitter<
-    HideableSidebarItemID[]
-  >()
-  readonly hiddenSidebarItems = this.getSignal<HideableSidebarItemID[]>(
-    SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS
-  )
 
   readonly allDisplayFields = signal<Array<{ id: DisplayField; name: string }>>(
     DEFAULT_DISPLAY_FIELDS
@@ -762,29 +749,6 @@ export class SettingsService {
     return this.storeSettings()
   }
 
-  sidebarItemIsHidden(item: HideableSidebarItemID): boolean {
-    return (
-      this.sidebarHiddenItemsEditing() ?? this.hiddenSidebarItems()
-    ).includes(item)
-  }
-
-  updateSidebarItemVisibility(
-    item: HideableSidebarItemID,
-    visible: boolean
-  ): void {
-    const hiddenItems = new Set(
-      this.sidebarHiddenItemsEditing() ?? this.hiddenSidebarItems()
-    )
-    if (visible) {
-      hiddenItems.delete(item)
-    } else {
-      hiddenItems.add(item)
-    }
-    const updatedHiddenItems = [...hiddenItems]
-    this.sidebarHiddenItemsEditing.set(updatedHiddenItems)
-    this.sidebarHiddenItemsEditingChanged.emit(updatedHiddenItems)
-  }
-
   updateSavedViewsVisibility(
     dashboardVisibleViewIds: number[],
     sidebarVisibleViewIds: number[]

src-ui/src/main.ts

@@ -89,6 +89,7 @@ import {
   exclamationTriangleFill,
   eye,
   fileEarmark,
+  fileEarmarkBreak,
   fileEarmarkCheck,
   fileEarmarkDiff,
   fileEarmarkFill,
@@ -99,6 +100,7 @@ import {
   fileEarmarkPlus,
   fileEarmarkRichtext,
   fileEarmarkSpreadsheet,
+  fileEarmarkRuled,
   fileText,
   files,
   filter,
@@ -338,6 +340,7 @@ const icons = {
   exclamationTriangleFill,
   eye,
   fileEarmark,
+  fileEarmarkBreak,
   fileEarmarkCheck,
   fileEarmarkDiff,
   fileEarmarkFill,
@@ -348,6 +351,7 @@ const icons = {
   fileEarmarkPlus,
   fileEarmarkRichtext,
   fileEarmarkSpreadsheet,
+  fileEarmarkRuled,
   files,
   fileText,
   filter,

src/documents/apps.py

@@ -13,8 +13,11 @@ class DocumentsConfig(AppConfig):
         from documents.signals.handlers import add_inbox_tags
         from documents.signals.handlers import add_or_update_document_in_llm_index
         from documents.signals.handlers import add_to_index
+        from documents.signals.handlers import capture_old_document_type
         from documents.signals.handlers import run_workflows_added
         from documents.signals.handlers import run_workflows_updated
+        from documents.signals.handlers import run_zone_ocr_extraction
+        from documents.signals.handlers import run_zone_ocr_on_type_change
         from documents.signals.handlers import send_websocket_document_updated
         from documents.signals.handlers import set_correspondent
         from documents.signals.handlers import set_document_type
@@ -28,8 +31,17 @@ class DocumentsConfig(AppConfig):
         document_consumption_finished.connect(set_storage_path)
         document_consumption_finished.connect(add_to_index)
         document_consumption_finished.connect(run_workflows_added)
-        document_consumption_finished.connect(add_to_index)
         document_consumption_finished.connect(add_or_update_document_in_llm_index)
+        document_consumption_finished.connect(run_zone_ocr_extraction)
+
+        from django.db.models.signals import post_save
+        from django.db.models.signals import pre_save
+
+        from documents.models import Document
+
+        pre_save.connect(capture_old_document_type, sender=Document)
+        post_save.connect(run_zone_ocr_on_type_change, sender=Document)
+
         document_updated.connect(run_workflows_updated)
         document_updated.connect(send_websocket_document_updated)
         document_updated.connect(add_or_update_document_in_llm_index)

src/documents/bulk_edit.py

@@ -28,7 +28,7 @@ from documents.models import DocumentType
 from documents.models import PaperlessTask
 from documents.models import StoragePath
 from documents.models import Tag
-from documents.permissions import set_permissions_for_objects
+from documents.permissions import set_permissions_for_object
 from documents.plugins.helpers import DocumentsStatusManager
 from documents.tasks import bulk_update_documents
 from documents.tasks import consume_file
@@ -299,55 +299,53 @@ def modify_custom_fields(
 ) -> Literal["OK"]:
     qs = Document.objects.filter(id__in=doc_ids).only("pk")
     affected_docs = list(qs.values_list("pk", flat=True))
-    # Ensure add_custom_fields is a list of (int, value) tuples, supports old API
+    # Ensure add_custom_fields is a list of tuples, supports old API
     add_custom_fields = (
-        [(int(field), value) for field, value in add_custom_fields.items()]
+        add_custom_fields.items()
         if isinstance(add_custom_fields, dict)
-        else [(int(field), None) for field in add_custom_fields]
+        else [(field, None) for field in add_custom_fields]
     )
 
-    # Resolved once, instead of re-querying the same field for every document
-    custom_fields_by_id: dict[int, CustomField] = CustomField.objects.in_bulk(
-        [field_id for field_id, _ in add_custom_fields],
-    )
-    # Passed to update_or_create() below rather than a bare id, so the FK is
-    # cached on the created instance and auditlog's post_save receiver does
-    # not reload it per row. Only needed for additions. content is deferred:
-    # the one field here that is both large and unused.
-    docs_by_id: dict[int, Document] = (
-        Document.objects.defer("content").in_bulk(affected_docs)
-        if add_custom_fields
-        else {}
-    )
+    custom_fields = CustomField.objects.filter(
+        id__in=[int(field) for field, _ in add_custom_fields],
+    ).distinct()
     for field_id, value in add_custom_fields:
-        custom_field = custom_fields_by_id[field_id]
-        value_field = CustomFieldInstance.TYPE_TO_DATA_STORE_NAME_MAP[
-            custom_field.data_type
-        ]
-        is_doclink = custom_field.data_type == CustomField.FieldDataType.DOCUMENTLINK
         for doc_id in affected_docs:
-            if is_doclink and value and doc_id in value:
-                # Prevent self-linking
-                continue
+            defaults = {}
+            custom_field = custom_fields.get(id=field_id)
+            if custom_field:
+                value_field = CustomFieldInstance.TYPE_TO_DATA_STORE_NAME_MAP[
+                    custom_field.data_type
+                ]
+                defaults[value_field] = value
+                if (
+                    custom_field.data_type == CustomField.FieldDataType.DOCUMENTLINK
+                    and value
+                    and doc_id in value
+                ):
+                    # Prevent self-linking
+                    continue
             CustomFieldInstance.objects.update_or_create(
-                document=docs_by_id[doc_id],
-                field=custom_field,
-                defaults={value_field: value},
+                document_id=doc_id,
+                field_id=field_id,
+                defaults=defaults,
             )
-            if is_doclink:
-                reflect_doclinks(docs_by_id[doc_id], custom_field, value)
+            if custom_field.data_type == CustomField.FieldDataType.DOCUMENTLINK:
+                doc = Document.objects.get(id=doc_id)
+                reflect_doclinks(doc, custom_field, value)
 
-    # For doc link fields that are being removed, remove symmetrical links.
-    # select_related avoids a per-instance reload of the document and field.
+    # For doc link fields that are being removed, remove symmetrical links
     for doclink_being_removed_instance in CustomFieldInstance.objects.filter(
         document_id__in=affected_docs,
         field__id__in=remove_custom_fields,
         field__data_type=CustomField.FieldDataType.DOCUMENTLINK,
         value_document_ids__isnull=False,
-    ).select_related("field", "document"):
+    ):
         for target_doc_id in doclink_being_removed_instance.value:
             remove_doclink(
-                document=doclink_being_removed_instance.document,
+                document=Document.objects.get(
+                    id=doclink_being_removed_instance.document.id,
+                ),
                 field=doclink_being_removed_instance.field,
                 target_doc_id=target_doc_id,
             )
@@ -433,13 +431,10 @@ def set_permissions(
     else:
         qs.update(owner=owner)
 
+    for doc in qs:
+        set_permissions_for_object(permissions=set_permissions, object=doc, merge=merge)
+
     affected_docs = list(qs.values_list("pk", flat=True))
-    set_permissions_for_objects(
-        permissions=set_permissions,
-        model=Document,
-        pks=affected_docs,
-        merge=merge,
-    )
 
     bulk_update_documents.apply_async(
         kwargs={"document_ids": affected_docs},
@@ -1183,13 +1178,10 @@ def remove_doclink(
     """
     Removes a 'symmetrical' link to `document` from the target document's existing custom field instance
     """
-    # select_related: a signal receiver (auditlog) touches .document/.field on
-    # the save() below, without this that is a per-call reload query
-    target_doc_field_instance = (
-        CustomFieldInstance.objects.filter(document_id=target_doc_id, field=field)
-        .select_related("document", "field")
-        .first()
-    )
+    target_doc_field_instance = CustomFieldInstance.objects.filter(
+        document_id=target_doc_id,
+        field=field,
+    ).first()
     if (
         target_doc_field_instance is not None
         and document.id in target_doc_field_instance.value

src/documents/classifier.py

@@ -34,27 +34,6 @@ from paperless.signed_pickle import signed_pickle_loads
 
 logger = logging.getLogger("paperless.classifier")
 
-
-def _predict_with_threshold(classifier, X, threshold: float) -> int | None:
-    """
-    Return the predicted class id, or None if:
-    - the prediction is -1 (no match), or
-    - the winning class probability is below the configured threshold.
-
-    Using predict_proba() instead of predict() lets us apply a minimum-confidence
-    cutoff so that uncertain predictions are discarded rather than assigned.
-    """
-    probas = classifier.predict_proba(X)[0]
-    best_idx = int(probas.argmax())
-    best_class = int(classifier.classes_[best_idx])
-
-    if best_class == -1:
-        return None
-    if threshold > 0.0 and probas[best_idx] < threshold:
-        return None
-    return best_class
-
-
 ADVANCED_TEXT_PROCESSING_ENABLED = (
     settings.NLTK_LANGUAGE is not None and settings.NLTK_ENABLED
 )
@@ -123,8 +102,7 @@ class DocumentClassifier:
     # v8 - Added storage path classifier
     # v9 - Changed from hashing to time/ids for re-train check
     # v10 - HMAC-signed model file
-    # v11 - Use sample_weight for balanced training; predict_proba with threshold
-    FORMAT_VERSION = 11
+    FORMAT_VERSION = 10
 
     HMAC_SIZE = 32  # SHA-256 digest length
 
@@ -346,13 +324,6 @@ class DocumentClassifier:
         from sklearn.preprocessing import LabelBinarizer
         from sklearn.preprocessing import MultiLabelBinarizer
 
-        # MLPClassifier does not support class_weight directly
-        # (https://github.com/scikit-learn/scikit-learn/issues/9113), so we use
-        # compute_sample_weight to balance classes during training and prevent
-        # over-represented correspondents from dominating predictions.
-        # https://scikit-learn.org/stable/modules/generated/sklearn.utils.class_weight.compute_sample_weight.html
-        from sklearn.utils.class_weight import compute_sample_weight
-
         # Step 2: vectorize data
         logger.debug("Vectorizing data...")
         notify("Vectorizing document content...")
@@ -398,7 +369,7 @@ class DocumentClassifier:
                 self.tags_binarizer = MultiLabelBinarizer()
                 labels_tags_vectorized = self.tags_binarizer.fit_transform(labels_tags)
 
-            self.tags_classifier = MLPClassifier(tol=0.01, random_state=0)
+            self.tags_classifier = MLPClassifier(tol=0.01)
             self.tags_classifier.fit(data_vectorized, labels_tags_vectorized)
         else:
             self.tags_classifier = None
@@ -409,12 +380,8 @@ class DocumentClassifier:
             notify(
                 f"Training correspondent classifier ({num_correspondents} correspondent(s))...",
             )
-            self.correspondent_classifier = MLPClassifier(tol=0.01, random_state=0)
-            self.correspondent_classifier.fit(
-                data_vectorized,
-                labels_correspondent,
-                sample_weight=compute_sample_weight("balanced", labels_correspondent),
-            )
+            self.correspondent_classifier = MLPClassifier(tol=0.01)
+            self.correspondent_classifier.fit(data_vectorized, labels_correspondent)
         else:
             self.correspondent_classifier = None
             logger.debug(
@@ -426,12 +393,8 @@ class DocumentClassifier:
             notify(
                 f"Training document type classifier ({num_document_types} type(s))...",
             )
-            self.document_type_classifier = MLPClassifier(tol=0.01, random_state=0)
-            self.document_type_classifier.fit(
-                data_vectorized,
-                labels_document_type,
-                sample_weight=compute_sample_weight("balanced", labels_document_type),
-            )
+            self.document_type_classifier = MLPClassifier(tol=0.01)
+            self.document_type_classifier.fit(data_vectorized, labels_document_type)
         else:
             self.document_type_classifier = None
             logger.debug(
@@ -443,11 +406,10 @@ class DocumentClassifier:
                 "Training storage paths classifier...",
             )
             notify(f"Training storage path classifier ({num_storage_paths} path(s))...")
-            self.storage_path_classifier = MLPClassifier(tol=0.01, random_state=0)
+            self.storage_path_classifier = MLPClassifier(tol=0.01)
             self.storage_path_classifier.fit(
                 data_vectorized,
                 labels_storage_path,
-                sample_weight=compute_sample_weight("balanced", labels_storage_path),
             )
         else:
             self.storage_path_classifier = None
@@ -584,24 +546,24 @@ class DocumentClassifier:
     def predict_correspondent(self, content: str) -> int | None:
         if self.correspondent_classifier:
             X = self._vectorize(content)
-            predicted_id = _predict_with_threshold(
-                self.correspondent_classifier,
-                X,
-                settings.CLASSIFIER_MATCH_THRESHOLD,
-            )
-            return predicted_id
-        return None
+            correspondent_id = self.correspondent_classifier.predict(X)
+            if correspondent_id != -1:
+                return correspondent_id
+            else:
+                return None
+        else:
+            return None
 
     def predict_document_type(self, content: str) -> int | None:
         if self.document_type_classifier:
             X = self._vectorize(content)
-            predicted_id = _predict_with_threshold(
-                self.document_type_classifier,
-                X,
-                settings.CLASSIFIER_MATCH_THRESHOLD,
-            )
-            return predicted_id
-        return None
+            document_type_id = self.document_type_classifier.predict(X)
+            if document_type_id != -1:
+                return document_type_id
+            else:
+                return None
+        else:
+            return None
 
     def predict_tags(self, content: str) -> list[int]:
         from sklearn.utils.multiclass import type_of_target
@@ -627,10 +589,10 @@ class DocumentClassifier:
     def predict_storage_path(self, content: str) -> int | None:
         if self.storage_path_classifier:
             X = self._vectorize(content)
-            predicted_id = _predict_with_threshold(
-                self.storage_path_classifier,
-                X,
-                settings.CLASSIFIER_MATCH_THRESHOLD,
-            )
-            return predicted_id
-        return None
+            storage_path_id = self.storage_path_classifier.predict(X)
+            if storage_path_id != -1:
+                return storage_path_id
+            else:
+                return None
+        else:
+            return None

src/documents/migrations/0022_ocrtemplate_ocrtemplatezone.py

@@ -0,0 +1,267 @@
+# Generated by Django 5.2.14 on 2026-06-16 17:36
+
+import django.core.validators
+import django.db.models.deletion
+import django.utils.timezone
+from django.db import migrations
+from django.db import models
+
+
+class Migration(migrations.Migration):
+    dependencies = [
+        ("documents", "0021_widen_workflow_integer_fields"),
+    ]
+
+    operations = [
+        migrations.CreateModel(
+            name="OcrTemplate",
+            fields=[
+                (
+                    "id",
+                    models.AutoField(
+                        auto_created=True,
+                        primary_key=True,
+                        serialize=False,
+                        verbose_name="ID",
+                    ),
+                ),
+                ("name", models.CharField(max_length=128, verbose_name="name")),
+                (
+                    "source_width",
+                    models.PositiveIntegerField(
+                        help_text="Width of the image the zones were drawn on (px)",
+                        validators=[django.core.validators.MinValueValidator(1)],
+                        verbose_name="source width",
+                    ),
+                ),
+                (
+                    "source_height",
+                    models.PositiveIntegerField(
+                        help_text="Height of the image the zones were drawn on (px)",
+                        validators=[django.core.validators.MinValueValidator(1)],
+                        verbose_name="source height",
+                    ),
+                ),
+                ("enabled", models.BooleanField(default=True, verbose_name="enabled")),
+                (
+                    "combine_formats",
+                    models.JSONField(
+                        blank=True,
+                        default=dict,
+                        help_text="Per-target format strings for combining several zones into one field, keyed by target (custom field id, or 'title'/'asn'/'created'). Tokens like {Zone Name} are replaced with that zone's value.",
+                        verbose_name="combine formats",
+                    ),
+                ),
+                (
+                    "created",
+                    models.DateTimeField(
+                        db_index=True,
+                        default=django.utils.timezone.now,
+                        editable=False,
+                        verbose_name="created",
+                    ),
+                ),
+                (
+                    "updated",
+                    models.DateTimeField(auto_now=True, verbose_name="updated"),
+                ),
+                (
+                    "document_type",
+                    models.ForeignKey(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="ocr_templates",
+                        to="documents.documenttype",
+                        verbose_name="document type",
+                    ),
+                ),
+                (
+                    "sample_document",
+                    models.ForeignKey(
+                        blank=True,
+                        help_text="Document used for previewing zones in the editor",
+                        null=True,
+                        on_delete=django.db.models.deletion.SET_NULL,
+                        related_name="+",
+                        to="documents.document",
+                        verbose_name="sample document",
+                    ),
+                ),
+            ],
+            options={
+                "verbose_name": "OCR template",
+                "verbose_name_plural": "OCR templates",
+                "ordering": ("name",),
+            },
+        ),
+        migrations.CreateModel(
+            name="OcrTemplateZone",
+            fields=[
+                (
+                    "id",
+                    models.AutoField(
+                        auto_created=True,
+                        primary_key=True,
+                        serialize=False,
+                        verbose_name="ID",
+                    ),
+                ),
+                (
+                    "name",
+                    models.CharField(
+                        help_text="Descriptive name for this zone (e.g. 'Invoice Number')",
+                        max_length=128,
+                        verbose_name="zone name",
+                    ),
+                ),
+                (
+                    "target",
+                    models.CharField(
+                        choices=[
+                            ("custom_field", "Custom field"),
+                            ("title", "Title"),
+                            ("asn", "Archive serial number"),
+                            ("created", "Date created"),
+                        ],
+                        default="custom_field",
+                        help_text="Where the extracted value is written: a custom field, or a built-in document field (title, ASN, created date)",
+                        max_length=20,
+                        verbose_name="target",
+                    ),
+                ),
+                (
+                    "page",
+                    models.IntegerField(
+                        blank=True,
+                        help_text="Page (1 = first, -1 = last; blank uses the template default)",
+                        null=True,
+                        verbose_name="page",
+                    ),
+                ),
+                (
+                    "x",
+                    models.PositiveIntegerField(
+                        help_text="Left edge (px)",
+                        verbose_name="x",
+                    ),
+                ),
+                (
+                    "y",
+                    models.PositiveIntegerField(
+                        help_text="Top edge (px)",
+                        verbose_name="y",
+                    ),
+                ),
+                (
+                    "width",
+                    models.PositiveIntegerField(
+                        help_text="Zone width (px)",
+                        validators=[django.core.validators.MinValueValidator(1)],
+                        verbose_name="width",
+                    ),
+                ),
+                (
+                    "height",
+                    models.PositiveIntegerField(
+                        help_text="Zone height (px)",
+                        validators=[django.core.validators.MinValueValidator(1)],
+                        verbose_name="height",
+                    ),
+                ),
+                (
+                    "zone_source_width",
+                    models.PositiveIntegerField(
+                        blank=True,
+                        help_text="Width of the page image this zone was drawn on (px). Falls back to template source_width if unset.",
+                        null=True,
+                        verbose_name="zone source width",
+                    ),
+                ),
+                (
+                    "zone_source_height",
+                    models.PositiveIntegerField(
+                        blank=True,
+                        help_text="Height of the page image this zone was drawn on (px). Falls back to template source_height if unset.",
+                        null=True,
+                        verbose_name="zone source height",
+                    ),
+                ),
+                (
+                    "ocr_language",
+                    models.CharField(
+                        default="deu+eng",
+                        help_text="Tesseract language code(s), e.g. 'deu+eng'",
+                        max_length=20,
+                        verbose_name="OCR language",
+                    ),
+                ),
+                (
+                    "transform",
+                    models.CharField(
+                        choices=[
+                            ("none", "None"),
+                            ("strip", "Strip whitespace"),
+                            ("uppercase", "Uppercase"),
+                            ("lowercase", "Lowercase"),
+                            ("numeric", "Numeric only"),
+                            (
+                                "strip_punctuation",
+                                "Remove leading/trailing punctuation",
+                            ),
+                            ("date", "Parse date"),
+                            ("qr_code", "Read QR/barcode"),
+                        ],
+                        default="strip",
+                        max_length=20,
+                        verbose_name="transform",
+                    ),
+                ),
+                (
+                    "date_format",
+                    models.CharField(
+                        blank=True,
+                        default="",
+                        help_text="Python strptime format for the 'Parse date' transform (e.g. %d.%m.%Y). Blank = auto-detect.",
+                        max_length=64,
+                        verbose_name="date format",
+                    ),
+                ),
+                (
+                    "validation_regex",
+                    models.CharField(
+                        blank=True,
+                        default="",
+                        help_text="Optional regex pattern — extracted text is only accepted if it matches",
+                        max_length=256,
+                        verbose_name="validation regex",
+                    ),
+                ),
+                ("order", models.PositiveIntegerField(default=0, verbose_name="order")),
+                (
+                    "custom_field",
+                    models.ForeignKey(
+                        blank=True,
+                        help_text="Target custom field (only used when target is 'custom_field')",
+                        null=True,
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="ocr_zones",
+                        to="documents.customfield",
+                        verbose_name="custom field",
+                    ),
+                ),
+                (
+                    "template",
+                    models.ForeignKey(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="zones",
+                        to="documents.ocrtemplate",
+                        verbose_name="template",
+                    ),
+                ),
+            ],
+            options={
+                "verbose_name": "OCR template zone",
+                "verbose_name_plural": "OCR template zones",
+                "ordering": ("template", "order"),
+            },
+        ),
+    ]

src/documents/models.py

@@ -375,7 +375,6 @@ class Document(SoftDeleteModel, ModelWithOwner):  # type: ignore[django-manager-
         If the queryset already annotated ``effective_content``, that value is used.
         """
         # Here to avoid circular import
-        from documents.versioning import LATEST_VERSION_CONTENT_PREFETCH_ATTR
         from documents.versioning import sort_versions_newest_first
         from documents.versioning import versions_newest_first
 
@@ -385,19 +384,6 @@ class Document(SoftDeleteModel, ModelWithOwner):  # type: ignore[django-manager-
         if self.root_document_id is not None or self.pk is None:
             return self.content
 
-        latest_version_prefetch = getattr(
-            self,
-            LATEST_VERSION_CONTENT_PREFETCH_ATTR,
-            None,
-        )
-        if latest_version_prefetch is not None:
-            # Empty list means prefetch ran and found no versions — use own content.
-            return (
-                latest_version_prefetch[0].content
-                if latest_version_prefetch
-                else self.content
-            )
-
         prefetched_cache = getattr(self, "_prefetched_objects_cache", None)
         prefetched_versions = (
             prefetched_cache.get("versions")
@@ -2047,3 +2033,248 @@ class WorkflowRun(SoftDeleteModel):
 
     def __str__(self) -> str:
         return f"WorkflowRun of {self.workflow} at {self.run_at} on {self.document}"
+
+
+class OcrTemplate(models.Model):
+    """
+    Defines a set of OCR extraction zones for a specific document type.
+
+    When a document of that type is consumed, each zone in the template is
+    cropped from the document image and OCR'd separately. The extracted text
+    is written to the configured custom field or built-in document field.
+    """
+
+    name = models.CharField(
+        _("name"),
+        max_length=128,
+    )
+
+    document_type = models.ForeignKey(
+        "documents.DocumentType",
+        on_delete=models.CASCADE,
+        related_name="ocr_templates",
+        verbose_name=_("document type"),
+        db_index=True,
+    )
+
+    source_width = models.PositiveIntegerField(
+        _("source width"),
+        validators=[MinValueValidator(1)],
+        help_text=_("Width of the image the zones were drawn on (px)"),
+    )
+
+    source_height = models.PositiveIntegerField(
+        _("source height"),
+        validators=[MinValueValidator(1)],
+        help_text=_("Height of the image the zones were drawn on (px)"),
+    )
+
+    sample_document = models.ForeignKey(
+        "documents.Document",
+        on_delete=models.SET_NULL,
+        null=True,
+        blank=True,
+        related_name="+",
+        verbose_name=_("sample document"),
+        help_text=_("Document used for previewing zones in the editor"),
+    )
+
+    enabled = models.BooleanField(_("enabled"), default=True)
+
+    combine_formats = models.JSONField(
+        _("combine formats"),
+        default=dict,
+        blank=True,
+        help_text=_(
+            "Per-target format strings for combining several zones into one "
+            "field, keyed by target (custom field id, or 'title'/'asn'/'created'). "
+            "Tokens like {Zone Name} are replaced with that zone's value.",
+        ),
+    )
+
+    created = models.DateTimeField(
+        _("created"),
+        default=timezone.now,
+        db_index=True,
+        editable=False,
+    )
+
+    updated = models.DateTimeField(
+        _("updated"),
+        auto_now=True,
+    )
+
+    class Meta:
+        ordering = ("name",)
+        verbose_name = _("OCR template")
+        verbose_name_plural = _("OCR templates")
+
+    def __str__(self) -> str:
+        return f"{self.name} ({self.document_type})"
+
+
+class OcrTemplateZone(models.Model):
+    """
+    A rectangular region within a document page to OCR and extract into a custom
+    field or built-in document field. Coordinates are relative to the source
+    image dimensions stored on the template.
+    """
+
+    template = models.ForeignKey(
+        OcrTemplate,
+        on_delete=models.CASCADE,
+        related_name="zones",
+        verbose_name=_("template"),
+    )
+
+    name = models.CharField(
+        _("zone name"),
+        max_length=128,
+        help_text=_("Descriptive name for this zone (e.g. 'Invoice Number')"),
+    )
+
+    class TargetType(models.TextChoices):
+        CUSTOM_FIELD = ("custom_field", _("Custom field"))
+        TITLE = ("title", _("Title"))
+        ASN = ("asn", _("Archive serial number"))
+        CREATED = ("created", _("Date created"))
+
+    target = models.CharField(
+        _("target"),
+        max_length=20,
+        choices=TargetType.choices,
+        default=TargetType.CUSTOM_FIELD,
+        help_text=_(
+            "Where the extracted value is written: a custom field, or a "
+            "built-in document field (title, ASN, created date)",
+        ),
+    )
+
+    custom_field = models.ForeignKey(
+        "documents.CustomField",
+        on_delete=models.CASCADE,
+        related_name="ocr_zones",
+        verbose_name=_("custom field"),
+        null=True,
+        blank=True,
+        help_text=_("Target custom field (only used when target is 'custom_field')"),
+    )
+
+    page = models.IntegerField(
+        _("page"),
+        null=True,
+        blank=True,
+        help_text=_("Page (1 = first, -1 = last; blank uses the template default)"),
+    )
+
+    x = models.PositiveIntegerField(_("x"), help_text=_("Left edge (px)"))
+    y = models.PositiveIntegerField(_("y"), help_text=_("Top edge (px)"))
+    width = models.PositiveIntegerField(
+        _("width"),
+        validators=[MinValueValidator(1)],
+        help_text=_("Zone width (px)"),
+    )
+    height = models.PositiveIntegerField(
+        _("height"),
+        validators=[MinValueValidator(1)],
+        help_text=_("Zone height (px)"),
+    )
+
+    # Per-zone source dimensions for coordinate scaling.
+    # Stored from the page image the zone was drawn on.
+    # If null, falls back to the template's source_width/source_height.
+    # This handles PDFs with mixed page sizes (e.g. landscape + portrait,
+    # or different paper formats across pages).
+    zone_source_width = models.PositiveIntegerField(
+        _("zone source width"),
+        null=True,
+        blank=True,
+        help_text=_(
+            "Width of the page image this zone was drawn on (px). "
+            "Falls back to template source_width if unset.",
+        ),
+    )
+    zone_source_height = models.PositiveIntegerField(
+        _("zone source height"),
+        null=True,
+        blank=True,
+        help_text=_(
+            "Height of the page image this zone was drawn on (px). "
+            "Falls back to template source_height if unset.",
+        ),
+    )
+
+    ocr_language = models.CharField(
+        _("OCR language"),
+        max_length=20,
+        default="deu+eng",
+        help_text=_("Tesseract language code(s), e.g. 'deu+eng'"),
+    )
+
+    class TransformType(models.TextChoices):
+        NONE = ("none", _("None"))
+        STRIP = ("strip", _("Strip whitespace"))
+        UPPERCASE = ("uppercase", _("Uppercase"))
+        LOWERCASE = ("lowercase", _("Lowercase"))
+        NUMERIC = ("numeric", _("Numeric only"))
+        STRIP_PUNCTUATION = (
+            "strip_punctuation",
+            _("Remove leading/trailing punctuation"),
+        )
+        DATE = ("date", _("Parse date"))
+        QR_CODE = ("qr_code", _("Read QR/barcode"))
+
+    transform = models.CharField(
+        _("transform"),
+        max_length=20,
+        choices=TransformType.choices,
+        default=TransformType.STRIP,
+    )
+
+    date_format = models.CharField(
+        _("date format"),
+        max_length=64,
+        blank=True,
+        default="",
+        help_text=_(
+            "Python strptime format for the 'Parse date' transform "
+            "(e.g. %d.%m.%Y). Blank = auto-detect.",
+        ),
+    )
+
+    validation_regex = models.CharField(
+        _("validation regex"),
+        max_length=256,
+        blank=True,
+        default="",
+        help_text=_(
+            "Optional regex pattern — extracted text is only accepted if it matches",
+        ),
+    )
+
+    order = models.PositiveIntegerField(_("order"), default=0)
+
+    class Meta:
+        ordering = ("template", "order")
+        verbose_name = _("OCR template zone")
+        verbose_name_plural = _("OCR template zones")
+
+    def __str__(self) -> str:
+        return f"{self.template.name} -> {self.name}"
+
+
+# Custom field data types that zone OCR can extract into. DOCUMENTLINK and
+# SELECT are excluded (they reference other objects, not free text). Single
+# source of truth for the serializer, the quick-create endpoint and the engine.
+OCR_SUPPORTED_FIELD_TYPES = frozenset(
+    {
+        CustomField.FieldDataType.STRING,
+        CustomField.FieldDataType.URL,
+        CustomField.FieldDataType.DATE,
+        CustomField.FieldDataType.INT,
+        CustomField.FieldDataType.FLOAT,
+        CustomField.FieldDataType.MONETARY,
+        CustomField.FieldDataType.LONG_TEXT,
+        CustomField.FieldDataType.BOOL,
+    },
+)

src/documents/permissions.py

@@ -173,179 +173,6 @@ def set_permissions_for_object(
                         )
 
 
-def _resolve_permissions(codenames: set[str], ctype: ContentType) -> list[Permission]:
-    """
-    Resolves `codenames` to Permission rows, raising like the single-object
-    assign_perm() this bulk path replaces does (via a `.get()` internally)
-    if any codename doesn't exist -- e.g. a client-supplied action name that
-    was never validated (BulkEditObjectsSerializer._validate_permissions
-    calls validate_set_permissions() only for its side-effecting id checks
-    and discards the filtered dict it returns, so an unrecognized action key
-    reaches this function as-is). A plain `.filter()` with no existence
-    check would otherwise silently build zero rows and no-op instead of
-    reporting the bad input.
-    """
-    permission_objs = list(
-        Permission.objects.filter(content_type=ctype, codename__in=codenames),
-    )
-    missing = codenames - {p.codename for p in permission_objs}
-    if missing:
-        raise Permission.DoesNotExist(
-            f"Permission matching query does not exist for codename(s): "
-            f"{', '.join(sorted(missing))}",
-        )
-    return permission_objs
-
-
-def _apply_bulk_permission_entry(
-    *,
-    perm_model: type[UserObjectPermission] | type[GroupObjectPermission],
-    identity_model: type[User] | type[Group],
-    identity_field: str,
-    ids: list[int],
-    codename: str,
-    permission_objs: list[Permission],
-    ctype: ContentType,
-    object_pks: list[str],
-    merge: bool,
-) -> None:
-    # Only the ids are needed to build permission rows (via `<field>_id=`),
-    # so avoid fetching full User/Group rows for identities that may not
-    # even end up being granted anything new.
-    add_ids = set(
-        identity_model.objects.filter(id__in=ids).values_list("id", flat=True),
-    )
-
-    if not merge:
-        existing_ids = set(
-            perm_model.objects.filter(
-                content_type=ctype,
-                object_pk__in=object_pks,
-                permission__codename=codename,
-            )
-            .values_list(f"{identity_field}_id", flat=True)
-            .distinct(),
-        )
-        remove_ids = existing_ids - add_ids
-        if remove_ids:
-            perm_model.objects.filter(
-                content_type=ctype,
-                object_pk__in=object_pks,
-                permission__codename=codename,
-                **{f"{identity_field}_id__in": remove_ids},
-            ).delete()
-
-    if not add_ids:
-        return
-
-    rows = [
-        perm_model(
-            content_type=ctype,
-            object_pk=pk,
-            permission=permission_obj,
-            **{f"{identity_field}_id": identity_id},
-        )
-        for permission_obj in permission_objs
-        for pk in object_pks
-        for identity_id in add_ids
-    ]
-    # ignore_conflicts skips only rows that already exist as an exact
-    # (identity, permission, object) match -- the same de-dup the
-    # underlying (user|group, permission, object_pk) unique constraint
-    # already enforces for the single-object assign_perm() this replaces,
-    # so it doesn't change what counts as "already granted". batch_size
-    # caps how many rows go into a single INSERT statement.
-    perm_model.objects.bulk_create(rows, ignore_conflicts=True, batch_size=1000)
-
-
-def set_permissions_for_objects(
-    permissions: dict,
-    model: type[Model],
-    pks: QuerySet | list,
-    *,
-    merge: bool = False,
-) -> None:
-    """
-    Bulk equivalent of set_permissions_for_object: applies the same
-    permission changes to every object identified by `pks` at once.
-
-    Takes a model + pks (rather than model instances) deliberately -- the
-    permission rows built below only ever need `pk`, `content_type`, and
-    identity ids, so callers shouldn't have to fetch full rows (with every
-    other field) just to hand them to this function.
-
-    Deliberately does not use guardian's queryset/list-aware assign_perm:
-    passing a list as the object routes to bulk_assign_perm, which skips
-    creating a direct permission row for anyone who already has the
-    permission via ANY group membership (it checks
-    ObjectPermissionChecker.has_perm, which is group-inheritance-aware) --
-    unlike the single-object assign_perm this replaces, which always
-    ensures a direct row via get_or_create regardless of group-derived
-    access. Losing that guarantee would mean a later revocation of the
-    group's grant silently strips access an admin explicitly asked to be
-    direct. Bulk-creating rows straight against the permission models
-    instead (see _apply_bulk_permission_entry) preserves the original
-    always-create-a-direct-row semantics while still batching every object
-    and every identity into one query per action, rather than one query per
-    (object, user) pair.
-    """
-    object_pks = [str(pk) for pk in pks]
-    if not object_pks:  # pragma: no cover
-        return
-
-    model_name = model.__name__.lower()
-    ctype = ContentType.objects.get_for_model(model)
-
-    # Every action is resolved up front, before anything is written, so an
-    # unrecognized action name (see _resolve_permissions) aborts the whole
-    # call instead of leaving the actions ahead of it already applied --
-    # BulkEditObjectsSerializer lets unknown keys through and its view turns
-    # the exception into a 400, so a half-applied change would otherwise be
-    # reported to the client as a failure.
-    permissions_by_action: dict[str, list[Permission]] = {}
-    for action, entry in permissions.items():
-        if "users" not in entry and "groups" not in entry:
-            continue
-        implied_codenames = {f"{action}_{model_name}"}
-        if action == "change":
-            # change gives view too
-            implied_codenames.add(f"view_{model_name}")
-        permissions_by_action[action] = _resolve_permissions(
-            implied_codenames,
-            ctype,
-        )
-
-    for action, entry in permissions.items():
-        codename = f"{action}_{model_name}"
-        permission_objs = permissions_by_action.get(action, [])
-
-        if "users" in entry:
-            _apply_bulk_permission_entry(
-                perm_model=UserObjectPermission,
-                identity_model=User,
-                identity_field="user",
-                ids=entry["users"],
-                codename=codename,
-                permission_objs=permission_objs,
-                ctype=ctype,
-                object_pks=object_pks,
-                merge=merge,
-            )
-
-        if "groups" in entry:
-            _apply_bulk_permission_entry(
-                perm_model=GroupObjectPermission,
-                identity_model=Group,
-                identity_field="group",
-                ids=entry["groups"],
-                codename=codename,
-                permission_objs=permission_objs,
-                ctype=ctype,
-                object_pks=object_pks,
-                merge=merge,
-            )
-
-
 def permitted_object_ids(
     user: User | None,
     model: type[Model],

src/documents/serialisers.py

@@ -56,6 +56,7 @@ if settings.AUDIT_LOG_ENABLED:
 from documents import bulk_edit
 from documents.data_models import DocumentSource
 from documents.filters import CustomFieldQueryParser
+from documents.models import OCR_SUPPORTED_FIELD_TYPES
 from documents.models import Correspondent
 from documents.models import CustomField
 from documents.models import CustomFieldInstance
@@ -63,6 +64,8 @@ from documents.models import Document
 from documents.models import DocumentType
 from documents.models import MatchingModel
 from documents.models import Note
+from documents.models import OcrTemplate
+from documents.models import OcrTemplateZone
 from documents.models import PaperlessTask
 from documents.models import SavedView
 from documents.models import SavedViewFilterRule
@@ -89,7 +92,6 @@ from documents.templating.utils import convert_format_str_to_template_format
 from documents.templating.workflows import validate_workflow_template
 from documents.validators import uri_validator
 from documents.validators import url_validator
-from documents.versioning import has_prefetched_effective_content
 from documents.versioning import sort_versions_newest_first
 
 if TYPE_CHECKING:
@@ -1153,14 +1155,8 @@ class DocumentSerializer(
 
     def to_representation(self, instance):
         doc = super().to_representation(instance)
-        if "content" in self.fields and has_prefetched_effective_content(instance):
-            # Only resolve version-aware content when it's cheap: an SQL
-            # annotation or a versions prefetch is already on the instance.
-            # A caller that set up neither (e.g. TrashView, GlobalSearchView,
-            # which build their own querysets) gets the document's own,
-            # unresolved content instead of paying for an extra per-instance
-            # query -- same as before effective_content resolution existed.
-            doc["content"] = instance.get_effective_content() or ""
+        if "content" in self.fields and hasattr(instance, "effective_content"):
+            doc["content"] = getattr(instance, "effective_content") or ""
         if self.truncate_content and "content" in self.fields:
             doc["content"] = doc.get("content")[0:550]
         return doc
@@ -1254,31 +1250,30 @@ class DocumentSerializer(
 
             validated_data["tags"] = list(final_tags)
         if validated_data.get("remove_inbox_tags"):
-            current_tag_ids = {t.pk for t in instance.tags.all()}
-            tags = (
-                validated_data["tags"]
+            tag_ids_being_added = (
+                [
+                    tag.id
+                    for tag in validated_data["tags"]
+                    if tag not in instance.tags.all()
+                ]
                 if "tags" in validated_data
-                else list(instance.tags.all())
+                else []
             )
-
-            # Tags newly added in this update, plus their ancestors, are kept
-            keep_ids: set[int] = set()
-            for tag in tags:
-                if tag.pk not in current_tag_ids:
-                    keep_ids.add(tag.pk)
-                    keep_ids.update(int(pk) for pk in tag.get_ancestors_pks())
-
-            # Remove inbox tags and their descendants, except those being kept
-            remove_ids: set[int] = set()
-            for inbox_tag in (
-                Tag.objects.filter(is_inbox_tag=True)
-                .exclude(pk__in=keep_ids)
-                .only("pk", "tn_descendants_pks")
-            ):
-                remove_ids.add(inbox_tag.pk)
-                remove_ids.update(int(pk) for pk in inbox_tag.get_descendants_pks())
-
-            validated_data["tags"] = [t for t in tags if t.pk not in remove_ids]
+            inbox_tags_not_being_added = Tag.objects.filter(is_inbox_tag=True).exclude(
+                id__in=tag_ids_being_added,
+            )
+            if "tags" in validated_data:
+                validated_data["tags"] = [
+                    tag
+                    for tag in validated_data["tags"]
+                    if tag not in inbox_tags_not_being_added
+                ]
+            else:
+                validated_data["tags"] = [
+                    tag
+                    for tag in instance.tags.all()
+                    if tag not in inbox_tags_not_being_added
+                ]
 
         if settings.AUDIT_LOG_ENABLED:
             with set_actor(self.user):
@@ -3670,3 +3665,129 @@ class StoragePathTestSerializer(SerializerWithPerms):
             document_field.queryset = Document.objects.filter(
                 id__in=permitted_document_ids(user),
             )
+
+
+class OcrTemplateZoneSerializer(serializers.ModelSerializer):
+    class Meta:
+        model = OcrTemplateZone
+        fields = [
+            "id",
+            "name",
+            "target",
+            "custom_field",
+            "page",
+            "x",
+            "y",
+            "width",
+            "height",
+            "ocr_language",
+            "transform",
+            "date_format",
+            "order",
+            "zone_source_width",
+            "zone_source_height",
+            "validation_regex",
+        ]
+
+    def validate_width(self, value):
+        if value < 1:
+            raise serializers.ValidationError("Width must be at least 1.")
+        return value
+
+    def validate_height(self, value):
+        if value < 1:
+            raise serializers.ValidationError("Height must be at least 1.")
+        return value
+
+    def validate_custom_field(self, value):
+        if value is None:
+            # Built-in target (title/asn/created) — no custom field required.
+            return value
+        if value.data_type not in OCR_SUPPORTED_FIELD_TYPES:
+            raise serializers.ValidationError(
+                f"Custom field type '{value.data_type}' is not supported for OCR extraction. "
+                f"Use string, integer, float, date, monetary, boolean, URL, or long text.",
+            )
+        return value
+
+
+class OcrTemplateSerializer(serializers.ModelSerializer):
+    zones = OcrTemplateZoneSerializer(many=True, required=False)
+
+    class Meta:
+        model = OcrTemplate
+        fields = [
+            "id",
+            "name",
+            "document_type",
+            "source_width",
+            "source_height",
+            "sample_document",
+            "enabled",
+            "combine_formats",
+            "created",
+            "updated",
+            "zones",
+        ]
+        read_only_fields = ["created", "updated"]
+
+    def validate_source_width(self, value):
+        if value < 1:
+            raise serializers.ValidationError("Source width must be at least 1.")
+        return value
+
+    def validate_source_height(self, value):
+        if value < 1:
+            raise serializers.ValidationError("Source height must be at least 1.")
+        return value
+
+    def validate_zones(self, zones_data):
+        """Validate zone coordinates are within the source dimensions."""
+        # source_width/height may not be in initial_data during partial updates
+        source_width = self.initial_data.get("source_width") or (
+            self.instance.source_width if self.instance else None
+        )
+        source_height = self.initial_data.get("source_height") or (
+            self.instance.source_height if self.instance else None
+        )
+
+        if source_width and source_height:
+            for zone in zones_data:
+                x = zone.get("x", 0)
+                y = zone.get("y", 0)
+                w = zone.get("width", 0)
+                h = zone.get("height", 0)
+                if x + w > int(source_width):
+                    raise serializers.ValidationError(
+                        f"Zone '{zone.get('name', '?')}' extends beyond source width "
+                        f"({x + w} > {source_width}).",
+                    )
+                if y + h > int(source_height):
+                    raise serializers.ValidationError(
+                        f"Zone '{zone.get('name', '?')}' extends beyond source height "
+                        f"({y + h} > {source_height}).",
+                    )
+
+        return zones_data
+
+    def create(self, validated_data):
+        zones_data = validated_data.pop("zones", [])
+        template = OcrTemplate.objects.create(**validated_data)
+        for zone_data in zones_data:
+            OcrTemplateZone.objects.create(template=template, **zone_data)
+        return template
+
+    def update(self, instance, validated_data):
+        zones_data = validated_data.pop("zones", None)
+
+        for attr, value in validated_data.items():
+            setattr(instance, attr, value)
+        instance.save()
+
+        if zones_data is not None:
+            # Replace all zones with the new set
+            instance.zones.all().delete()
+            for zone_data in zones_data:
+                OcrTemplateZone.objects.create(template=instance, **zone_data)
+
+        return instance

src/documents/signals/handlers.py

@@ -1398,6 +1398,76 @@ def close_connection_pool_on_worker_init(**kwargs) -> None:
             conn.close_pool()
 
 
+def run_zone_ocr_extraction(sender, document, original_file=None, **kwargs):
+    """
+    Run zone-based OCR extraction if the document's type has an active template.
+    """
+    try:
+        from documents.zone_ocr import run_zone_extraction
+
+        run_zone_extraction(document, Path(original_file) if original_file else None)
+    except Exception:
+        logger.exception(
+            "Zone OCR extraction failed for document %s",
+            document.pk,
+        )
+
+
+def capture_old_document_type(sender, instance, **kwargs):
+    """pre_save: remember the document's previous type so the post_save handler
+    can tell whether the type actually changed (vs. every other save)."""
+    if instance.pk:
+        instance._old_document_type_id = (
+            Document.objects.filter(pk=instance.pk)
+            .values_list("document_type_id", flat=True)
+            .first()
+        )
+    else:
+        instance._old_document_type_id = None
+
+
+def run_zone_ocr_on_type_change(sender, instance, *, created=False, **kwargs):
+    """
+    Run zone OCR only when a document's TYPE actually changes (and the new type
+    has an enabled template). NOT on every save — zone OCR overwrites fields, so
+    re-running it on each edit would clobber the user's changes. Newly created
+    documents are handled by the consumption signal, and the user can always
+    trigger extraction manually via the run-zone-ocr action.
+    """
+    if created or not instance.pk or not instance.document_type_id:
+        return
+
+    # Only proceed if the type changed compared to what was in the DB before.
+    old_type = getattr(instance, "_old_document_type_id", None)
+    if old_type == instance.document_type_id:
+        return
+
+    from documents.models import OcrTemplate
+
+    if not OcrTemplate.objects.filter(
+        document_type_id=instance.document_type_id,
+        enabled=True,
+    ).exists():
+        return
+
+    try:
+        from documents.zone_ocr import run_zone_extraction
+
+        doc_path = instance.archive_path or instance.source_path
+        if doc_path and Path(doc_path).is_file():
+            logger.info(
+                "Zone OCR: running extraction for document %d (type %d)",
+                instance.pk,
+                instance.document_type_id,
+            )
+            run_zone_extraction(instance, None)
+    except Exception:
+        logger.exception(
+            "Zone OCR extraction failed for document %s",
+            instance.pk,
+        )
+
+
 @worker_process_shutdown.connect
 def close_connection_pool_on_worker_shutdown(**kwargs) -> None:  # pragma: no cover
     """

src/documents/tests/test_api_chat.py

@@ -38,42 +38,6 @@ class TestChatStreamingViewInputValidation(APITestCase):
             )
         assert resp.status_code == status.HTTP_400_BAD_REQUEST
 
-    def test_answer_is_not_compressed(self) -> None:
-        """
-        GIVEN:
-            - A client that accepts compressed responses
-        WHEN:
-            - It asks the chat endpoint a question
-        THEN:
-            - The answer is streamed unencoded, chunk for chunk
-
-        The stream compressors buffer, so a compressed answer arrives in one
-        piece. The view cannot opt out by flagging the request: DRF's request
-        wrapper proxies reads but keeps writes to itself, so the flag never
-        reaches the Django request the middleware sees.
-        """
-        chunks = [f"token{i} " for i in range(40)]
-        with (
-            mock.patch(
-                "documents.views.AIConfig",
-                return_value=self._mock_ai_enabled(),
-            ),
-            mock.patch(
-                "documents.views.stream_chat_with_documents",
-                return_value=iter(chunks),
-            ),
-        ):
-            resp = self.client.post(
-                "/api/documents/chat/",
-                {"q": "What is in my archive?"},
-                format="json",
-                HTTP_ACCEPT_ENCODING="gzip, deflate, br, zstd",
-            )
-
-        assert resp.status_code == status.HTTP_200_OK
-        assert not resp.has_header("Content-Encoding")
-        assert list(resp.streaming_content) == [c.encode() for c in chunks]
-
     def test_missing_question_is_rejected(self) -> None:
         with mock.patch(
             "documents.views.AIConfig",

src/documents/tests/test_api_objects.py

@@ -2,15 +2,10 @@ import datetime
 import json
 from unittest import mock
 
-from django.contrib.auth.models import Group
 from django.contrib.auth.models import Permission
 from django.contrib.auth.models import User
-from django.db import connection
 from django.test import override_settings
-from django.test.utils import CaptureQueriesContext
 from guardian.shortcuts import assign_perm
-from guardian.shortcuts import get_groups_with_perms
-from guardian.shortcuts import get_users_with_perms
 from rest_framework import status
 from rest_framework.test import APITestCase
 
@@ -847,66 +842,6 @@ class TestBulkEditObjects(APITestCase):
         self.assertEqual(response.status_code, status.HTTP_200_OK)
         self.assertEqual(StoragePath.objects.count(), 0)
 
-    def test_bulk_objects_set_permissions_batched_across_object_count(
-        self,
-    ) -> None:
-        """
-        GIVEN:
-            - Many tags are being bulk-edited to set permissions at once
-        WHEN:
-            - bulk_edit_objects API endpoint is called with set_permissions
-              operation over a small batch vs. a much larger one
-        THEN:
-            - Permissions are applied correctly at both scales
-            - Query count does not grow with the number of tags, i.e. each
-              user/group is applied across all tags with one batched call
-              rather than one call per (tag, identity) pair
-        """
-        group1 = Group.objects.create(name="perm-group")
-        permissions = {
-            "view": {"users": [self.user1.id, self.user2.id], "groups": [group1.id]},
-            "change": {"users": [self.user1.id], "groups": [group1.id]},
-        }
-
-        def run_with_n_tags(n: int) -> int:
-            tags = [Tag.objects.create(name=f"perm-tag-{n}-{i}") for i in range(n)]
-            with CaptureQueriesContext(connection) as ctx:
-                response = self.client.post(
-                    "/api/bulk_edit_objects/",
-                    json.dumps(
-                        {
-                            "objects": [t.id for t in tags],
-                            "object_type": "tags",
-                            "operation": "set_permissions",
-                            "permissions": permissions,
-                            "merge": False,
-                        },
-                    ),
-                    content_type="application/json",
-                )
-            self.assertEqual(response.status_code, status.HTTP_200_OK)
-            for tag in tags:
-                self.assertEqual(get_users_with_perms(tag).count(), 2)
-                self.assertEqual(get_groups_with_perms(tag).count(), 1)
-            return len(ctx.captured_queries)
-
-        small_batch_queries = run_with_n_tags(5)
-        large_batch_queries = run_with_n_tags(50)
-
-        # A tolerance rather than equality, matching the N+1 check in
-        # test_views.py: bulk_create's batch_size caps rows per INSERT, so a
-        # large enough selection does legitimately add statements, and the
-        # per-process ContentType cache makes the first run carry an extra
-        # query. Neither can hide a regression to per-object assignment,
-        # which would be ~10x the small-batch count here.
-        self.assertLessEqual(
-            large_batch_queries,
-            small_batch_queries + 5,
-            "Permission assignment appears to scale with object count: "
-            f"{small_batch_queries} queries for 5 tags vs. "
-            f"{large_batch_queries} for 50",
-        )
-
     def test_bulk_objects_delete_all_filtered(self) -> None:
         """
         GIVEN:

src/documents/tests/test_api_ocr_templates.py

@@ -0,0 +1,449 @@
+"""Tests for the OCR Template API."""
+
+import json
+
+from django.contrib.auth.models import User
+from rest_framework import status
+from rest_framework.test import APITestCase
+
+from documents.models import CustomField
+from documents.models import DocumentType
+from documents.models import OcrTemplate
+from documents.models import OcrTemplateZone
+from documents.tests.utils import DirectoriesMixin
+
+
+class TestOcrTemplatesAPI(DirectoriesMixin, APITestCase):
+    ENDPOINT = "/api/ocr_templates/"
+
+    def setUp(self) -> None:
+        self.user = User.objects.create_superuser(username="temp_admin")
+        self.client.force_authenticate(user=self.user)
+
+        self.doc_type = DocumentType.objects.create(name="Invoice")
+        self.custom_field_text = CustomField.objects.create(
+            name="Invoice Number",
+            data_type=CustomField.FieldDataType.STRING,
+        )
+        self.custom_field_date = CustomField.objects.create(
+            name="Invoice Date",
+            data_type=CustomField.FieldDataType.DATE,
+        )
+        self.custom_field_int = CustomField.objects.create(
+            name="Amount",
+            data_type=CustomField.FieldDataType.INT,
+        )
+        self.custom_field_doclink = CustomField.objects.create(
+            name="Related Docs",
+            data_type=CustomField.FieldDataType.DOCUMENTLINK,
+        )
+
+        return super().setUp()
+
+    def _make_template_data(self, **overrides):
+        data = {
+            "name": "Invoice Template",
+            "document_type": self.doc_type.pk,
+            "default_page": 0,
+            "source_width": 2480,
+            "source_height": 3508,
+            "enabled": True,
+            "zones": [],
+        }
+        data.update(overrides)
+        return data
+
+    def _make_zone_data(self, **overrides):
+        data = {
+            "name": "Zone 1",
+            "custom_field": self.custom_field_text.pk,
+            "x": 100,
+            "y": 100,
+            "width": 200,
+            "height": 50,
+            "ocr_language": "deu+eng",
+            "transform": "strip",
+            "order": 0,
+        }
+        data.update(overrides)
+        return data
+
+    # --- Create ---
+
+    def test_create_template(self):
+        """
+        GIVEN:
+            - A document type and custom fields exist
+        WHEN:
+            - API request to create an OCR template with one zone
+        THEN:
+            - The template and zone are created
+        """
+        data = self._make_template_data(
+            zones=[
+                self._make_zone_data(
+                    name="Invoice Number",
+                    x=1500,
+                    y=200,
+                    width=800,
+                    height=100,
+                ),
+            ],
+        )
+        resp = self.client.post(
+            self.ENDPOINT,
+            data=json.dumps(data),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_201_CREATED)
+
+        result = resp.json()
+        self.assertEqual(result["name"], "Invoice Template")
+        self.assertEqual(result["document_type"], self.doc_type.pk)
+        self.assertEqual(len(result["zones"]), 1)
+        self.assertEqual(result["zones"][0]["name"], "Invoice Number")
+        self.assertEqual(OcrTemplate.objects.count(), 1)
+        self.assertEqual(OcrTemplateZone.objects.count(), 1)
+
+    def test_create_template_multiple_zones(self):
+        """
+        GIVEN:
+            - Multiple custom fields exist
+        WHEN:
+            - A template with multiple zones is created
+        THEN:
+            - All zones are created
+        """
+        data = self._make_template_data(
+            zones=[
+                self._make_zone_data(
+                    name="Invoice Number",
+                    custom_field=self.custom_field_text.pk,
+                ),
+                self._make_zone_data(
+                    name="Invoice Date",
+                    custom_field=self.custom_field_date.pk,
+                    order=1,
+                ),
+            ],
+        )
+        resp = self.client.post(
+            self.ENDPOINT,
+            data=json.dumps(data),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_201_CREATED)
+        self.assertEqual(len(resp.json()["zones"]), 2)
+        self.assertEqual(OcrTemplateZone.objects.count(), 2)
+
+    def test_create_template_no_zones(self):
+        """
+        GIVEN:
+            - Valid template data without zones
+        WHEN:
+            - Template is created
+        THEN:
+            - Template is created with no zones
+        """
+        data = self._make_template_data()
+        resp = self.client.post(
+            self.ENDPOINT,
+            data=json.dumps(data),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_201_CREATED)
+        self.assertEqual(len(resp.json()["zones"]), 0)
+
+    # --- Validation ---
+
+    def test_create_template_zero_source_width_rejected(self):
+        """
+        GIVEN:
+            - Template data with source_width=0
+        WHEN:
+            - Create is attempted
+        THEN:
+            - 400 error is returned
+        """
+        data = self._make_template_data(source_width=0)
+        resp = self.client.post(
+            self.ENDPOINT,
+            data=json.dumps(data),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_400_BAD_REQUEST)
+
+    def test_create_template_zero_source_height_rejected(self):
+        data = self._make_template_data(source_height=0)
+        resp = self.client.post(
+            self.ENDPOINT,
+            data=json.dumps(data),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_400_BAD_REQUEST)
+
+    def test_create_zone_zero_width_rejected(self):
+        data = self._make_template_data(
+            zones=[self._make_zone_data(width=0)],
+        )
+        resp = self.client.post(
+            self.ENDPOINT,
+            data=json.dumps(data),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_400_BAD_REQUEST)
+
+    def test_create_zone_zero_height_rejected(self):
+        data = self._make_template_data(
+            zones=[self._make_zone_data(height=0)],
+        )
+        resp = self.client.post(
+            self.ENDPOINT,
+            data=json.dumps(data),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_400_BAD_REQUEST)
+
+    def test_create_zone_exceeds_source_width_rejected(self):
+        """Zone that extends beyond the source image width should be rejected."""
+        data = self._make_template_data(
+            source_width=1000,
+            zones=[self._make_zone_data(x=800, width=300)],  # 800+300 > 1000
+        )
+        resp = self.client.post(
+            self.ENDPOINT,
+            data=json.dumps(data),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_400_BAD_REQUEST)
+
+    def test_create_zone_exceeds_source_height_rejected(self):
+        data = self._make_template_data(
+            source_height=1000,
+            zones=[self._make_zone_data(y=900, height=200)],  # 900+200 > 1000
+        )
+        resp = self.client.post(
+            self.ENDPOINT,
+            data=json.dumps(data),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_400_BAD_REQUEST)
+
+    def test_create_zone_unsupported_custom_field_type_rejected(self):
+        """DOCUMENTLINK and SELECT fields can't be populated via OCR."""
+        data = self._make_template_data(
+            zones=[self._make_zone_data(custom_field=self.custom_field_doclink.pk)],
+        )
+        resp = self.client.post(
+            self.ENDPOINT,
+            data=json.dumps(data),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_400_BAD_REQUEST)
+
+    # --- List ---
+
+    def test_list_templates(self):
+        template = OcrTemplate.objects.create(
+            name="Test Template",
+            document_type=self.doc_type,
+            source_width=2480,
+            source_height=3508,
+        )
+        OcrTemplateZone.objects.create(
+            template=template,
+            name="Zone 1",
+            custom_field=self.custom_field_text,
+            x=100,
+            y=100,
+            width=200,
+            height=50,
+        )
+
+        resp = self.client.get(self.ENDPOINT)
+        self.assertEqual(resp.status_code, status.HTTP_200_OK)
+        data = resp.json()
+        self.assertEqual(data["count"], 1)
+        self.assertEqual(len(data["results"][0]["zones"]), 1)
+
+    def test_list_empty(self):
+        resp = self.client.get(self.ENDPOINT)
+        self.assertEqual(resp.status_code, status.HTTP_200_OK)
+        self.assertEqual(resp.json()["count"], 0)
+
+    # --- Update ---
+
+    def test_update_template_replaces_zones(self):
+        """PUT should replace all zones with the new set."""
+        template = OcrTemplate.objects.create(
+            name="Old Name",
+            document_type=self.doc_type,
+            source_width=2480,
+            source_height=3508,
+        )
+        OcrTemplateZone.objects.create(
+            template=template,
+            name="Old Zone",
+            custom_field=self.custom_field_text,
+            x=0,
+            y=0,
+            width=100,
+            height=100,
+        )
+
+        data = self._make_template_data(
+            name="New Name",
+            zones=[
+                self._make_zone_data(
+                    name="New Zone",
+                    custom_field=self.custom_field_date.pk,
+                ),
+            ],
+        )
+        resp = self.client.put(
+            f"{self.ENDPOINT}{template.pk}/",
+            data=json.dumps(data),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_200_OK)
+
+        template.refresh_from_db()
+        self.assertEqual(template.name, "New Name")
+        self.assertEqual(OcrTemplateZone.objects.count(), 1)
+        self.assertEqual(OcrTemplateZone.objects.first().name, "New Zone")
+
+    # --- Delete ---
+
+    def test_delete_template_cascades_zones(self):
+        template = OcrTemplate.objects.create(
+            name="To Delete",
+            document_type=self.doc_type,
+            source_width=2480,
+            source_height=3508,
+        )
+        OcrTemplateZone.objects.create(
+            template=template,
+            name="Zone",
+            custom_field=self.custom_field_text,
+            x=0,
+            y=0,
+            width=100,
+            height=100,
+        )
+
+        resp = self.client.delete(f"{self.ENDPOINT}{template.pk}/")
+        self.assertEqual(resp.status_code, status.HTTP_204_NO_CONTENT)
+        self.assertEqual(OcrTemplate.objects.count(), 0)
+        self.assertEqual(OcrTemplateZone.objects.count(), 0)
+
+    def test_delete_nonexistent_returns_404(self):
+        resp = self.client.delete(f"{self.ENDPOINT}99999/")
+        self.assertEqual(resp.status_code, status.HTTP_404_NOT_FOUND)
+
+    # --- Patch ---
+
+    def test_patch_toggle_enabled(self):
+        template = OcrTemplate.objects.create(
+            name="Toggle Test",
+            document_type=self.doc_type,
+            source_width=2480,
+            source_height=3508,
+            enabled=True,
+        )
+
+        resp = self.client.patch(
+            f"{self.ENDPOINT}{template.pk}/",
+            data=json.dumps({"enabled": False}),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_200_OK)
+        template.refresh_from_db()
+        self.assertFalse(template.enabled)
+
+    def test_patch_preserves_zones(self):
+        """PATCH without zones field should not delete existing zones."""
+        template = OcrTemplate.objects.create(
+            name="Patch Test",
+            document_type=self.doc_type,
+            source_width=2480,
+            source_height=3508,
+        )
+        OcrTemplateZone.objects.create(
+            template=template,
+            name="Existing Zone",
+            custom_field=self.custom_field_text,
+            x=0,
+            y=0,
+            width=100,
+            height=100,
+        )
+
+        resp = self.client.patch(
+            f"{self.ENDPOINT}{template.pk}/",
+            data=json.dumps({"name": "Updated Name"}),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_200_OK)
+        self.assertEqual(OcrTemplateZone.objects.count(), 1)
+
+    # --- Auth ---
+
+    def test_unauthenticated_rejected(self):
+        self.client.logout()
+        resp = self.client.get(self.ENDPOINT)
+        self.assertIn(
+            resp.status_code,
+            (status.HTTP_401_UNAUTHORIZED, status.HTTP_403_FORBIDDEN),
+        )
+
+    # --- Quick create field ---
+
+    def test_quick_create_field(self):
+        """Creating a custom field inline from the template editor."""
+        resp = self.client.post(
+            f"{self.ENDPOINT}quick-create-field/",
+            data=json.dumps({"name": "New Field", "data_type": "string"}),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_201_CREATED)
+        data = resp.json()
+        self.assertEqual(data["name"], "New Field")
+        self.assertEqual(data["data_type"], "string")
+        self.assertTrue(data["created"])
+        self.assertTrue(CustomField.objects.filter(name="New Field").exists())
+
+    def test_quick_create_field_existing(self):
+        """If a field with the same name exists, return it without creating."""
+        resp = self.client.post(
+            f"{self.ENDPOINT}quick-create-field/",
+            data=json.dumps({"name": "Invoice Number", "data_type": "string"}),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_200_OK)
+        data = resp.json()
+        self.assertEqual(data["id"], self.custom_field_text.pk)
+        self.assertFalse(data["created"])
+
+    def test_quick_create_field_empty_name_rejected(self):
+        resp = self.client.post(
+            f"{self.ENDPOINT}quick-create-field/",
+            data=json.dumps({"name": "", "data_type": "string"}),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_400_BAD_REQUEST)
+
+    def test_quick_create_field_unsupported_type_rejected(self):
+        resp = self.client.post(
+            f"{self.ENDPOINT}quick-create-field/",
+            data=json.dumps({"name": "Bad Field", "data_type": "documentlink"}),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_400_BAD_REQUEST)
+
+    def test_quick_create_field_select_type_rejected(self):
+        resp = self.client.post(
+            f"{self.ENDPOINT}quick-create-field/",
+            data=json.dumps({"name": "Bad Field", "data_type": "select"}),
+            content_type="application/json",
+        )
+        self.assertEqual(resp.status_code, status.HTTP_400_BAD_REQUEST)

src/documents/tests/test_bulk_edit.py

@@ -5,11 +5,8 @@ from unittest import mock
 
 import pikepdf
 from django.contrib.auth.models import Group
-from django.contrib.auth.models import Permission
 from django.contrib.auth.models import User
-from django.db import connection
 from django.test import TestCase
-from django.test.utils import CaptureQueriesContext
 from guardian.shortcuts import assign_perm
 from guardian.shortcuts import get_groups_with_perms
 from guardian.shortcuts import get_users_with_perms
@@ -22,7 +19,6 @@ from documents.models import Document
 from documents.models import DocumentType
 from documents.models import StoragePath
 from documents.models import Tag
-from documents.permissions import set_permissions_for_objects
 from documents.tests.utils import DirectoriesMixin
 
 
@@ -519,178 +515,6 @@ class TestBulkEdit(DirectoriesMixin, TestCase):
         )
         self.assertEqual(groups_with_perms.count(), 2)
 
-    @mock.patch("documents.tasks.bulk_update_documents.apply_async")
-    def test_set_permissions_batched_across_document_count(
-        self,
-        m,
-    ) -> None:
-        """
-        GIVEN:
-            - Many documents are being bulk-edited to set permissions at once
-        WHEN:
-            - set_permissions runs over a small batch vs. a much larger one
-        THEN:
-            - Permissions are applied correctly at both scales
-            - Query count does not grow with the number of documents, i.e.
-              each user/group is applied across all documents with one
-              batched call rather than one call per (document, identity)
-              pair
-        """
-        permissions = {
-            "view": {
-                "users": [self.user1.id, self.user2.id],
-                "groups": [self.group2.id],
-            },
-            "change": {
-                "users": [self.user1.id],
-                "groups": [self.group2.id],
-            },
-        }
-
-        def run_with_n_documents(n: int) -> int:
-            docs = [
-                Document.objects.create(checksum=f"perm-{n}-{i}", title=f"perm-{n}-{i}")
-                for i in range(n)
-            ]
-            with CaptureQueriesContext(connection) as ctx:
-                bulk_edit.set_permissions(
-                    [doc.id for doc in docs],
-                    set_permissions=permissions,
-                    owner=self.owner,
-                    merge=False,
-                )
-            for doc in docs:
-                self.assertEqual(get_users_with_perms(doc).count(), 2)
-                self.assertEqual(get_groups_with_perms(doc).count(), 1)
-            return len(ctx.captured_queries)
-
-        small_batch_queries = run_with_n_documents(5)
-        large_batch_queries = run_with_n_documents(50)
-
-        # A tolerance rather than equality, matching the N+1 check in
-        # test_views.py: bulk_create's batch_size caps rows per INSERT, so a
-        # large enough selection does legitimately add statements, and the
-        # per-process ContentType cache makes the first run carry an extra
-        # query. Neither can hide a regression to per-document assignment,
-        # which would be ~10x the small-batch count here.
-        self.assertLessEqual(
-            large_batch_queries,
-            small_batch_queries + 5,
-            "Permission assignment appears to scale with document count: "
-            f"{small_batch_queries} queries for 5 documents vs. "
-            f"{large_batch_queries} for 50",
-        )
-
-    @mock.patch("documents.tasks.bulk_update_documents.apply_async")
-    def test_set_permissions_grants_direct_perm_even_if_already_granted_via_group(
-        self,
-        m,
-    ) -> None:
-        """
-        GIVEN:
-            - A user already has view access to a document via group
-              membership, with no direct grant of their own
-        WHEN:
-            - set_permissions explicitly grants that same user direct view
-              access via bulk_edit
-        THEN:
-            - A direct permission grant is created for the user, not skipped
-              because they already have equivalent access via the group
-
-        Regression test: guardian's queryset-aware assign_perm() (routed to
-        when the target is a list/queryset) skips creating a direct row for
-        anyone whose ObjectPermissionChecker.has_perm() already returns True
-        -- which includes group-derived access. The single-object assign_perm
-        this bulk path replaces has no such check; it always ensures a
-        direct row via get_or_create. Losing that guarantee would mean
-        revoking the group's grant later silently strips access that was
-        supposed to be explicit.
-        """
-        self.doc1.owner = self.user1
-        self.doc1.save()
-        self.user1.groups.add(self.group1)
-        assign_perm("view_document", self.group1, self.doc1)
-
-        bulk_edit.set_permissions(
-            [self.doc1.id],
-            set_permissions={
-                "view": {"users": [self.user1.id], "groups": []},
-            },
-            merge=True,
-        )
-
-        direct_users = get_users_with_perms(
-            self.doc1,
-            only_with_perms_in=["view_document"],
-            with_group_users=False,
-        )
-        self.assertIn(self.user1, direct_users)
-
-    def test_set_permissions_for_objects_raises_for_unknown_action(self) -> None:
-        """
-        GIVEN:
-            - An unrecognized permission action name with users to grant it
-              to
-        WHEN:
-            - set_permissions_for_objects is called
-        THEN:
-            - Permission.DoesNotExist is raised, not a silent no-op
-
-        Regression test: the endpoint that calls this
-        (BulkEditObjectPermissionsView) never actually validates action
-        names against the raw client-supplied permissions dict --
-        BulkEditObjectsSerializer._validate_permissions calls
-        validate_set_permissions() only for its side-effecting user/group id
-        checks and discards the filtered dict it returns -- so a bogus
-        action key reaches this function as-is. Resolving the Permission via
-        a bare `.filter()` (which returns empty instead of raising) would
-        silently drop the grant and report success.
-        """
-        with self.assertRaises(Permission.DoesNotExist):
-            set_permissions_for_objects(
-                {"not_a_real_action": {"users": [self.user1.id], "groups": []}},
-                Document,
-                [self.doc1.pk],
-            )
-
-    def test_set_permissions_for_objects_unknown_action_applies_nothing(
-        self,
-    ) -> None:
-        """
-        GIVEN:
-            - A permissions dict with a valid action ordered ahead of an
-              unrecognized one
-        WHEN:
-            - set_permissions_for_objects is called
-        THEN:
-            - Permission.DoesNotExist is raised
-            - The valid action ahead of it is not applied either
-
-        Every action is resolved before any row is written, so a bad action
-        name cannot leave a half-applied change behind. That matters because
-        BulkEditObjectsView turns this exception into a 400: without the
-        up-front resolution the client would be told the request failed
-        while the leading action had already been committed.
-        """
-        with self.assertRaises(Permission.DoesNotExist):
-            set_permissions_for_objects(
-                {
-                    "view": {"users": [self.user1.id], "groups": []},
-                    "not_a_real_action": {"users": [self.user1.id], "groups": []},
-                },
-                Document,
-                [self.doc1.pk],
-            )
-
-        self.assertNotIn(
-            self.user1,
-            get_users_with_perms(
-                self.doc1,
-                only_with_perms_in=["view_document"],
-                with_group_users=False,
-            ),
-        )
-
     @mock.patch("documents.models.Document.delete")
     def test_delete_documents_old_uuid_field(self, m) -> None:
         m.side_effect = Exception("Data too long for column 'transaction_id' at row 1")

src/documents/tests/test_classifier.py

@@ -3,7 +3,6 @@ import warnings
 from pathlib import Path
 from unittest import mock
 
-import numpy as np
 import pytest
 from django.conf import settings
 from django.test import TestCase
@@ -12,7 +11,6 @@ from django.test import override_settings
 from documents.classifier import ClassifierModelCorruptError
 from documents.classifier import DocumentClassifier
 from documents.classifier import IncompatibleClassifierVersionError
-from documents.classifier import _predict_with_threshold
 from documents.classifier import load_classifier
 from documents.models import Correspondent
 from documents.models import Document
@@ -627,103 +625,6 @@ class TestClassifier(DirectoriesMixin, TestCase):
         self.assertEqual(self.classifier.predict_storage_path(doc1.content), sp.pk)
         self.assertIsNone(self.classifier.predict_storage_path(doc2.content))
 
-    def test_predict_rejects_prediction_below_match_threshold(self) -> None:
-        """
-        GIVEN:
-            - Classifiers trained against test data with confident predictions
-        WHEN:
-            - CLASSIFIER_MATCH_THRESHOLD exceeds the model's confidence
-        THEN:
-            - Every predict_* method discards the match in favor of no match
-        """
-        c1 = Correspondent.objects.create(
-            name="c1",
-            matching_algorithm=Correspondent.MATCH_AUTO,
-        )
-        dt1 = DocumentType.objects.create(
-            name="dt1",
-            matching_algorithm=DocumentType.MATCH_AUTO,
-        )
-        sp1 = StoragePath.objects.create(
-            name="sp1",
-            matching_algorithm=StoragePath.MATCH_AUTO,
-        )
-
-        doc1 = Document.objects.create(
-            title="doc1",
-            content="this is a document from c1",
-            correspondent=c1,
-            document_type=dt1,
-            storage_path=sp1,
-            checksum="A",
-        )
-        Document.objects.create(
-            title="doc2",
-            content="this is a document from no one",
-            checksum="B",
-        )
-
-        self.classifier.train()
-
-        predictors = {
-            "correspondent": self.classifier.predict_correspondent,
-            "document_type": self.classifier.predict_document_type,
-            "storage_path": self.classifier.predict_storage_path,
-        }
-        # No real prediction can reach a confidence this high, so this
-        # isolates the threshold check from the model's actual output.
-        with override_settings(CLASSIFIER_MATCH_THRESHOLD=0.999999):
-            for name, predict in predictors.items():
-                with self.subTest(field=name):
-                    self.assertIsNone(predict(doc1.content))
-
-    def test_train_uses_balanced_sample_weight(self) -> None:
-        """
-        GIVEN:
-            - A training set with correspondents, document types and storage paths
-        WHEN:
-            - The classifier is trained
-        THEN:
-            - Each MLP classifier is fit with balanced sample weights, so that
-              over-represented classes don't dominate predictions
-        """
-        c1 = Correspondent.objects.create(
-            name="c1",
-            matching_algorithm=Correspondent.MATCH_AUTO,
-        )
-        dt1 = DocumentType.objects.create(
-            name="dt1",
-            matching_algorithm=DocumentType.MATCH_AUTO,
-        )
-        sp1 = StoragePath.objects.create(
-            name="sp1",
-            matching_algorithm=StoragePath.MATCH_AUTO,
-        )
-
-        Document.objects.create(
-            title="doc1",
-            content="this is a document from c1",
-            correspondent=c1,
-            document_type=dt1,
-            storage_path=sp1,
-            checksum="A",
-        )
-        Document.objects.create(
-            title="doc2",
-            content="this is a document from no one",
-            checksum="B",
-        )
-
-        with mock.patch(
-            "sklearn.utils.class_weight.compute_sample_weight",
-            return_value=None,
-        ) as mocked_compute_sample_weight:
-            self.classifier.train()
-
-        self.assertEqual(mocked_compute_sample_weight.call_count, 3)
-        for call in mocked_compute_sample_weight.call_args_list:
-            self.assertEqual(call.args[0], "balanced")
-
     def test_one_tag_predict(self) -> None:
         t1 = Tag.objects.create(name="t1", matching_algorithm=Tag.MATCH_AUTO, pk=12)
 
@@ -909,52 +810,6 @@ class TestClassifier(DirectoriesMixin, TestCase):
             load_classifier(raise_exception=True)
 
 
-class _StubProbaClassifier:
-    """
-    A fake scikit-learn classifier exposing just enough of the API for
-    `_predict_with_threshold`: `classes_` and `predict_proba`.
-    """
-
-    def __init__(self, classes: list[int], probabilities: list[float]) -> None:
-        self.classes_ = np.array(classes)
-        self._probabilities = np.array([probabilities])
-
-    def predict_proba(self, X) -> np.ndarray:
-        return self._probabilities
-
-
-@pytest.mark.parametrize(
-    ("classes", "probabilities", "threshold", "expected"),
-    [
-        # confident prediction above the threshold is returned
-        ([-1, 3], [0.1, 0.9], 0.6, 3),
-        # prediction below the threshold is discarded
-        ([-1, 3], [0.45, 0.55], 0.6, None),
-        # boundary: exactly at the threshold is accepted, not discarded
-        ([-1, 3], [0.4, 0.6], 0.6, 3),
-        # the winning class is the "no match" pseudo-class, regardless of its
-        # own confidence
-        ([-1, 3], [0.99, 0.01], 0.0, None),
-        # threshold of 0.0 disables the confidence check entirely
-        ([-1, 3], [0.45, 0.55], 0.0, 3),
-    ],
-)
-def test_predict_with_threshold(classes, probabilities, threshold, expected) -> None:
-    classifier = _StubProbaClassifier(classes, probabilities)
-    result = _predict_with_threshold(classifier, X=None, threshold=threshold)
-    assert result == expected
-
-
-def test_classifier_match_threshold_default() -> None:
-    """
-    GIVEN:
-        - No PAPERLESS_CLASSIFIER_MATCH_THRESHOLD environment variable is set
-    THEN:
-        - The classifier match threshold defaults to 0.6
-    """
-    assert settings.CLASSIFIER_MATCH_THRESHOLD == 0.6
-
-
 def test_preprocess_content() -> None:
     """
     GIVEN:

src/documents/tests/test_document_list_effective_content_annotation.py

@@ -1,457 +0,0 @@
-from __future__ import annotations
-
-from types import SimpleNamespace
-from typing import TYPE_CHECKING
-
-import pytest
-from django.db import connection
-from django.test.utils import CaptureQueriesContext
-from rest_framework import status
-
-from documents.models import Document
-from documents.tests.factories import DocumentFactory
-from documents.versioning import LATEST_VERSION_CONTENT_PREFETCH_ATTR
-from documents.versioning import has_prefetched_effective_content
-from documents.versioning import latest_version_content_prefetch
-from documents.views import DocumentViewSet
-
-if TYPE_CHECKING:
-    from rest_framework.test import APIClient
-
-
-class TestNeedsEffectiveContentAnnotation:
-    """
-    DocumentViewSet._needs_effective_content_annotation() decides whether
-    the effective_content correlated subquery is worth attaching to the
-    queryset at all -- see TestDocumentListEffectiveContentAnnotation below
-    for why. This only checks that decision's own logic (a plain query-param
-    membership test), not that Django/DRF's filtering machinery works.
-    """
-
-    @pytest.mark.parametrize(
-        ("params", "expected"),
-        [
-            ({}, False),
-            ({"ordering": "-added"}, False),
-            ({"tags__id__in": "1,2"}, False),
-            ({"search": ""}, False),
-            ({"search": "   "}, False),
-            ({"content__icontains": ""}, False),
-            ({"search": "foo"}, True),
-            ({"title_content": "foo"}, True),
-            ({"content__istartswith": "foo"}, True),
-            ({"content__iendswith": "foo"}, True),
-            ({"content__icontains": "foo"}, True),
-            ({"content__iexact": "foo"}, True),
-        ],
-    )
-    def test_detects_content_filter_params(
-        self,
-        params: dict[str, str],
-        expected: bool,  # noqa: FBT001
-    ) -> None:
-        """
-        GIVEN:
-            - A view bound to a request carrying the given query params
-        WHEN:
-            - Checking whether the effective_content annotation is needed
-        THEN:
-            - It is needed only for requests that actually filter on it
-        """
-        view = DocumentViewSet()
-        view.request = SimpleNamespace(query_params=params)
-
-        assert view._needs_effective_content_annotation() is expected
-
-
-class TestNeedsEffectiveContentPrefetch:
-    """
-    DocumentViewSet._needs_effective_content_prefetch() decides whether the
-    single-version content prefetch is worth attaching. It has to read the
-    `fields` param exactly the way get_serializer() does, or a request whose
-    response includes content ends up without the prefetch and pays
-    get_effective_content()'s per-instance fallback instead.
-    """
-
-    @pytest.mark.parametrize(
-        ("params", "expected"),
-        [
-            pytest.param({}, True, id="no-fields-param-keeps-every-field"),
-            pytest.param({"fields": ""}, True, id="blank-fields-keeps-every-field"),
-            pytest.param(
-                {"fields": "id,content"},
-                True,
-                id="content-among-requested-fields",
-            ),
-            pytest.param({"fields": "content"}, True, id="content-only"),
-            pytest.param({"fields": "id"}, False, id="content-not-requested"),
-            pytest.param(
-                {"fields": "id,title"},
-                False,
-                id="several-fields-without-content",
-            ),
-        ],
-    )
-    def test_detects_whether_content_can_reach_the_response(
-        self,
-        params: dict[str, str],
-        expected: bool,  # noqa: FBT001
-    ) -> None:
-        """
-        GIVEN:
-            - A view bound to a request carrying the given query params
-        WHEN:
-            - Checking whether the content prefetch is needed
-        THEN:
-            - It is needed exactly when get_serializer() would emit content,
-              which treats a blank `fields` the same as an absent one
-        """
-        view = DocumentViewSet()
-        view.request = SimpleNamespace(query_params=params)
-
-        assert view._needs_effective_content_prefetch() is expected
-
-
-@pytest.mark.django_db
-class TestDocumentListEffectiveContentAnnotation:
-    """
-    DocumentViewSet.get_queryset() only attaches the effective_content
-    correlated subquery when a request actually filters on it. Attaching it
-    unconditionally re-executes it once per candidate row before the page's
-    LIMIT is applied -- fine on SQLite/Postgres, but pathological on
-    MariaDB's default cardinality estimation for the root_document_id
-    self-join once candidate counts get large (see the root_document_id /
-    effective_content perf investigation).
-    """
-
-    def test_list_without_content_filter_skips_annotation_but_returns_latest_content(
-        self,
-        admin_client: APIClient,
-    ) -> None:
-        """
-        GIVEN:
-            - A root document whose latest version has different content
-        WHEN:
-            - Listing documents with no search/content-filter param
-        THEN:
-            - The response still reflects the latest version's content
-            - The database never evaluates effective_content per row
-        """
-        root = DocumentFactory(content="old-root-content")
-        DocumentFactory(
-            root_document=root,
-            version_index=1,
-            content="new-version-content",
-        )
-
-        with CaptureQueriesContext(connection) as ctx:
-            response = admin_client.get("/api/documents/?fields=id,content")
-
-        assert response.status_code == status.HTTP_200_OK
-        assert response.data["results"] == [
-            {"id": root.id, "content": "new-version-content"},
-        ]
-        assert not any(
-            "effective_content" in query["sql"] for query in ctx.captured_queries
-        )
-
-    @pytest.mark.parametrize(
-        "fields_param",
-        [
-            pytest.param("", id="blank-fields"),
-            pytest.param("id,content", id="content-requested"),
-        ],
-    )
-    def test_content_resolves_without_a_query_per_document(
-        self,
-        admin_client: APIClient,
-        fields_param: str,
-    ) -> None:
-        """
-        GIVEN:
-            - One versioned root document, then two more
-        WHEN:
-            - Listing documents with a `fields` param that keeps content
-        THEN:
-            - Every root's content resolves to its latest version's
-            - The query count does not grow with the number of documents,
-              i.e. a blank `fields` does not skip the prefetch and fall back
-              to loading each root's deferred version content
-        """
-        first = DocumentFactory(content="first-root-content")
-        DocumentFactory(
-            root_document=first,
-            version_index=1,
-            content="first-version-content",
-        )
-
-        with CaptureQueriesContext(connection) as one_document:
-            response = admin_client.get(f"/api/documents/?fields={fields_param}")
-
-        assert response.status_code == status.HTTP_200_OK
-        assert [r["content"] for r in response.data["results"]] == [
-            "first-version-content",
-        ]
-
-        for index in range(2):
-            root = DocumentFactory(content=f"root-content-{index}")
-            DocumentFactory(
-                root_document=root,
-                version_index=1,
-                content=f"version-content-{index}",
-            )
-        with CaptureQueriesContext(connection) as three_documents:
-            response = admin_client.get(f"/api/documents/?fields={fields_param}")
-
-        assert response.status_code == status.HTTP_200_OK
-        assert sorted(r["content"] for r in response.data["results"]) == [
-            "first-version-content",
-            "version-content-0",
-            "version-content-1",
-        ]
-        assert len(_get_document_queries(three_documents)) == len(
-            _get_document_queries(one_document),
-        )
-
-    def test_list_without_content_field_skips_prefetch_and_omits_content(
-        self,
-        admin_client: APIClient,
-    ) -> None:
-        """
-        GIVEN:
-            - A versioned root document
-        WHEN:
-            - Listing documents without asking for content
-        THEN:
-            - Content is neither serialized nor resolved
-            - Nothing pays for the prefetch or the per-instance fallback
-        """
-        root = DocumentFactory(content="root-content")
-        DocumentFactory(
-            root_document=root,
-            version_index=1,
-            content="version-content",
-        )
-
-        with CaptureQueriesContext(connection) as ctx:
-            response = admin_client.get("/api/documents/?fields=id")
-
-        assert response.status_code == status.HTTP_200_OK
-        assert response.data["results"] == [{"id": root.id}]
-        assert _get_effective_content_fallback_queries(ctx) == []
-        # Only the list query itself reads a content column: no extra query
-        # for the skipped prefetch, none for a per-instance fallback
-        content_queries = [
-            query
-            for query in ctx.captured_queries
-            if '"documents_document"."content"' in query["sql"]
-        ]
-        assert len(content_queries) == 1
-
-    def test_latest_version_content_prefetch_carries_only_the_newest_version(
-        self,
-    ) -> None:
-        """
-        GIVEN:
-            - A root document with two versions
-        WHEN:
-            - Fetching the root through latest_version_content_prefetch()
-        THEN:
-            - The prefetch carries only the single newest version, not every
-              historical version's content (the whole point of not reusing
-              the metadata-only "versions" prefetch for this)
-        """
-        root = DocumentFactory(content="root-content")
-        DocumentFactory(
-            root_document=root,
-            version_index=1,
-            content="older-version-content",
-        )
-        DocumentFactory(
-            root_document=root,
-            version_index=2,
-            content="newest-version-content",
-        )
-
-        fetched_root = (
-            Document.objects.filter(pk=root.pk)
-            .prefetch_related(
-                latest_version_content_prefetch(),
-            )
-            .get()
-        )
-
-        latest = getattr(fetched_root, LATEST_VERSION_CONTENT_PREFETCH_ATTR)
-        assert [v.content for v in latest] == ["newest-version-content"]
-
-
-class TestHasPrefetchedEffectiveContent:
-    """
-    DocumentSerializer.to_representation() only calls get_effective_content()
-    when has_prefetched_effective_content() says it's cheap -- otherwise a
-    caller that never set up an annotation or prefetch (TrashView,
-    GlobalSearchView, which build their own querysets and don't display
-    content at all) would pay for a per-instance query nobody asked for.
-    """
-
-    def test_false_with_no_annotation_or_prefetch(self) -> None:
-        """
-        GIVEN:
-            - A document the ORM never annotated or prefetched for
-        WHEN:
-            - Asking whether its effective content is already resolved
-        THEN:
-            - It is not, so the serializer must leave it alone
-        """
-        document = DocumentFactory.build()
-
-        assert has_prefetched_effective_content(document) is False
-
-    def test_true_with_effective_content_annotation(self) -> None:
-        """
-        GIVEN:
-            - A document carrying the queryset's effective_content annotation
-        WHEN:
-            - Asking whether its effective content is already resolved
-        THEN:
-            - It is, straight off the annotation
-        """
-        document = DocumentFactory.build()
-        document.effective_content = "resolved"
-
-        assert has_prefetched_effective_content(document) is True
-
-    def test_true_with_lean_prefetch_attr_even_when_empty(self) -> None:
-        """
-        GIVEN:
-            - A document the lean content prefetch ran for, finding no versions
-        WHEN:
-            - Asking whether its effective content is already resolved
-        THEN:
-            - It is: an empty prefetch is an answer, not a missing one
-        """
-        document = DocumentFactory.build()
-        setattr(document, LATEST_VERSION_CONTENT_PREFETCH_ATTR, [])
-
-        assert has_prefetched_effective_content(document) is True
-
-    def test_true_with_metadata_versions_prefetch_cache(self) -> None:
-        """
-        GIVEN:
-            - A document carrying only the metadata "versions" prefetch
-        WHEN:
-            - Asking whether its effective content is already resolved
-        THEN:
-            - It is, via get_effective_content()'s prefetch-cache branch
-        """
-        document = DocumentFactory.build()
-        document._prefetched_objects_cache = {"versions": []}
-
-        assert has_prefetched_effective_content(document) is True
-
-
-def _get_document_queries(
-    ctx: CaptureQueriesContext,
-) -> list[dict[str, str]]:
-    """
-    The queries a list request spends on the documents themselves, i.e.
-    everything but the one-time django_content_type lookup guardian's
-    permission filtering makes. That lookup is process-cached, and the
-    autouse fixture in conftest clears the cache before every test, so it
-    lands in whichever request happens to run first and never repeats --
-    counting it makes a request look like it costs one query more than the
-    identical request after it.
-    """
-    return [q for q in ctx.captured_queries if '"django_content_type"' not in q["sql"]]
-
-
-def _get_effective_content_fallback_queries(
-    ctx: CaptureQueriesContext,
-) -> list[dict[str, str]]:
-    """
-    Document.get_effective_content()'s per-instance fallback (no annotation,
-    no prefetch) is a `.values_list("content", flat=True).first()` query --
-    a SELECT of just the content column. Distinct from get_versions()'s own,
-    unrelated per-instance metadata query (id/checksum/added/etc, no
-    content) run to build the "versions" response field, which isn't part
-    of what this test file covers.
-    """
-    return [
-        q
-        for q in ctx.captured_queries
-        if q["sql"].startswith('SELECT "documents_document"."content" FROM')
-    ]
-
-
-@pytest.mark.django_db
-class TestTrashAndGlobalSearchEffectiveContentIsNeverPerInstance:
-    """
-    TrashView and GlobalSearchView serialize Document instances with
-    DocumentSerializer too, but build their querysets independently of
-    DocumentViewSet.get_queryset(). TrashView doesn't display content at all,
-    so it keeps the document's own unresolved content; GlobalSearchView
-    annotates effective_content itself, so it shows the latest version's.
-    Neither should ever fall back to a per-instance query.
-    """
-
-    def test_trash_list_shows_unresolved_content_with_no_extra_query(
-        self,
-        admin_client: APIClient,
-    ) -> None:
-        """
-        GIVEN:
-            - A trashed root document whose own content differs from what a
-              version would have had (also trashed, deletion cascades)
-        WHEN:
-            - Listing trash
-        THEN:
-            - The response shows the document's own content
-            - Nothing ever queries for versions to resolve it
-        """
-        root = DocumentFactory(content="own-content")
-        DocumentFactory(
-            root_document=root,
-            version_index=1,
-            content="version-content",
-        )
-        root.delete()
-
-        with CaptureQueriesContext(connection) as ctx:
-            response = admin_client.get("/api/trash/")
-
-        assert response.status_code == status.HTTP_200_OK
-        [result] = [r for r in response.data["results"] if r["id"] == root.id]
-        assert result["content"] == "own-content"
-        assert _get_effective_content_fallback_queries(ctx) == []
-
-    def test_global_search_db_only_shows_latest_version_content_with_no_extra_query(
-        self,
-        admin_client: APIClient,
-    ) -> None:
-        """
-        GIVEN:
-            - A root document, findable by title, whose own content differs
-              from its latest version's
-        WHEN:
-            - Using the global search endpoint's db_only mode
-        THEN:
-            - The response shows the latest version's content, resolved by
-              GlobalSearchView's own effective_content annotation
-            - There is no per-instance fallback query
-        """
-        root = DocumentFactory(title="findme", content="own-content")
-        DocumentFactory(
-            root_document=root,
-            version_index=1,
-            content="version-content",
-        )
-
-        with CaptureQueriesContext(connection) as ctx:
-            response = admin_client.get(
-                "/api/search/?query=findme&db_only=true",
-            )
-
-        assert response.status_code == status.HTTP_200_OK
-        [result] = [d for d in response.data["documents"] if d["id"] == root.id]
-        assert result["content"] == "version-content"
-        assert _get_effective_content_fallback_queries(ctx) == []

src/documents/tests/test_tag_hierarchy.py

@@ -2,7 +2,6 @@ from unittest import mock
 
 from django.contrib.auth.models import Permission
 from django.contrib.auth.models import User
-from rest_framework import status
 from rest_framework.test import APITestCase
 
 from documents import bulk_edit
@@ -109,44 +108,6 @@ class TestTagHierarchy(DirectoriesMixin, APITestCase):
         self.document.refresh_from_db()
         assert self.document.tags.count() == 0
 
-    def test_remove_inbox_tags_removes_nested_children(self) -> None:
-        inbox = Tag.objects.create(name="Inbox", is_inbox_tag=True)
-        nested = Tag.objects.create(name="Nested", tn_parent=inbox)
-        self.document.add_nested_tags([nested])
-
-        resp = self.client.patch(
-            f"/api/documents/{self.document.pk}/",
-            {"title": "new title", "remove_inbox_tags": True},
-            format="json",
-        )
-        assert resp.status_code == status.HTTP_200_OK
-        self.document.refresh_from_db()
-        assert self.document.tags.count() == 0
-
-        # A subsequent save must not re-add the inbox tag as an ancestor
-        resp = self.client.patch(
-            f"/api/documents/{self.document.pk}/",
-            {"title": "another title", "tags": [], "remove_inbox_tags": True},
-            format="json",
-        )
-        assert resp.status_code == status.HTTP_200_OK
-        self.document.refresh_from_db()
-        assert self.document.tags.count() == 0
-
-    def test_remove_inbox_tags_keeps_inbox_when_nested_child_added(self) -> None:
-        inbox = Tag.objects.create(name="Inbox", is_inbox_tag=True)
-        nested = Tag.objects.create(name="Nested", tn_parent=inbox)
-        self.document.add_nested_tags([inbox])
-
-        self.client.patch(
-            f"/api/documents/{self.document.pk}/",
-            {"tags": [nested.pk], "remove_inbox_tags": True},
-            format="json",
-        )
-        self.document.refresh_from_db()
-        tags = set(self.document.tags.values_list("pk", flat=True))
-        assert tags == {inbox.pk, nested.pk}
-
     def test_bulk_edit_respects_hierarchy(self) -> None:
         bulk_edit.add_tag([self.document.pk], self.child.pk)
         self.document.refresh_from_db()

src/documents/tests/test_zone_ocr.py

@@ -0,0 +1,454 @@
+"""Tests for the zone-based OCR extraction engine."""
+
+import tempfile
+from pathlib import Path
+from unittest.mock import MagicMock
+from unittest.mock import patch
+
+from django.test import TestCase
+
+from documents.models import CustomField
+from documents.models import CustomFieldInstance
+from documents.models import Document
+from documents.models import DocumentType
+from documents.models import OcrTemplate
+from documents.models import OcrTemplateZone
+from documents.zone_ocr import _apply_transform
+from documents.zone_ocr import _convert_value
+from documents.zone_ocr import _detect_mime
+from documents.zone_ocr import _resolve_doc_path
+from documents.zone_ocr import run_zone_extraction
+
+
+class TestApplyTransform(TestCase):
+    """Tests for the _apply_transform function."""
+
+    def test_strip(self):
+        self.assertEqual(_apply_transform("  hello  ", "strip"), "hello")
+
+    def test_none_transform(self):
+        self.assertEqual(_apply_transform("  hello  ", "none"), "hello")
+
+    def test_uppercase(self):
+        self.assertEqual(_apply_transform("hello world", "uppercase"), "HELLO WORLD")
+
+    def test_lowercase(self):
+        self.assertEqual(_apply_transform("HELLO WORLD", "lowercase"), "hello world")
+
+    def test_numeric_basic(self):
+        self.assertEqual(_apply_transform("INV-2026-001", "numeric"), "2026-001")
+
+    def test_numeric_with_currency(self):
+        self.assertEqual(_apply_transform("€1,234.56", "numeric"), "1,234.56")
+
+    def test_numeric_empty_result_falls_back(self):
+        self.assertEqual(_apply_transform("abc", "numeric"), "abc")
+
+    def test_date_dmy_dots(self):
+        self.assertEqual(_apply_transform("13.04.2026", "date_dmy"), "2026-04-13")
+
+    def test_date_dmy_slashes(self):
+        self.assertEqual(_apply_transform("01/12/2025", "date_dmy"), "2025-12-01")
+
+    def test_date_dmy_two_digit_year(self):
+        self.assertEqual(_apply_transform("13.04.26", "date_dmy"), "2026-04-13")
+
+    def test_date_dmy_with_prefix(self):
+        self.assertEqual(_apply_transform("Date: 01/12/2025", "date_dmy"), "2025-12-01")
+
+    def test_date_dmy_invalid_falls_back(self):
+        self.assertEqual(_apply_transform("32.13.2026", "date_dmy"), "32.13.2026")
+
+    def test_date_dmy_no_match_falls_back(self):
+        self.assertEqual(_apply_transform("not a date", "date_dmy"), "not a date")
+
+    def test_date_ymd_dashes(self):
+        self.assertEqual(_apply_transform("2026-04-13", "date_ymd"), "2026-04-13")
+
+    def test_date_ymd_slashes(self):
+        self.assertEqual(_apply_transform("2026/04/13", "date_ymd"), "2026-04-13")
+
+    def test_date_ymd_invalid_falls_back(self):
+        self.assertEqual(_apply_transform("2026-13-32", "date_ymd"), "2026-13-32")
+
+    def test_empty_string(self):
+        self.assertEqual(_apply_transform("", "strip"), "")
+
+    def test_whitespace_only(self):
+        self.assertEqual(_apply_transform("   ", "strip"), "")
+
+    def test_unknown_transform_strips(self):
+        self.assertEqual(_apply_transform("  hello  ", "unknown"), "hello")
+
+
+class TestConvertValue(TestCase):
+    """Tests for the _convert_value function."""
+
+    def test_string(self):
+        self.assertEqual(
+            _convert_value("Hello", CustomField.FieldDataType.STRING),
+            "Hello",
+        )
+
+    def test_string_truncation(self):
+        result = _convert_value("x" * 200, CustomField.FieldDataType.STRING)
+        self.assertEqual(len(result), 128)
+
+    def test_url(self):
+        self.assertEqual(
+            _convert_value("https://example.com", CustomField.FieldDataType.URL),
+            "https://example.com",
+        )
+
+    def test_long_text(self):
+        long = "x" * 500
+        self.assertEqual(
+            _convert_value(long, CustomField.FieldDataType.LONG_TEXT),
+            long,
+        )
+
+    def test_int_simple(self):
+        self.assertEqual(_convert_value("42", CustomField.FieldDataType.INT), 42)
+
+    def test_int_with_noise(self):
+        self.assertEqual(_convert_value("INV-123", CustomField.FieldDataType.INT), 123)
+
+    def test_int_negative(self):
+        self.assertEqual(_convert_value("-42", CustomField.FieldDataType.INT), -42)
+
+    def test_int_empty_returns_none(self):
+        self.assertIsNone(_convert_value("abc", CustomField.FieldDataType.INT))
+
+    def test_int_only_dash_returns_none(self):
+        self.assertIsNone(_convert_value("-", CustomField.FieldDataType.INT))
+
+    def test_float_simple(self):
+        self.assertAlmostEqual(
+            _convert_value("1234.56", CustomField.FieldDataType.FLOAT),
+            1234.56,
+        )
+
+    def test_float_european_format(self):
+        self.assertAlmostEqual(
+            _convert_value("1.234,56", CustomField.FieldDataType.FLOAT),
+            1234.56,
+        )
+
+    def test_float_us_format(self):
+        self.assertAlmostEqual(
+            _convert_value("1,234.56", CustomField.FieldDataType.FLOAT),
+            1234.56,
+        )
+
+    def test_float_comma_only(self):
+        self.assertAlmostEqual(
+            _convert_value("1234,56", CustomField.FieldDataType.FLOAT),
+            1234.56,
+        )
+
+    def test_float_empty_returns_none(self):
+        self.assertIsNone(_convert_value("abc", CustomField.FieldDataType.FLOAT))
+
+    def test_float_only_separator_returns_none(self):
+        self.assertIsNone(_convert_value(",", CustomField.FieldDataType.FLOAT))
+
+    def test_date_iso(self):
+        self.assertEqual(
+            _convert_value("2026-04-13", CustomField.FieldDataType.DATE),
+            "2026-04-13",
+        )
+
+    def test_date_invalid_returns_none(self):
+        self.assertIsNone(_convert_value("not a date", CustomField.FieldDataType.DATE))
+
+    def test_date_invalid_values_returns_none(self):
+        self.assertIsNone(_convert_value("2026-13-32", CustomField.FieldDataType.DATE))
+
+    def test_monetary_simple(self):
+        self.assertEqual(
+            _convert_value("123.45", CustomField.FieldDataType.MONETARY),
+            "123.45",
+        )
+
+    def test_monetary_european(self):
+        self.assertEqual(
+            _convert_value("1.234,56", CustomField.FieldDataType.MONETARY),
+            "1234.56",
+        )
+
+    def test_monetary_with_currency_symbol(self):
+        self.assertEqual(
+            _convert_value("€1,234.56", CustomField.FieldDataType.MONETARY),
+            "1234.56",
+        )
+
+    def test_monetary_empty_returns_none(self):
+        self.assertIsNone(_convert_value("CHF", CustomField.FieldDataType.MONETARY))
+
+    def test_bool_true(self):
+        for val in ("true", "True", "yes", "1", "ja", "x", "X"):
+            self.assertTrue(
+                _convert_value(val, CustomField.FieldDataType.BOOL),
+                f"Expected True for {val!r}",
+            )
+
+    def test_bool_false(self):
+        for val in ("false", "False", "no", "0", "nein"):
+            self.assertFalse(
+                _convert_value(val, CustomField.FieldDataType.BOOL),
+                f"Expected False for {val!r}",
+            )
+
+    def test_bool_unknown_returns_none(self):
+        self.assertIsNone(_convert_value("maybe", CustomField.FieldDataType.BOOL))
+
+    def test_unsupported_type_returns_none(self):
+        self.assertIsNone(
+            _convert_value("test", CustomField.FieldDataType.DOCUMENTLINK),
+        )
+        self.assertIsNone(
+            _convert_value("test", CustomField.FieldDataType.SELECT),
+        )
+
+    def test_empty_string_returns_none(self):
+        self.assertIsNone(_convert_value("", CustomField.FieldDataType.STRING))
+
+
+class TestDetectMime(TestCase):
+    """Tests for _detect_mime."""
+
+    def test_pdf_extension(self):
+        self.assertEqual(_detect_mime(Path("test.pdf")), "application/pdf")
+
+    def test_png_extension(self):
+        self.assertEqual(_detect_mime(Path("test.png")), "image/png")
+
+    def test_jpg_extension(self):
+        self.assertEqual(_detect_mime(Path("test.jpg")), "image/jpeg")
+
+    def test_unknown_extension(self):
+        self.assertIsNone(_detect_mime(Path("test.xyz")))
+
+    def test_webp_extension(self):
+        self.assertEqual(_detect_mime(Path("test.webp")), "image/webp")
+
+
+class TestResolveDocPath(TestCase):
+    """Tests for _resolve_doc_path."""
+
+    def test_returns_none_when_no_files_exist(self):
+        doc = MagicMock()
+        doc.has_archive_version = False
+        doc.source_path = Path("/nonexistent/source.pdf")
+        result = _resolve_doc_path(doc, None)
+        self.assertIsNone(result)
+
+    def test_returns_original_file_as_fallback(self):
+        doc = MagicMock()
+        doc.has_archive_version = False
+        doc.source_path = Path("/nonexistent/source.pdf")
+
+        with tempfile.NamedTemporaryFile(suffix=".pdf") as f:
+            result = _resolve_doc_path(doc, Path(f.name))
+            self.assertEqual(result, Path(f.name))
+
+    def test_returns_none_for_none_original_file(self):
+        doc = MagicMock()
+        doc.has_archive_version = False
+        doc.source_path = Path("/nonexistent/source.pdf")
+        result = _resolve_doc_path(doc, None)
+        self.assertIsNone(result)
+
+
+class TestRunZoneExtraction(TestCase):
+    """Tests for the full extraction pipeline."""
+
+    def setUp(self):
+        self.doc_type = DocumentType.objects.create(name="Invoice")
+        self.custom_field = CustomField.objects.create(
+            name="Invoice Number",
+            data_type=CustomField.FieldDataType.STRING,
+        )
+
+    def test_skips_document_without_type(self):
+        doc = Document.objects.create(
+            title="No Type",
+            content="test",
+            mime_type="application/pdf",
+        )
+        run_zone_extraction(doc, Path("/nonexistent"))
+        self.assertEqual(CustomFieldInstance.objects.count(), 0)
+
+    def test_skips_document_without_matching_template(self):
+        other_type = DocumentType.objects.create(name="Other")
+        doc = Document.objects.create(
+            title="No Template",
+            content="test",
+            mime_type="application/pdf",
+            document_type=other_type,
+        )
+        run_zone_extraction(doc, Path("/nonexistent"))
+        self.assertEqual(CustomFieldInstance.objects.count(), 0)
+
+    def test_skips_disabled_template(self):
+        template = OcrTemplate.objects.create(
+            name="Disabled",
+            document_type=self.doc_type,
+            source_width=2480,
+            source_height=3508,
+            enabled=False,
+        )
+        OcrTemplateZone.objects.create(
+            template=template,
+            name="Zone",
+            custom_field=self.custom_field,
+            x=0,
+            y=0,
+            width=100,
+            height=50,
+        )
+
+        doc = Document.objects.create(
+            title="Test",
+            content="test",
+            mime_type="application/pdf",
+            document_type=self.doc_type,
+        )
+        run_zone_extraction(doc, Path("/nonexistent"))
+        self.assertEqual(CustomFieldInstance.objects.count(), 0)
+
+    def test_skips_template_with_no_zones(self):
+        OcrTemplate.objects.create(
+            name="Empty",
+            document_type=self.doc_type,
+            source_width=2480,
+            source_height=3508,
+            enabled=True,
+        )
+
+        doc = Document.objects.create(
+            title="Test",
+            content="test",
+            mime_type="application/pdf",
+            document_type=self.doc_type,
+        )
+
+        with tempfile.NamedTemporaryFile(suffix=".pdf") as f:
+            f.write(b"%PDF-1.4 fake")
+            f.flush()
+            run_zone_extraction(doc, Path(f.name))
+        self.assertEqual(CustomFieldInstance.objects.count(), 0)
+
+    @patch("documents.zone_ocr._process_template")
+    def test_calls_process_for_enabled_template(self, mock_process):
+        template = OcrTemplate.objects.create(
+            name="Active",
+            document_type=self.doc_type,
+            source_width=2480,
+            source_height=3508,
+            enabled=True,
+        )
+        OcrTemplateZone.objects.create(
+            template=template,
+            name="Zone",
+            custom_field=self.custom_field,
+            x=0,
+            y=0,
+            width=100,
+            height=50,
+        )
+
+        doc = Document.objects.create(
+            title="Test",
+            content="test",
+            mime_type="application/pdf",
+            document_type=self.doc_type,
+        )
+
+        with tempfile.NamedTemporaryFile(suffix=".pdf") as f:
+            f.write(b"%PDF-1.4 fake")
+            f.flush()
+            run_zone_extraction(doc, Path(f.name))
+
+        self.assertTrue(mock_process.called)
+
+    @patch("documents.zone_ocr._process_template")
+    def test_handles_process_exception_gracefully(self, mock_process):
+        """A failing template should not prevent other templates from running."""
+        mock_process.side_effect = RuntimeError("test error")
+
+        template = OcrTemplate.objects.create(
+            name="Failing",
+            document_type=self.doc_type,
+            source_width=2480,
+            source_height=3508,
+            enabled=True,
+        )
+        OcrTemplateZone.objects.create(
+            template=template,
+            name="Zone",
+            custom_field=self.custom_field,
+            x=0,
+            y=0,
+            width=100,
+            height=50,
+        )
+
+        doc = Document.objects.create(
+            title="Test",
+            content="test",
+            mime_type="application/pdf",
+            document_type=self.doc_type,
+        )
+
+        with tempfile.NamedTemporaryFile(suffix=".pdf") as f:
+            f.write(b"%PDF-1.4 fake")
+            f.flush()
+            # Should not raise
+            run_zone_extraction(doc, Path(f.name))
+
+    def test_handles_none_original_file(self):
+        """Should not crash when original_file is None."""
+        doc = Document.objects.create(
+            title="Test",
+            content="test",
+            mime_type="application/pdf",
+            document_type=self.doc_type,
+        )
+        # No template, so it exits early — but shouldn't crash on None
+        run_zone_extraction(doc, None)
+
+    @patch("documents.zone_ocr._process_template")
+    def test_multiple_templates_all_process(self, mock_process):
+        """Multiple enabled templates for the same type should all run."""
+        for i in range(3):
+            template = OcrTemplate.objects.create(
+                name=f"Template {i}",
+                document_type=self.doc_type,
+                source_width=2480,
+                source_height=3508,
+                enabled=True,
+            )
+            OcrTemplateZone.objects.create(
+                template=template,
+                name=f"Zone {i}",
+                custom_field=self.custom_field,
+                x=0,
+                y=0,
+                width=100,
+                height=50,
+            )
+
+        doc = Document.objects.create(
+            title="Test",
+            content="test",
+            mime_type="application/pdf",
+            document_type=self.doc_type,
+        )
+
+        with tempfile.NamedTemporaryFile(suffix=".pdf") as f:
+            f.write(b"%PDF-1.4 fake")
+            f.flush()
+            run_zone_extraction(doc, Path(f.name))
+
+        self.assertEqual(mock_process.call_count, 3)

src/documents/versioning.py

@@ -7,12 +7,9 @@ from typing import Any
 
 from django.db.models import F
 from django.db.models import OuterRef
-from django.db.models import Prefetch
 from django.db.models import QuerySet
 from django.db.models import Subquery
-from django.db.models import Window
 from django.db.models.functions import Coalesce
-from django.db.models.functions import RowNumber
 
 from documents.models import Document
 
@@ -49,68 +46,6 @@ def annotate_effective_content(documents: QuerySet[Document]) -> QuerySet[Docume
     )
 
 
-LATEST_VERSION_CONTENT_PREFETCH_ATTR = "_latest_version_content_prefetch"
-
-
-def latest_version_content_prefetch() -> Prefetch:
-    """
-    A Prefetch for Document.versions scoped to just the newest version's
-    content, for get_effective_content()'s fallback when no SQL annotation
-    is present.
-
-    Deliberately not merged into a metadata-only "versions" prefetch (the one
-    used for the serialized versions list): that one fetches every historical
-    version of every document, and pulling full OCR content for versions
-    nobody will read wastes DB transfer/memory at scale. This one is windowed
-    down to a single row per root, then bounded by Prefetch's own IN-list to
-    whatever page/result set it's attached to -- one cheap bulk query total,
-    not one per document and not one per version.
-    """
-    return Prefetch(
-        "versions",
-        queryset=(
-            Document.objects.filter(
-                root_document_id__isnull=False,
-                deleted_at__isnull=True,
-            )
-            .annotate(
-                rn=Window(
-                    RowNumber(),
-                    partition_by=F("root_document_id"),
-                    order_by=[
-                        F("version_index").desc(nulls_last=True),
-                        F("id").desc(),
-                    ],
-                ),
-            )
-            .filter(rn=1)
-            .only("id", "root_document_id", "content")
-        ),
-        to_attr=LATEST_VERSION_CONTENT_PREFETCH_ATTR,
-    )
-
-
-def has_prefetched_effective_content(document: Document) -> bool:
-    """
-    True if document.get_effective_content() can answer without an extra
-    per-instance query -- an SQL ``effective_content`` annotation, the lean
-    latest_version_content_prefetch(), or the metadata-only "versions"
-    prefetch is already present on the instance.
-
-    Callers that haven't set any of those up (e.g. views that build their
-    own querysets independently of DocumentViewSet.get_queryset(), like
-    TrashView or GlobalSearchView) intentionally don't pay for version-aware
-    content resolution -- see DocumentSerializer.to_representation(), which
-    uses this to decide whether to call get_effective_content() at all.
-    """
-    if hasattr(document, "effective_content"):
-        return True
-    if getattr(document, LATEST_VERSION_CONTENT_PREFETCH_ATTR, None) is not None:
-        return True
-    prefetched_cache = getattr(document, "_prefetched_objects_cache", None)
-    return isinstance(prefetched_cache, dict) and "versions" in prefetched_cache
-
-
 def sort_versions_newest_first(documents: list[Document]) -> list[Document]:
     """
     Same sorting as versions_newest_first()

src/documents/views.py

@@ -3,6 +3,7 @@ import logging
 import os
 import platform
 import re
+import subprocess
 import tempfile
 import zipfile
 from collections import defaultdict
@@ -36,6 +37,7 @@ from django.db.migrations.recorder import MigrationRecorder
 from django.db.models import Avg
 from django.db.models import Case
 from django.db.models import Count
+from django.db.models import F
 from django.db.models import IntegerField
 from django.db.models import Max
 from django.db.models import Model
@@ -136,26 +138,26 @@ from documents.filters import CustomFieldFilterSet
 from documents.filters import DocumentFilterSet
 from documents.filters import DocumentsOrderingFilter
 from documents.filters import DocumentTypeFilterSet
-from documents.filters import EffectiveContentFilter
 from documents.filters import PaperlessTaskFilterSet
 from documents.filters import PermittedObjectsFilter
 from documents.filters import ShareLinkBundleFilterSet
 from documents.filters import ShareLinkFilterSet
 from documents.filters import StoragePathFilterSet
 from documents.filters import TagFilterSet
-from documents.filters import TitleContentFilter
 from documents.mail import EmailAttachment
 from documents.mail import send_email
 from documents.matching import match_correspondents
 from documents.matching import match_document_types
 from documents.matching import match_storage_paths
 from documents.matching import match_tags
+from documents.models import OCR_SUPPORTED_FIELD_TYPES
 from documents.models import Correspondent
 from documents.models import CustomField
 from documents.models import CustomFieldInstance
 from documents.models import Document
 from documents.models import DocumentType
 from documents.models import Note
+from documents.models import OcrTemplate
 from documents.models import PaperlessTask
 from documents.models import SavedView
 from documents.models import ShareLink
@@ -180,7 +182,7 @@ from documents.permissions import has_perms_owner_aware
 from documents.permissions import has_system_status_permission
 from documents.permissions import permitted_document_ids
 from documents.permissions import permitted_object_ids
-from documents.permissions import set_permissions_for_objects
+from documents.permissions import set_permissions_for_object
 from documents.permissions import user_is_unrestricted
 from documents.plugins.date_parsing import get_date_parser
 from documents.schema import generate_object_with_permissions_schema
@@ -202,6 +204,7 @@ from documents.serialisers import EmailSerializer
 from documents.serialisers import MergeDocumentsAsVersionsSerializer
 from documents.serialisers import MergeDocumentsSerializer
 from documents.serialisers import NotesSerializer
+from documents.serialisers import OcrTemplateSerializer
 from documents.serialisers import PostDocumentSerializer
 from documents.serialisers import RemovePasswordDocumentsSerializer
 from documents.serialisers import ReprocessDocumentsSerializer
@@ -237,7 +240,6 @@ from documents.versioning import annotate_effective_content
 from documents.versioning import get_latest_version_for_root
 from documents.versioning import get_request_version_param
 from documents.versioning import get_root_document
-from documents.versioning import latest_version_content_prefetch
 from documents.versioning import resolve_requested_version_for_root
 from documents.versioning import versions_newest_first
 from paperless import version
@@ -1086,59 +1088,12 @@ class DocumentViewSet(
             ],
         }
 
-    @classmethod
-    def _content_filter_params(cls) -> tuple[str, ...]:
-        """
-        Query params whose filtering needs effective_content evaluated in SQL
-        against every candidate row -- see
-        _needs_effective_content_annotation(). Derived rather than
-        hand-maintained so a new content-filtering param counts automatically.
-        """
-        params = [
-            name
-            for name, f in DocumentFilterSet.declared_filters.items()
-            if isinstance(f, (TitleContentFilter, EffectiveContentFilter))
-        ]
-        if "effective_content" in cls.search_fields:
-            params.append(SearchFilter().search_param)
-        return tuple(params)
-
-    def _needs_effective_content_annotation(self) -> bool:
-        # effective_content is a per-row correlated subquery resolving each
-        # document's latest version. Filtering *on* it forces the database to
-        # evaluate it for every candidate row before reaching the LIMIT, which
-        # the root_document_id self-join makes pathological on MariaDB
-        # specifically once real candidate counts get large; otherwise the
-        # "versions" prefetch + Document.get_effective_content() resolves only
-        # the page that survives pagination. Every param here is deprecated in
-        # favor of the Tantivy-backed search endpoint (see filters.py's
-        # TitleContentFilter/EffectiveContentFilter docs), so pay that cost
-        # only when one is actually used. Blank values don't count, matching
-        # how those filters themselves no-op on them -- an empty `?search=`
-        # applies no predicate.
-        params = self.request.query_params
-        return any(
-            params.get(param, "").strip() for param in self._content_filter_params()
-        )
-
-    def _requested_fields(self) -> list[str] | None:
-        # The sparse-fieldset `fields` param, as DynamicFieldsModelSerializer
-        # wants it: None means "no restriction, serialize everything", which
-        # a blank value means too. get_queryset() and get_serializer() both
-        # branch on this, and they have to read it identically -- a queryset
-        # that skips the content prefetch for a response that still
-        # serializes content reintroduces get_effective_content()'s
-        # per-instance fallback.
-        fields_param = self.request.query_params.get("fields")
-        return fields_param.split(",") if fields_param else None
-
-    def _needs_effective_content_prefetch(self) -> bool:
-        # The prefetch spares get_effective_content() a per-instance fallback
-        # query, but only earns itself when content can reach the response.
-        fields = self._requested_fields()
-        return fields is None or "content" in fields
-
     def get_queryset(self):
+        latest_version_content = Subquery(
+            versions_newest_first(
+                Document.objects.filter(root_document=OuterRef("pk")),
+            ).values("content")[:1],
+        )
         # A correlated subquery avoids the LEFT JOIN + Count() this used to
         # be, which forced a GROUP BY aggregate over every matching document
         # before the query could even be sorted or limited.
@@ -1158,43 +1113,40 @@ class DocumentViewSet(
         # ObjectFilter.filter(). A blanket .distinct() here forces the
         # database to fully sort and dedupe every visible document before
         # it can apply LIMIT, which is disastrous at scale.
-        prefetches = [
-            Prefetch(
-                "versions",
-                queryset=Document.objects.only(
-                    "id",
-                    "added",
-                    "checksum",
-                    "version_label",
-                    "root_document_id",
-                    "version_index",
-                ),
-            ),
-            "tags",
-            Prefetch(
-                "custom_fields",
-                queryset=CustomFieldInstance.objects.select_related("field"),
-            ),
-            # NotesSerializer nests the author, this avoids query per note
-            Prefetch("notes", queryset=Note.objects.select_related("user")),
-        ]
-        if self._needs_effective_content_prefetch():
-            prefetches.append(latest_version_content_prefetch())
-        queryset = (
+        return (
             Document.objects.filter(root_document__isnull=True)
             .order_by("-created", "-id")
+            .annotate(effective_content=Coalesce(latest_version_content, F("content")))
             .annotate(num_notes=Coalesce(note_count, 0))
             .select_related("correspondent", "storage_path", "document_type", "owner")
-            .prefetch_related(*prefetches)
+            .prefetch_related(
+                Prefetch(
+                    "versions",
+                    queryset=Document.objects.only(
+                        "id",
+                        "added",
+                        "checksum",
+                        "version_label",
+                        "root_document_id",
+                        "version_index",
+                    ),
+                ),
+                "tags",
+                Prefetch(
+                    "custom_fields",
+                    queryset=CustomFieldInstance.objects.select_related("field"),
+                ),
+                # NotesSerializer nests the author, this avoids query per note
+                Prefetch("notes", queryset=Note.objects.select_related("user")),
+            )
         )
-        if self._needs_effective_content_annotation():
-            queryset = annotate_effective_content(queryset)
-        return queryset
 
     def get_serializer(self, *args, **kwargs):
+        fields_param = self.request.query_params.get("fields", None)
+        fields = fields_param.split(",") if fields_param else None
         truncate_content = self.request.query_params.get("truncate_content", "False")
         kwargs.setdefault("context", self.get_serializer_context())
-        kwargs.setdefault("fields", self._requested_fields())
+        kwargs.setdefault("fields", fields)
         kwargs.setdefault("truncate_content", truncate_content.lower() in ["true", "1"])
         try:
             full_perms = get_boolean(
@@ -2220,6 +2172,73 @@ class DocumentViewSet(
             },
         ),
     )
+    @action(methods=["post"], detail=True, url_path="run-zone-ocr")
+    def run_zone_ocr(self, request, pk=None):
+        """Run zone-based OCR extraction on this document."""
+        try:
+            document = Document.objects.get(pk=pk)
+        except Document.DoesNotExist:
+            raise Http404
+
+        if not document.document_type_id:
+            return Response(
+                {"error": "Document has no type assigned"},
+                status=status.HTTP_400_BAD_REQUEST,
+            )
+
+        templates = OcrTemplate.objects.filter(
+            document_type_id=document.document_type_id,
+            enabled=True,
+        )
+        if not templates.exists():
+            return Response(
+                {"error": "No OCR templates found for this document type"},
+                status=status.HTTP_404_NOT_FOUND,
+            )
+
+        doc_path = document.archive_path or document.source_path
+        if not doc_path or not Path(doc_path).is_file():
+            return Response(
+                {"error": "Document file not found"},
+                status=status.HTTP_404_NOT_FOUND,
+            )
+
+        from documents.zone_ocr import run_zone_extraction
+
+        run_zone_extraction(document, None)
+
+        # Collect results
+        results = []
+        builtin_labels = {"title": "Title", "asn": "ASN", "created": "Created"}
+        for template in templates.prefetch_related("zones", "zones__custom_field"):
+            for zone in template.zones.all():
+                target = getattr(zone, "target", None) or "custom_field"
+                if target == "custom_field" and zone.custom_field_id:
+                    cf_instance = document.custom_fields.filter(
+                        field=zone.custom_field,
+                    ).first()
+                    field_name = zone.custom_field.name
+                    value = cf_instance.value if cf_instance else None
+                else:
+                    field_name = builtin_labels.get(target, target)
+                    value = {
+                        "title": document.title,
+                        "asn": document.archive_serial_number,
+                        "created": document.created.isoformat()
+                        if document.created
+                        else None,
+                    }.get(target)
+                results.append(
+                    {
+                        "template": template.name,
+                        "zone": zone.name,
+                        "custom_field": field_name,
+                        "value": value,
+                    },
+                )
+
+        return Response({"results": results})
+
     @action(
         methods=["delete"],
         detail=True,
@@ -2380,6 +2399,7 @@ class ChatStreamingView(GenericAPIView[Any]):
     serializer_class = ChatStreamingSerializer
 
     def post(self, request, *args, **kwargs):
+        request.compress_exempt = True
         ai_config = AIConfig()
         if not ai_config.ai_enabled:
             return HttpResponseBadRequest("AI is required for this feature")
@@ -5018,12 +5038,12 @@ class BulkEditObjectsView(PassUserMixin):
                     qs_owner_update.update(owner=owner)
 
                 if "permissions" in serializer.validated_data:
-                    set_permissions_for_objects(
-                        permissions=permissions,
-                        model=object_class,
-                        pks=qs.values_list("pk", flat=True),
-                        merge=merge,
-                    )
+                    for obj in qs:
+                        set_permissions_for_object(
+                            permissions=permissions,
+                            object=obj,
+                            merge=merge,
+                        )
 
             except Exception as e:
                 logger.warning(
@@ -5593,3 +5613,224 @@ def serve_logo(request: HttpRequest, filename: str | None = None) -> FileRespons
         filename=logo_name,
         as_attachment=True,
     )
+
+
+class OcrTemplateViewSet(ModelViewSet):
+    """CRUD for OCR templates with zone definitions."""
+
+    queryset = (
+        OcrTemplate.objects.all()
+        .prefetch_related(
+            "zones",
+            "zones__custom_field",
+        )
+        .order_by("name")
+    )
+    serializer_class = OcrTemplateSerializer
+    permission_classes = (IsAuthenticated, PaperlessObjectPermissions)
+    pagination_class = StandardPagination
+
+    @action(
+        detail=False,
+        methods=["get"],
+        url_path=r"document-page-image/(?P<doc_id>[0-9]+)/(?P<page>[0-9]+)",
+    )
+    def document_page_image(self, request, doc_id=None, page=None):
+        """Render a specific page of a document as a PNG image.
+
+        Used by the frontend template editor to display document pages
+        as images that users can draw zones on.
+        """
+        try:
+            document = Document.objects.get(pk=doc_id)
+        except Document.DoesNotExist:
+            raise Http404("Document not found")
+
+        page_num = int(page)
+
+        # Validate page number
+        if document.page_count and page_num >= document.page_count:
+            raise Http404(
+                f"Page {page_num} out of range (document has {document.page_count} pages)",
+            )
+
+        doc_path = document.archive_path or document.source_path
+        if not doc_path or not Path(doc_path).is_file():
+            raise Http404("Document file not found")
+
+        # Check if document is an image (single page, no PDF rendering needed)
+        if document.mime_type and document.mime_type.startswith("image/"):
+            content = Path(doc_path).read_bytes()
+            return HttpResponse(content, content_type=document.mime_type)
+
+        with tempfile.TemporaryDirectory(dir=settings.SCRATCH_DIR) as tmp_dir:
+            output_prefix = Path(tmp_dir) / "page"
+            try:
+                subprocess.run(
+                    [
+                        "pdftoppm",
+                        "-png",
+                        "-r",
+                        "150",  # Lower DPI for preview
+                        "-f",
+                        str(page_num + 1),
+                        "-l",
+                        str(page_num + 1),
+                        str(doc_path),
+                        str(output_prefix),
+                    ],
+                    check=True,
+                    capture_output=True,
+                    timeout=30,
+                )
+            except subprocess.CalledProcessError as e:
+                raise Http404(
+                    f"Failed to render page: {e.stderr.decode(errors='replace')[:200]}",
+                )
+            except FileNotFoundError:
+                raise Http404("pdftoppm not available - is poppler-utils installed?")
+
+            rendered = sorted(Path(tmp_dir).glob("page-*.png"))
+            if not rendered:
+                raise Http404("No rendered page found")
+
+            content = rendered[0].read_bytes()
+
+        return HttpResponse(content, content_type="image/png")
+
+    @action(detail=False, methods=["post"], url_path="test-zone")
+    def test_zone(self, request):
+        """Run OCR on a single ad-hoc zone of a document and return what it
+        yields: the raw OCR text, the transformed value, and whether the
+        validation regex matches. Non-destructive - writes nothing. Used by the
+        editor's per-zone test so a user can tune the zone/regex before saving.
+
+        Accepts: {"document": <id>, "zone": {x, y, width, height, page,
+        ocr_language, transform, validation_regex, zone_source_width,
+        zone_source_height}}.
+        """
+        from documents.models import OcrTemplateZone
+        from documents.zone_ocr import extract_zone_preview
+
+        zone_data = request.data.get("zone") or {}
+
+        try:
+            document = Document.objects.get(pk=request.data.get("document"))
+        except (Document.DoesNotExist, ValueError, TypeError):
+            return Response(
+                {"error": "Document not found"},
+                status=status.HTTP_404_NOT_FOUND,
+            )
+
+        doc_path = document.archive_path or document.source_path
+        if not doc_path or not Path(doc_path).is_file():
+            return Response(
+                {"error": "Document file not found"},
+                status=status.HTTP_404_NOT_FOUND,
+            )
+
+        try:
+            zone = OcrTemplateZone(
+                name=zone_data.get("name") or "test",
+                x=int(zone_data.get("x", 0)),
+                y=int(zone_data.get("y", 0)),
+                width=int(zone_data.get("width", 0)),
+                height=int(zone_data.get("height", 0)),
+                page=zone_data.get("page"),
+                ocr_language=zone_data.get("ocr_language") or "eng",
+                transform=zone_data.get("transform") or "strip",
+                date_format=zone_data.get("date_format") or "",
+                validation_regex=zone_data.get("validation_regex") or "",
+            )
+        except (ValueError, TypeError):
+            return Response(
+                {"error": "Invalid zone definition"},
+                status=status.HTTP_400_BAD_REQUEST,
+            )
+
+        if zone.width < 2 or zone.height < 2:
+            return Response(
+                {"error": "Zone is too small to test"},
+                status=status.HTTP_400_BAD_REQUEST,
+            )
+
+        result = extract_zone_preview(
+            Path(doc_path),
+            zone,
+            int(zone_data.get("zone_source_width") or 0),
+            int(zone_data.get("zone_source_height") or 0),
+            document.page_count,
+        )
+
+        regex_match = None
+        if zone.validation_regex and result.get("value") is not None:
+            try:
+                regex_match = (
+                    re.fullmatch(zone.validation_regex, result["value"]) is not None
+                )
+            except re.error:
+                regex_match = None
+
+        return Response(
+            {
+                "raw_text": result.get("raw_text"),
+                "value": result.get("value"),
+                "regex": zone.validation_regex,
+                "regex_match": regex_match,
+            },
+        )
+
+    @action(detail=False, methods=["post"], url_path="quick-create-field")
+    def quick_create_field(self, request):
+        """Create a custom field inline from the template editor.
+
+        Accepts: {"name": "Invoice Number", "data_type": "string"}
+        Returns the created field so the frontend can immediately use it.
+        """
+        name = request.data.get("name", "").strip()
+        data_type = request.data.get("data_type", "").strip()
+
+        if not name:
+            return Response(
+                {"error": "Field name is required"},
+                status=status.HTTP_400_BAD_REQUEST,
+            )
+
+        if data_type not in OCR_SUPPORTED_FIELD_TYPES:
+            return Response(
+                {
+                    "error": f"Unsupported data type '{data_type}'. "
+                    f"Supported: {', '.join(sorted(OCR_SUPPORTED_FIELD_TYPES))}",
+                },
+                status=status.HTTP_400_BAD_REQUEST,
+            )
+
+        # Check if field already exists
+        existing = CustomField.objects.filter(name=name).first()
+        if existing:
+            return Response(
+                {
+                    "id": existing.pk,
+                    "name": existing.name,
+                    "data_type": existing.data_type,
+                    "created": False,
+                },
+            )
+
+        # Check user has permission to create custom fields
+        if not request.user.has_perm("documents.add_customfield"):
+            return Response(
+                {"error": "You don't have permission to create custom fields"},
+                status=status.HTTP_403_FORBIDDEN,
+            )
+
+        field = CustomField.objects.create(name=name, data_type=data_type)
+        return Response(
+            {
+                "id": field.pk,
+                "name": field.name,
+                "data_type": field.data_type,
+                "created": True,
+            },
+            status=status.HTTP_201_CREATED,
+        )

src/documents/zone_ocr.py

@@ -0,0 +1,757 @@
+"""
+Zone-based OCR extraction engine.
+
+After a document is consumed, this module checks if the document's type has
+an active OCR template. If so, it renders the relevant pages as images,
+crops each zone, runs Tesseract OCR on the crop, applies transforms,
+and writes the results to the mapped custom fields.
+"""
+
+from __future__ import annotations
+
+import logging
+import re
+import string
+import subprocess
+import tempfile
+from datetime import date
+from datetime import datetime
+from pathlib import Path
+
+from django.conf import settings
+from PIL import Image
+
+from documents.models import CustomField
+from documents.models import CustomFieldInstance
+from documents.models import Document
+from documents.models import OcrTemplate
+from documents.models import OcrTemplateZone
+
+logger = logging.getLogger("paperless.zone_ocr")
+
+
+def run_zone_extraction(
+    document: Document,
+    original_file: Path | None,
+) -> None:
+    """
+    Run zone-based OCR extraction for a document if its type has an active template.
+    Called from the document_consumption_finished signal handler.
+    """
+    if not document.document_type_id:
+        return
+
+    templates = OcrTemplate.objects.filter(
+        document_type_id=document.document_type_id,
+        enabled=True,
+    ).prefetch_related("zones", "zones__custom_field")
+
+    if not templates.exists():
+        return
+
+    # Resolve the document file: prefer archive (PDF/A), then source, then signal arg
+    doc_path = _resolve_doc_path(document, original_file)
+    if doc_path is None:
+        logger.warning(
+            "Zone OCR: no accessible file for document %d",
+            document.pk,
+        )
+        return
+
+    for template in templates:
+        zones = list(template.zones.all())
+        if not zones:
+            continue
+
+        logger.info(
+            "Zone OCR: processing template '%s' for document %d (%d zones)",
+            template.name,
+            document.pk,
+            len(zones),
+        )
+
+        try:
+            _process_template(document, doc_path, template, zones)
+        except Exception:
+            logger.exception(
+                "Zone OCR: error processing template '%s' for document %d",
+                template.name,
+                document.pk,
+            )
+
+
+def _resolve_doc_path(
+    document: Document,
+    original_file: Path | None,
+) -> Path | None:
+    """Find an accessible file for the document."""
+    candidates = []
+    if document.has_archive_version:
+        candidates.append(document.archive_path)
+    candidates.append(document.source_path)
+    if original_file is not None:
+        candidates.append(original_file)
+
+    for path in candidates:
+        if path is not None and Path(path).is_file():
+            return Path(path)
+    return None
+
+
+def _resolve_page_idx(page_value, page_count) -> int:
+    """Resolve a 1-indexed page (1 = first, -1 = last) to a 0-indexed image
+    index. A blank page_value defaults to the first page."""
+    if page_value is None:
+        return 0
+    if page_value == -1:
+        return (page_count - 1) if page_count else 0
+    if page_value >= 1:
+        return page_value - 1
+    return 0
+
+
+def _process_template(
+    document: Document,
+    doc_path: Path,
+    template: OcrTemplate,
+    zones: list[OcrTemplateZone],
+) -> None:
+    """Process all zones in a template against a document.
+
+    Each zone is OCR'd independently, then zones are grouped by their target
+    field and each field is written exactly once. When several zones share a
+    field, their values are combined via the template's per-field format string
+    (or joined in order if none is set) — this avoids the zones overwriting each
+    other's value.
+    """
+    pages_needed: set[int] = {
+        _resolve_page_idx(zone.page, document.page_count) for zone in zones
+    }
+
+    with tempfile.TemporaryDirectory(dir=settings.SCRATCH_DIR) as tmp_dir:
+        tmp_path = Path(tmp_dir)
+
+        page_images = _render_pages(
+            doc_path,
+            pages_needed,
+            tmp_path,
+            document.page_count,
+        )
+
+        # Pass 1: OCR every zone into a value (or None if it failed/was rejected).
+        zone_values: dict[int, str | None] = {}
+        for zone in zones:
+            page_idx = _resolve_page_idx(zone.page, document.page_count)
+
+            if page_idx not in page_images:
+                logger.warning(
+                    "Zone OCR: page %d not available for zone '%s'",
+                    page_idx,
+                    zone.name,
+                )
+                continue
+
+            src_w = zone.zone_source_width or template.source_width
+            src_h = zone.zone_source_height or template.source_height
+
+            extracted = _extract_zone(
+                page_images[page_idx],
+                zone,
+                src_w,
+                src_h,
+                tmp_path,
+            )
+
+            if (
+                extracted is not None
+                and zone.validation_regex
+                and not re.fullmatch(zone.validation_regex, extracted)
+            ):
+                logger.info(
+                    "Zone OCR: '%s' value %r rejected by regex '%s'",
+                    zone.name,
+                    extracted[:100],
+                    zone.validation_regex,
+                )
+                extracted = None
+
+            zone_values[id(zone)] = extracted
+
+    # Pass 2: group zones by target field and write each field once.
+    grouped: dict[str, list[OcrTemplateZone]] = {}
+    for zone in zones:
+        grouped.setdefault(_field_key(zone), []).append(zone)
+
+    combine_formats = template.combine_formats or {}
+    for key, field_zones in grouped.items():
+        value = _combine_field_value(
+            combine_formats.get(key, ""),
+            field_zones,
+            zone_values,
+        )
+        if not value:
+            continue
+
+        target_zone = field_zones[0]
+        _write_zone_value(document, target_zone, value)
+        logger.info(
+            "Zone OCR: %s = %r (from %d zone(s))",
+            _zone_target_label(target_zone),
+            value[:100] if len(value) > 100 else value,
+            len(field_zones),
+        )
+
+
+def _field_key(zone: OcrTemplateZone) -> str:
+    """Identify a zone's target field. Custom fields key by id, built-in targets
+    by their name. Matches the key used in OcrTemplate.combine_formats and on the
+    frontend field select."""
+    target = getattr(zone, "target", None) or "custom_field"
+    if target == "custom_field" and zone.custom_field_id:
+        return str(zone.custom_field_id)
+    return target
+
+
+def _combine_field_value(
+    fmt: str,
+    field_zones: list[OcrTemplateZone],
+    zone_values: dict[int, str | None],
+) -> str:
+    """Combine the OCR values of all zones targeting one field.
+
+    With a format string, `{Zone Name}` tokens are replaced by that zone's value
+    and literal text is kept; separators left dangling by an empty token are
+    cleaned up. Without a format, the zone values are joined in order by a space.
+    """
+    values = {z.name: (zone_values.get(id(z)) or "") for z in field_zones}
+
+    if not fmt:
+        parts = [zone_values.get(id(z)) or "" for z in field_zones]
+        return " ".join(p for p in parts if p).strip()
+
+    def _replace(match: re.Match) -> str:
+        return values.get(match.group(1).strip(), "")
+
+    combined = re.sub(r"\{([^{}]+)\}", _replace, fmt)
+    # Tidy up separators an empty token may have left behind.
+    combined = re.sub(r"\s{2,}", " ", combined)
+    combined = re.sub(r"([^\w\s])\s*\1+", r"\1", combined)
+    return combined.strip().strip("-/.,;:| \t")
+
+
+def _render_pages(
+    doc_path: Path,
+    pages: set[int],
+    tmp_dir: Path,
+    page_count: int | None,
+) -> dict[int, Path]:
+    """Render specific PDF pages as PNG images using pdftoppm (poppler-utils)."""
+    result: dict[int, Path] = {}
+    mime = _detect_mime(doc_path)
+
+    if mime and mime.startswith("image/"):
+        # Single-image document — use it directly as page 0.
+        result[0] = doc_path
+        return result
+
+    # Callers pass already-resolved 0-indexed page numbers (see _resolve_page_idx).
+    for actual_page in pages:
+        if actual_page < 0:
+            logger.warning("Zone OCR: invalid page index %d", actual_page)
+            continue
+
+        output_prefix = tmp_dir / f"page_{actual_page}"
+        try:
+            subprocess.run(
+                [
+                    "pdftoppm",
+                    "-png",
+                    "-r",
+                    "300",
+                    "-f",
+                    str(actual_page + 1),  # pdftoppm is 1-indexed
+                    "-l",
+                    str(actual_page + 1),
+                    str(doc_path),
+                    str(output_prefix),
+                ],
+                check=True,
+                capture_output=True,
+                timeout=60,
+            )
+        except subprocess.TimeoutExpired:
+            logger.error("Zone OCR: pdftoppm timed out for page %d", actual_page)
+            continue
+        except subprocess.CalledProcessError as e:
+            logger.error(
+                "Zone OCR: pdftoppm failed for page %d: %s",
+                actual_page,
+                e.stderr.decode(errors="replace") if e.stderr else str(e),
+            )
+            continue
+        except FileNotFoundError:
+            logger.error("Zone OCR: pdftoppm not found — is poppler-utils installed?")
+            return result  # No point trying other pages
+
+        # pdftoppm names output as prefix-NNNN.png
+        rendered = sorted(tmp_dir.glob(f"page_{actual_page}-*.png"))
+        if rendered:
+            result[actual_page] = rendered[0]
+
+    return result
+
+
+def _crop_zone(
+    page_img: Path,
+    zone: OcrTemplateZone,
+    source_width: int,
+    source_height: int,
+    tmp_dir: Path,
+) -> Image.Image | None:
+    """Crop a zone from the page image and return the PIL Image."""
+    try:
+        with Image.open(page_img) as img:
+            img_width, img_height = img.size
+
+            scale_x = img_width / source_width
+            scale_y = img_height / source_height
+
+            crop_left = int(zone.x * scale_x)
+            crop_top = int(zone.y * scale_y)
+            crop_right = int((zone.x + zone.width) * scale_x)
+            crop_bottom = int((zone.y + zone.height) * scale_y)
+
+            # Clamp to the image so an oversized zone can't crop out of bounds.
+            crop_left = max(0, min(crop_left, img_width))
+            crop_top = max(0, min(crop_top, img_height))
+            crop_right = max(crop_left + 1, min(crop_right, img_width))
+            crop_bottom = max(crop_top + 1, min(crop_bottom, img_height))
+
+            if crop_right - crop_left < 2 or crop_bottom - crop_top < 2:
+                logger.warning("Zone OCR: crop too small for zone '%s'", zone.name)
+                return None
+
+            return img.crop((crop_left, crop_top, crop_right, crop_bottom)).copy()
+    except Exception:
+        logger.exception("Zone OCR: crop failed for zone '%s'", zone.name)
+        return None
+
+
+def _read_barcode(cropped: Image.Image, zone_name: str) -> str | None:
+    """Read QR/barcode from a cropped image using zxingcpp."""
+    try:
+        import zxingcpp
+
+        results = zxingcpp.read_barcodes(cropped)
+        if results:
+            text = results[0].text
+            logger.debug(
+                "Zone OCR: barcode found in zone '%s': %s",
+                zone_name,
+                text[:100],
+            )
+            return text
+        logger.debug("Zone OCR: no barcode found in zone '%s'", zone_name)
+        return None
+    except ImportError:
+        logger.error("Zone OCR: zxingcpp not available — install zxing-cpp")
+        return None
+    except Exception:
+        logger.exception("Zone OCR: barcode read failed for zone '%s'", zone_name)
+        return None
+
+
+def _ocr_text(cropped: Image.Image, zone: OcrTemplateZone, tmp_dir: Path) -> str | None:
+    """OCR a cropped image with Tesseract."""
+    crop_path = tmp_dir / f"zone_{zone.pk}.png"
+    cropped.save(crop_path)
+
+    try:
+        proc = subprocess.run(
+            [
+                "tesseract",
+                str(crop_path),
+                "stdout",
+                "-l",
+                zone.ocr_language,
+                "--psm",
+                "6",  # Assume uniform block of text
+            ],
+            capture_output=True,
+            text=True,
+            timeout=30,
+            check=True,
+        )
+        return proc.stdout.strip() or None
+    except subprocess.TimeoutExpired:
+        logger.error("Zone OCR: Tesseract timed out for zone '%s'", zone.name)
+        return None
+    except subprocess.CalledProcessError as e:
+        logger.error(
+            "Zone OCR: Tesseract failed for zone '%s': %s",
+            zone.name,
+            e.stderr[:200] if e.stderr else str(e),
+        )
+        return None
+    except FileNotFoundError:
+        logger.error("Zone OCR: Tesseract not found — is tesseract-ocr installed?")
+        return None
+
+
+def _extract_zone(
+    page_img: Path,
+    zone: OcrTemplateZone,
+    source_width: int,
+    source_height: int,
+    tmp_dir: Path,
+) -> str | None:
+    """Crop a zone from the page image and extract text via OCR or barcode reader."""
+    cropped = _crop_zone(page_img, zone, source_width, source_height, tmp_dir)
+    if cropped is None:
+        return None
+
+    # QR/barcode zones skip Tesseract entirely
+    if zone.transform == "qr_code":
+        text = _read_barcode(cropped, zone.name)
+        if not text:
+            return None
+        return _apply_transform(
+            text,
+            zone.transform,
+            getattr(zone, "date_format", "") or "",
+        )
+
+    text = _ocr_text(cropped, zone, tmp_dir)
+    if not text:
+        return None
+
+    return _apply_transform(
+        text,
+        zone.transform,
+        getattr(zone, "date_format", "") or "",
+    )
+
+
+def extract_zone_preview(
+    doc_path: Path,
+    zone: OcrTemplateZone,
+    source_width: int,
+    source_height: int,
+    page_count: int | None,
+) -> dict:
+    """Non-destructive single-zone extraction for the editor's per-zone test.
+
+    Renders the zone's page, crops it, runs OCR (or the barcode reader) and
+    applies the transform — WITHOUT writing any custom field. Returns the raw
+    OCR text and the transformed value so the user can see what the zone yields
+    (and tune the validation regex) before saving.
+    """
+    # zone.page is 1-indexed (1 = first, -1 = last); resolve to a 0-indexed
+    # image index exactly like the production extraction path does.
+    page_idx = _resolve_page_idx(zone.page, page_count)
+    with tempfile.TemporaryDirectory(dir=settings.SCRATCH_DIR) as tmp_dir:
+        tmp_path = Path(tmp_dir)
+        page_images = _render_pages(doc_path, {page_idx}, tmp_path, page_count)
+        if page_idx not in page_images:
+            return {"raw_text": None, "value": None}
+
+        if not source_width or not source_height:
+            with Image.open(page_images[page_idx]) as im:
+                source_width, source_height = im.size
+
+        cropped = _crop_zone(
+            page_images[page_idx],
+            zone,
+            source_width,
+            source_height,
+            tmp_path,
+        )
+        if cropped is None:
+            return {"raw_text": None, "value": None}
+
+        if zone.transform == "qr_code":
+            raw_text = _read_barcode(cropped, zone.name)
+        else:
+            raw_text = _ocr_text(cropped, zone, tmp_path)
+
+        value = (
+            _apply_transform(
+                raw_text,
+                zone.transform,
+                getattr(zone, "date_format", "") or "",
+            )
+            if raw_text
+            else None
+        )
+        return {"raw_text": raw_text, "value": value}
+
+
+def _parse_date(text: str, fmt: str) -> str:
+    """Parse a date from OCR text. With a Python strptime `fmt`, try that first;
+    otherwise (or on failure) fall back to dateparser auto-detection. Returns an
+    ISO date string, or the original text if nothing parses."""
+    text = text.strip()
+    if not text:
+        return text
+    if fmt:
+        try:
+            return datetime.strptime(text, fmt).date().isoformat()
+        except ValueError:
+            pass
+    try:
+        import dateparser
+
+        parsed = dateparser.parse(
+            text,
+            settings={
+                "PREFER_DAY_OF_MONTH": "first",
+                "RETURN_AS_TIMEZONE_AWARE": False,
+            },
+        )
+        if parsed:
+            return parsed.date().isoformat()
+    except Exception:
+        logger.debug("Zone OCR: dateparser failed for %r", text[:50])
+    return text
+
+
+def _apply_transform(text: str, transform: str, date_format: str = "") -> str:
+    """Apply post-processing transform to extracted text."""
+    text = text.strip()
+    if not text:
+        return text
+
+    if transform in ("strip", "none"):
+        return text
+    elif transform == "date":
+        return _parse_date(text, date_format)
+    elif transform == "uppercase":
+        return text.upper()
+    elif transform == "lowercase":
+        return text.lower()
+    elif transform == "numeric":
+        result = re.sub(r"[^\d.,\-]", "", text)
+        return result if result else text
+    elif transform == "strip_punctuation":
+        return text.strip(string.punctuation + " \t\r\n")
+    elif transform == "qr_code":
+        # Barcode/QR content as read by _read_barcode.
+        return text
+    return text
+
+
+def _zone_target_label(zone: OcrTemplateZone) -> str:
+    """Human label of a zone's write target (for logging)."""
+    target = getattr(zone, "target", None) or "custom_field"
+    if target == "custom_field":
+        return zone.custom_field.name if zone.custom_field_id else "(no field)"
+    return {"title": "Title", "asn": "ASN", "created": "Created"}.get(target, target)
+
+
+def _parse_created_datetime(value: str):
+    """Parse an extracted value into a tz-aware datetime for document.created.
+
+    Prefers an ISO date (the zone should use a date transform); falls back to
+    dateparser. Returns None if no date can be parsed.
+    """
+    from django.utils import timezone as djtz
+
+    m = re.search(r"(\d{4})-(\d{2})-(\d{2})", value)
+    if m:
+        try:
+            dt = datetime(int(m[1]), int(m[2]), int(m[3]))
+            return djtz.make_aware(dt) if djtz.is_naive(dt) else dt
+        except ValueError:
+            pass
+    try:
+        import dateparser
+
+        parsed = dateparser.parse(
+            value,
+            settings={"RETURN_AS_TIMEZONE_AWARE": False},
+        )
+        if parsed:
+            return djtz.make_aware(parsed) if djtz.is_naive(parsed) else parsed
+    except Exception:
+        logger.debug("Zone OCR: dateparser failed for created value %r", value[:50])
+    return None
+
+
+def _write_zone_value(
+    document: Document,
+    zone: OcrTemplateZone,
+    value: str,
+) -> None:
+    """Write an extracted value to the zone's target — a custom field, or a
+    built-in document field (title / archive_serial_number / created)."""
+    target = getattr(zone, "target", None) or "custom_field"
+
+    if target == "custom_field":
+        if zone.custom_field_id:
+            _write_custom_field(document, zone.custom_field, value)
+        else:
+            logger.debug("Zone OCR: zone '%s' has no custom field set", zone.name)
+        return
+
+    if target == "title":
+        document.title = value[:128]
+        document.save(update_fields=["title"])
+    elif target == "asn":
+        digits = re.sub(r"[^\d]", "", value)
+        if not digits:
+            logger.debug(
+                "Zone OCR: ASN zone '%s' produced no digits (%r)",
+                zone.name,
+                value[:50],
+            )
+            return
+        document.archive_serial_number = int(digits)
+        document.save(update_fields=["archive_serial_number"])
+    elif target == "created":
+        parsed = _parse_created_datetime(value)
+        if parsed is None:
+            logger.debug(
+                "Zone OCR: created zone '%s' could not parse a date (%r)",
+                zone.name,
+                value[:50],
+            )
+            return
+        document.created = parsed
+        document.save(update_fields=["created"])
+
+
+def _write_custom_field(
+    document: Document,
+    custom_field: CustomField,
+    value: str,
+) -> None:
+    """Write an extracted value to a document's custom field."""
+    typed_value = _convert_value(value, custom_field.data_type)
+    if typed_value is None:
+        logger.debug(
+            "Zone OCR: skipping custom field '%s' — value conversion returned None",
+            custom_field.name,
+        )
+        return
+
+    value_field_name = CustomFieldInstance.get_value_field_name(custom_field.data_type)
+
+    CustomFieldInstance.objects.update_or_create(
+        document=document,
+        field=custom_field,
+        defaults={value_field_name: typed_value},
+    )
+
+
+def _convert_value(value: str, data_type: str) -> object | None:
+    """Convert an extracted OCR string to the appropriate type for the custom field."""
+    if not value:
+        return None
+
+    try:
+        if data_type in (
+            CustomField.FieldDataType.STRING,
+            CustomField.FieldDataType.URL,
+        ):
+            return value[:128]
+
+        elif data_type == CustomField.FieldDataType.LONG_TEXT:
+            return value
+
+        elif data_type == CustomField.FieldDataType.INT:
+            digits = re.sub(r"[^\d\-]", "", value)
+            # Handle edge case: only dashes or empty
+            digits = digits.lstrip("-") or ""
+            if not digits:
+                return None
+            # Restore leading minus if original had one
+            if value.strip().startswith("-"):
+                digits = "-" + digits
+            return int(digits)
+
+        elif data_type == CustomField.FieldDataType.FLOAT:
+            # Handle European format: 1.234,56 → 1234.56
+            cleaned = re.sub(r"[^\d.,\-]", "", value)
+            if not cleaned or cleaned in (".", ",", "-"):
+                return None
+            # If both . and , present, the last one is the decimal separator
+            if "," in cleaned and "." in cleaned:
+                if cleaned.rindex(",") > cleaned.rindex("."):
+                    # European: 1.234,56
+                    cleaned = cleaned.replace(".", "").replace(",", ".")
+                else:
+                    # US: 1,234.56
+                    cleaned = cleaned.replace(",", "")
+            elif "," in cleaned:
+                # Only comma — treat as decimal separator
+                cleaned = cleaned.replace(",", ".")
+            return float(cleaned)
+
+        elif data_type == CustomField.FieldDataType.DATE:
+            match = re.search(r"(\d{4})-(\d{2})-(\d{2})", value)
+            if match:
+                y, m, d = match.groups()
+                # Validate the date
+                date(int(y), int(m), int(d))
+                return f"{y}-{m}-{d}"
+            return None
+
+        elif data_type == CustomField.FieldDataType.MONETARY:
+            cleaned = re.sub(r"[^\d.,\-]", "", value)
+            if not cleaned or cleaned in (".", ",", "-"):
+                return None
+            if "," in cleaned and "." in cleaned:
+                if cleaned.rindex(",") > cleaned.rindex("."):
+                    cleaned = cleaned.replace(".", "").replace(",", ".")
+                else:
+                    cleaned = cleaned.replace(",", "")
+            elif "," in cleaned:
+                cleaned = cleaned.replace(",", ".")
+            # Validate it parses as a number
+            float(cleaned)
+            return cleaned
+
+        elif data_type == CustomField.FieldDataType.BOOL:
+            lower = value.lower().strip()
+            if lower in ("true", "yes", "1", "ja", "oui", "si", "x"):
+                return True
+            elif lower in ("false", "no", "0", "nein", "non"):
+                return False
+            return None
+
+        else:
+            # Unsupported types (DOCUMENTLINK, SELECT) — can't OCR into these
+            logger.debug(
+                "Zone OCR: unsupported custom field type %s for OCR extraction",
+                data_type,
+            )
+            return None
+
+    except (ValueError, TypeError) as e:
+        logger.warning("Zone OCR: could not convert %r to %s: %s", value, data_type, e)
+        return None
+
+
+def _detect_mime(path: Path) -> str | None:
+    """Detect MIME type of a file."""
+    try:
+        import magic
+
+        return magic.from_file(str(path), mime=True)
+    except ImportError:
+        pass
+    except Exception:
+        logger.debug("Zone OCR: magic failed for %s, falling back to extension", path)
+
+    suffix = path.suffix.lower()
+    return {
+        ".pdf": "application/pdf",
+        ".png": "image/png",
+        ".jpg": "image/jpeg",
+        ".jpeg": "image/jpeg",
+        ".tiff": "image/tiff",
+        ".tif": "image/tiff",
+        ".webp": "image/webp",
+        ".bmp": "image/bmp",
+        ".gif": "image/gif",
+    }.get(suffix)

src/paperless/middleware.py

@@ -1,23 +1,8 @@
-from compression_middleware.middleware import CompressionMiddleware
 from django.conf import settings
 
 from paperless import version
 
 
-class StreamAwareCompressionMiddleware(CompressionMiddleware):
-    """
-    Bypasses compression for server-sent streams (text/event-stream).
-
-    See https://github.com/friedelwolff/django-compression-middleware/pull/7
-    """
-
-    def process_response(self, request, response):
-        content_type = response.headers.get("Content-Type", "")
-        if content_type.startswith("text/event-stream"):
-            return response
-        return super().process_response(request, response)
-
-
 class ApiVersionMiddleware:
     def __init__(self, get_response):
         self.get_response = get_response

src/paperless/settings/__init__.py

@@ -10,6 +10,7 @@ from pathlib import Path
 from typing import Final
 from urllib.parse import urlparse
 
+from compression_middleware.middleware import CompressionMiddleware
 from django.core.exceptions import ImproperlyConfigured
 from django.utils.translation import gettext_lazy as _
 from dotenv import load_dotenv
@@ -95,13 +96,6 @@ MODEL_FILE = get_path_from_env(
     "PAPERLESS_MODEL_FILE",
     DATA_DIR / "classification_model.pickle",
 )
-
-# Minimum confidence (0.0-1.0) for the ML classifier to assign a correspondent,
-# document type, or storage path. 0.0 disables the threshold.
-CLASSIFIER_MATCH_THRESHOLD: Final[float] = get_float_from_env(
-    "PAPERLESS_CLASSIFIER_MATCH_THRESHOLD",
-    0.6,
-)
 LLM_INDEX_DIR = DATA_DIR / "llm_index"
 LLM_INDEX_LOCK = LLM_INDEX_DIR / "index.lock"
 # Cross-process read/write lock guarding the LLM index compaction/migration
@@ -200,10 +194,22 @@ MIDDLEWARE = [
     "allauth.account.middleware.AccountMiddleware",
 ]
 
-# Optional to enable compression. The subclass leaves server-sent events
-# uncompressed; see paperless.middleware.StreamAwareCompressionMiddleware.
+# Optional to enable compression
 if get_bool_from_env("PAPERLESS_ENABLE_COMPRESSION", "yes"):  # pragma: no cover
-    MIDDLEWARE.insert(0, "paperless.middleware.StreamAwareCompressionMiddleware")
+    MIDDLEWARE.insert(0, "compression_middleware.middleware.CompressionMiddleware")
+
+# Workaround to not compress streaming responses (e.g. chat).
+# See https://github.com/friedelwolff/django-compression-middleware/pull/7
+original_process_response = CompressionMiddleware.process_response
+
+
+def patched_process_response(self, request, response):
+    if getattr(request, "compress_exempt", False):
+        return response
+    return original_process_response(self, request, response)
+
+
+CompressionMiddleware.process_response = patched_process_response
 
 ROOT_URLCONF = "paperless.urls"
 

src/paperless/tests/test_compression_middleware.py

@@ -1,54 +0,0 @@
-from django.http import HttpResponse
-from django.http import StreamingHttpResponse
-from django.test import RequestFactory
-from django.test import TestCase
-
-from paperless.middleware import StreamAwareCompressionMiddleware
-
-
-class TestStreamAwareCompressionMiddleware(TestCase):
-    def setUp(self) -> None:
-        super().setUp()
-        self.factory = RequestFactory()
-        self.middleware = StreamAwareCompressionMiddleware(lambda request: None)
-
-    def _request(self):
-        return self.factory.get(
-            "/api/documents/chat/",
-            HTTP_ACCEPT_ENCODING="gzip, deflate, br, zstd",
-        )
-
-    def test_event_stream_is_not_compressed(self) -> None:
-        """
-        GIVEN:
-            - A server-sent event response produced chunk by chunk
-        WHEN:
-            - The compression middleware processes it
-        THEN:
-            - It is passed through unencoded, one wire chunk per source chunk
-        """
-        chunks = [f"token{i} ".encode() for i in range(40)]
-        response = StreamingHttpResponse(
-            iter(chunks),
-            content_type="text/event-stream",
-        )
-
-        response = self.middleware.process_response(self._request(), response)
-
-        assert not response.has_header("Content-Encoding")
-        assert list(response.streaming_content) == chunks
-
-    def test_regular_response_is_still_compressed(self) -> None:
-        """
-        GIVEN:
-            - An ordinary response large enough to be worth compressing
-        WHEN:
-            - The compression middleware processes it
-        THEN:
-            - It is compressed as before
-        """
-        response = HttpResponse(b"a" * 5000, content_type="application/json")
-
-        response = self.middleware.process_response(self._request(), response)
-
-        assert response.has_header("Content-Encoding")

src/paperless/urls.py

@@ -29,6 +29,7 @@ from documents.views import IndexView
 from documents.views import LogViewSet
 from documents.views import MergeDocumentsAsVersionsView
 from documents.views import MergeDocumentsView
+from documents.views import OcrTemplateViewSet
 from documents.views import PostDocumentView
 from documents.views import RemoteVersionView
 from documents.views import RemovePasswordDocumentsView
@@ -87,6 +88,7 @@ api_router.register(r"workflow_triggers", WorkflowTriggerViewSet)
 api_router.register(r"workflow_actions", WorkflowActionViewSet)
 api_router.register(r"workflows", WorkflowViewSet)
 api_router.register(r"custom_fields", CustomFieldViewSet)
+api_router.register(r"ocr_templates", OcrTemplateViewSet)
 api_router.register(r"config", ApplicationConfigurationViewSet)
 api_router.register(r"processed_mail", ProcessedMailViewSet)
 

src/paperless_ai/client.py

@@ -40,6 +40,7 @@ LLM_SYSTEM_PROMPT = (
 
 # openai-python rejects empty keys since 2.34.0, "fake" is the stand-in from
 # llama-index's own OpenAILike docs https://docs.llamaindex.ai/en/stable/api_reference/llms/openai_like/
+# TODO: remove pending resolution of https://github.com/openai/openai-python/issues/3224
 PLACEHOLDER_API_KEY: Final = "fake"