.env.example

@@ -0,0 +1,36 @@
+# Datenbank (PostgreSQL)
+POSTGRES_USER=archive_user
+POSTGRES_PASSWORD=change-me
+POSTGRES_DB=family_archive_db
+
+# Object Storage (MinIO)
+MINIO_ROOT_USER=minio_admin
+MINIO_ROOT_PASSWORD=change-me
+MINIO_DEFAULT_BUCKETS=archive-documents
+
+# Ports (für Zugriff vom Host/NAS)
+PORT_DB=5432
+PORT_MINIO_API=9000
+PORT_MINIO_CONSOLE=9001
+PORT_BACKEND=8080
+PORT_FRONTEND=5173
+
+# Mailpit — local mail catcher (dev only, included in docker-compose)
+# Web UI:  http://localhost:8025
+# SMTP:    localhost:1025  (used automatically by the backend container)
+PORT_MAILPIT_UI=8100
+PORT_MAILPIT_SMTP=1025
+
+# OCR Training — set a secret token to protect the /train and /segtrain endpoints on the
+# Python OCR microservice. Leave empty to disable token authentication (development only).
+# OCR_TRAINING_TOKEN=change-me-in-production
+
+# Production SMTP — uncomment and fill in to send real emails instead of catching them
+# APP_BASE_URL=https://your-domain.example.com
+# MAIL_HOST=smtp.example.com
+# MAIL_PORT=587
+# MAIL_USERNAME=your-smtp-user
+# MAIL_PASSWORD=your-smtp-password
+# MAIL_SMTP_AUTH=true
+# MAIL_STARTTLS_ENABLE=true
+# APP_MAIL_FROM=noreply@your-domain.example.com

backend/pom.xml

@@ -152,6 +152,13 @@
 			<artifactId>springdoc-openapi-starter-webmvc-ui</artifactId>
 			<version>3.0.2</version>
 		</dependency>
+
+		<!-- PDF rendering for training data export -->
+		<dependency>
+			<groupId>org.apache.pdfbox</groupId>
+			<artifactId>pdfbox</artifactId>
+			<version>3.0.4</version>
+		</dependency>
 	</dependencies>
 
 

backend/src/main/java/org/raddatz/familienarchiv/controller/DocumentController.java

@@ -12,6 +12,7 @@ import java.util.UUID;
 
 
 import io.swagger.v3.oas.annotations.Parameter;
+import io.swagger.v3.oas.annotations.responses.ApiResponse;
 import org.raddatz.familienarchiv.dto.DocumentSearchResult;
 import org.raddatz.familienarchiv.dto.DocumentUpdateDTO;
 import org.raddatz.familienarchiv.dto.DocumentVersionSummary;
@@ -21,6 +22,7 @@ import org.raddatz.familienarchiv.exception.ErrorCode;
 import org.raddatz.familienarchiv.model.Document;
 import org.raddatz.familienarchiv.dto.DocumentSort;
 import org.raddatz.familienarchiv.model.DocumentStatus;
+import org.raddatz.familienarchiv.model.TrainingLabel;
 import org.raddatz.familienarchiv.model.DocumentVersion;
 import org.raddatz.familienarchiv.security.Permission;
 import org.raddatz.familienarchiv.security.RequirePermission;
@@ -35,9 +37,11 @@ import org.springframework.core.io.InputStreamResource;
 import org.springframework.web.bind.annotation.DeleteMapping;
 import org.springframework.web.bind.annotation.GetMapping;
 import org.springframework.web.bind.annotation.ModelAttribute;
+import org.springframework.web.bind.annotation.PatchMapping;
 import org.springframework.web.bind.annotation.PathVariable;
 import org.springframework.web.bind.annotation.PostMapping;
 import org.springframework.web.bind.annotation.PutMapping;
+import org.springframework.web.bind.annotation.RequestBody;
 import org.springframework.web.bind.annotation.RequestMapping;
 import org.springframework.web.bind.annotation.RequestParam;
 import org.springframework.web.bind.annotation.RequestPart;
@@ -208,6 +212,30 @@ public class DocumentController {
         return ResponseEntity.ok(DocumentSearchResult.of(results));
     }
 
+    // --- TRAINING LABELS ---
+
+    public record TrainingLabelRequest(String label, boolean enrolled) {}
+
+    @PatchMapping("/{id}/training-labels")
+    @RequirePermission(Permission.WRITE_ALL)
+    @ApiResponse(responseCode = "204")
+    public ResponseEntity<Void> patchTrainingLabel(
+            @PathVariable UUID id,
+            @RequestBody TrainingLabelRequest req) {
+        TrainingLabel label;
+        try {
+            label = TrainingLabel.valueOf(req.label());
+        } catch (IllegalArgumentException e) {
+            throw new ResponseStatusException(HttpStatus.BAD_REQUEST, "Unknown training label: " + req.label());
+        }
+        if (req.enrolled()) {
+            documentService.addTrainingLabel(id, label);
+        } else {
+            documentService.removeTrainingLabel(id, label);
+        }
+        return ResponseEntity.noContent().build();
+    }
+
     // --- VERSIONS ---
 
     @GetMapping("/{id}/versions")

backend/src/main/java/org/raddatz/familienarchiv/controller/OcrController.java

@@ -7,16 +7,23 @@ import org.raddatz.familienarchiv.dto.OcrStatusDTO;
 import org.raddatz.familienarchiv.dto.TriggerOcrDTO;
 import org.raddatz.familienarchiv.model.AppUser;
 import org.raddatz.familienarchiv.model.OcrJob;
+import org.raddatz.familienarchiv.model.OcrTrainingRun;
 import org.raddatz.familienarchiv.security.Permission;
 import org.raddatz.familienarchiv.security.RequirePermission;
 import org.raddatz.familienarchiv.service.OcrBatchService;
 import org.raddatz.familienarchiv.service.OcrProgressService;
 import org.raddatz.familienarchiv.service.OcrService;
+import org.raddatz.familienarchiv.service.OcrTrainingService;
+import org.raddatz.familienarchiv.service.SegmentationTrainingExportService;
+import org.raddatz.familienarchiv.service.TrainingDataExportService;
 import org.raddatz.familienarchiv.service.UserService;
+import org.springframework.http.HttpHeaders;
 import org.springframework.http.HttpStatus;
 import org.springframework.http.MediaType;
+import org.springframework.http.ResponseEntity;
 import org.springframework.security.core.Authentication;
 import org.springframework.web.bind.annotation.*;
+import org.springframework.web.servlet.mvc.method.annotation.StreamingResponseBody;
 import org.springframework.web.servlet.mvc.method.annotation.SseEmitter;
 
 import jakarta.validation.Valid;
@@ -32,6 +39,9 @@ public class OcrController {
     private final OcrBatchService ocrBatchService;
     private final OcrProgressService ocrProgressService;
     private final UserService userService;
+    private final TrainingDataExportService trainingDataExportService;
+    private final SegmentationTrainingExportService segmentationTrainingExportService;
+    private final OcrTrainingService ocrTrainingService;
 
     @PostMapping("/api/documents/{documentId}/ocr")
     @ResponseStatus(HttpStatus.ACCEPTED)
@@ -41,7 +51,8 @@ public class OcrController {
             @RequestBody TriggerOcrDTO dto,
             Authentication authentication) {
         UUID userId = resolveUserId(authentication);
-        UUID jobId = ocrService.startOcr(documentId, dto.getScriptType(), userId);
+        UUID jobId = ocrService.startOcr(documentId, dto.getScriptType(), userId,
+                Boolean.TRUE.equals(dto.getUseExistingAnnotations()));
         return Map.of("jobId", jobId);
     }
 
@@ -75,6 +86,54 @@ public class OcrController {
         return ocrService.getDocumentOcrStatus(documentId);
     }
 
+    @GetMapping("/api/ocr/training-data/export")
+    @RequirePermission(Permission.ADMIN)
+    public ResponseEntity<StreamingResponseBody> exportTrainingData() {
+        if (trainingDataExportService.queryEligibleBlocks().isEmpty()) {
+            return ResponseEntity.noContent().build();
+        }
+        StreamingResponseBody body = trainingDataExportService.exportToZip();
+        return ResponseEntity.ok()
+                .contentType(MediaType.parseMediaType("application/zip"))
+                .header(HttpHeaders.CONTENT_DISPOSITION, "attachment; filename=\"training-data.zip\"")
+                .body(body);
+    }
+
+    @GetMapping("/api/ocr/segmentation-training-data/export")
+    @RequirePermission(Permission.ADMIN)
+    public ResponseEntity<StreamingResponseBody> exportSegmentationTrainingData() {
+        if (segmentationTrainingExportService.querySegmentationBlocks().isEmpty()) {
+            return ResponseEntity.noContent().build();
+        }
+        StreamingResponseBody body = segmentationTrainingExportService.exportToZip();
+        return ResponseEntity.ok()
+                .contentType(MediaType.parseMediaType("application/zip"))
+                .header(HttpHeaders.CONTENT_DISPOSITION, "attachment; filename=\"segmentation-data.zip\"")
+                .body(body);
+    }
+
+    @PostMapping("/api/ocr/train")
+    @ResponseStatus(HttpStatus.CREATED)
+    @RequirePermission(Permission.ADMIN)
+    public OcrTrainingRun triggerTraining(Authentication authentication) {
+        UUID userId = resolveUserId(authentication);
+        return ocrTrainingService.triggerTraining(userId);
+    }
+
+    @PostMapping("/api/ocr/segtrain")
+    @ResponseStatus(HttpStatus.CREATED)
+    @RequirePermission(Permission.ADMIN)
+    public OcrTrainingRun triggerSegTraining(Authentication authentication) {
+        UUID userId = resolveUserId(authentication);
+        return ocrTrainingService.triggerSegTraining(userId);
+    }
+
+    @GetMapping("/api/ocr/training-info")
+    @RequirePermission(Permission.ADMIN)
+    public OcrTrainingService.TrainingInfoResponse getTrainingInfo() {
+        return ocrTrainingService.getTrainingInfo();
+    }
+
     private UUID resolveUserId(Authentication authentication) {
         if (authentication == null || !authentication.isAuthenticated()) return null;
         try {

backend/src/main/java/org/raddatz/familienarchiv/dto/TriggerOcrDTO.java

@@ -10,4 +10,5 @@ import org.raddatz.familienarchiv.model.ScriptType;
 @AllArgsConstructor
 public class TriggerOcrDTO {
     private ScriptType scriptType;
+    private Boolean useExistingAnnotations = false;
 }

backend/src/main/java/org/raddatz/familienarchiv/exception/ErrorCode.java

@@ -49,8 +49,6 @@ public enum ErrorCode {
     // --- Annotations ---
     /** The annotation with the given ID does not exist. 404 */
     ANNOTATION_NOT_FOUND,
-    /** The new annotation overlaps an existing one on the same page. 409 */
-    ANNOTATION_OVERLAP,
 
     // --- Transcription Blocks ---
     /** The transcription block with the given ID does not exist. 404 */
@@ -75,6 +73,8 @@ public enum ErrorCode {
     OCR_DOCUMENT_NOT_UPLOADED,
     /** OCR processing failed for the document. 500 */
     OCR_PROCESSING_FAILED,
+    /** A training run is already in progress. 409 */
+    TRAINING_ALREADY_RUNNING,
 
     // --- Generic ---
     /** Request validation failed (missing or malformed fields). 400 */

backend/src/main/java/org/raddatz/familienarchiv/model/Document.java

@@ -110,4 +110,11 @@ public class Document {
     @JoinTable(name = "document_tags", joinColumns = @JoinColumn(name = "document_id"), inverseJoinColumns = @JoinColumn(name = "tag_id"))
     @Builder.Default
     private Set<Tag> tags = new HashSet<>();
+
+    @ElementCollection(fetch = FetchType.EAGER)
+    @CollectionTable(name = "document_training_labels", joinColumns = @JoinColumn(name = "document_id"))
+    @Column(name = "label")
+    @Enumerated(EnumType.STRING)
+    @Builder.Default
+    private Set<TrainingLabel> trainingLabels = new HashSet<>();
 }

backend/src/main/java/org/raddatz/familienarchiv/model/OcrTrainingRun.java

@@ -0,0 +1,69 @@
+package org.raddatz.familienarchiv.model;
+
+import io.swagger.v3.oas.annotations.media.Schema;
+import jakarta.persistence.*;
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+import org.hibernate.annotations.CreationTimestamp;
+
+import java.time.Instant;
+import java.util.UUID;
+
+@Entity
+@Table(name = "ocr_training_runs")
+@Data
+@NoArgsConstructor
+@AllArgsConstructor
+@Builder
+public class OcrTrainingRun {
+
+    @Id
+    @GeneratedValue(strategy = GenerationType.UUID)
+    @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+    private UUID id;
+
+    @Enumerated(EnumType.STRING)
+    @Column(nullable = false)
+    @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+    private TrainingStatus status;
+
+    @Column(name = "block_count", nullable = false)
+    @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+    private int blockCount;
+
+    @Column(name = "document_count", nullable = false)
+    @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+    private int documentCount;
+
+    @Column(name = "model_name", nullable = false)
+    @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+    private String modelName;
+
+    @Column(name = "cer")
+    private Double cer;
+
+    @Column(name = "loss")
+    private Double loss;
+
+    @Column(name = "accuracy")
+    private Double accuracy;
+
+    @Column(name = "epochs")
+    private Integer epochs;
+
+    @Column(name = "error_message")
+    private String errorMessage;
+
+    @Column(name = "triggered_by")
+    private UUID triggeredBy;
+
+    @CreationTimestamp
+    @Column(name = "created_at", nullable = false, updatable = false)
+    @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
+    private Instant createdAt;
+
+    @Column(name = "completed_at")
+    private Instant completedAt;
+}

backend/src/main/java/org/raddatz/familienarchiv/model/TrainingLabel.java

@@ -0,0 +1,6 @@
+package org.raddatz.familienarchiv.model;
+
+public enum TrainingLabel {
+    KURRENT_RECOGNITION,
+    KURRENT_SEGMENTATION
+}

backend/src/main/java/org/raddatz/familienarchiv/model/TrainingStatus.java

@@ -0,0 +1,7 @@
+package org.raddatz.familienarchiv.model;
+
+public enum TrainingStatus {
+    RUNNING,
+    DONE,
+    FAILED
+}

backend/src/main/java/org/raddatz/familienarchiv/model/TranscriptionBlock.java

@@ -30,8 +30,7 @@ public class TranscriptionBlock {
     @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
     private UUID documentId;
 
-    @Column(nullable = false, columnDefinition = "TEXT")
+    @Column(columnDefinition = "TEXT")
-    @Schema(requiredMode = Schema.RequiredMode.REQUIRED)
     private String text;
 
     @Column(length = 200)

backend/src/main/java/org/raddatz/familienarchiv/repository/OcrTrainingRunRepository.java

@@ -0,0 +1,16 @@
+package org.raddatz.familienarchiv.repository;
+
+import org.raddatz.familienarchiv.model.OcrTrainingRun;
+import org.raddatz.familienarchiv.model.TrainingStatus;
+import org.springframework.data.jpa.repository.JpaRepository;
+
+import java.util.List;
+import java.util.Optional;
+import java.util.UUID;
+
+public interface OcrTrainingRunRepository extends JpaRepository<OcrTrainingRun, UUID> {
+
+    Optional<OcrTrainingRun> findFirstByStatus(TrainingStatus status);
+
+    List<OcrTrainingRun> findTop5ByOrderByCreatedAtDesc();
+}

backend/src/main/java/org/raddatz/familienarchiv/repository/TranscriptionBlockRepository.java

@@ -2,6 +2,7 @@ package org.raddatz.familienarchiv.repository;
 
 import org.raddatz.familienarchiv.model.TranscriptionBlock;
 import org.springframework.data.jpa.repository.JpaRepository;
+import org.springframework.data.jpa.repository.Query;
 
 import java.util.List;
 import java.util.Optional;
@@ -13,5 +14,27 @@ public interface TranscriptionBlockRepository extends JpaRepository<Transcriptio
 
     Optional<TranscriptionBlock> findByIdAndDocumentId(UUID id, UUID documentId);
 
+    Optional<TranscriptionBlock> findByAnnotationId(UUID annotationId);
+
+    void deleteByAnnotationId(UUID annotationId);
+
     int countByDocumentId(UUID documentId);
+
+    @Query("""
+            SELECT b FROM TranscriptionBlock b
+            JOIN DocumentAnnotation a ON a.id = b.annotationId
+            JOIN Document d ON d.id = b.documentId
+            WHERE b.reviewed = true
+            AND 'KURRENT_RECOGNITION' MEMBER OF d.trainingLabels
+            """)
+    List<TranscriptionBlock> findEligibleKurrentBlocks();
+
+    @Query("""
+            SELECT b FROM TranscriptionBlock b
+            JOIN DocumentAnnotation a ON a.id = b.annotationId
+            JOIN Document d ON d.id = b.documentId
+            WHERE b.source = 'MANUAL'
+            AND 'KURRENT_SEGMENTATION' MEMBER OF d.trainingLabels
+            """)
+    List<TranscriptionBlock> findSegmentationBlocks();
 }

backend/src/main/java/org/raddatz/familienarchiv/service/AnnotationService.java

@@ -6,6 +6,7 @@ import org.raddatz.familienarchiv.exception.DomainException;
 import org.raddatz.familienarchiv.exception.ErrorCode;
 import org.raddatz.familienarchiv.model.DocumentAnnotation;
 import org.raddatz.familienarchiv.repository.AnnotationRepository;
+import org.raddatz.familienarchiv.repository.TranscriptionBlockRepository;
 import org.springframework.stereotype.Service;
 import org.springframework.transaction.annotation.Transactional;
 
@@ -17,6 +18,7 @@ import java.util.UUID;
 public class AnnotationService {
 
     private final AnnotationRepository annotationRepository;
+    private final TranscriptionBlockRepository blockRepository;
 
     public List<DocumentAnnotation> listAnnotations(UUID documentId) {
         return annotationRepository.findByDocumentId(documentId);
@@ -24,15 +26,6 @@ public class AnnotationService {
 
     @Transactional
     public DocumentAnnotation createAnnotation(UUID documentId, CreateAnnotationDTO dto, UUID userId, String fileHash) {
-        List<DocumentAnnotation> existing =
-                annotationRepository.findByDocumentIdAndPageNumber(documentId, dto.getPageNumber());
-
-        boolean overlaps = existing.stream().anyMatch(a -> overlaps(a, dto));
-        if (overlaps) {
-            throw DomainException.conflict(
-                    ErrorCode.ANNOTATION_OVERLAP, "Annotation overlaps an existing one on this page");
-        }
-
         DocumentAnnotation annotation = DocumentAnnotation.builder()
                 .documentId(documentId)
                 .pageNumber(dto.getPageNumber())
@@ -79,6 +72,7 @@ public class AnnotationService {
             throw DomainException.forbidden("Only the annotation author can delete it");
         }
 
+        blockRepository.deleteByAnnotationId(annotationId);
         annotationRepository.delete(annotation);
     }
 
@@ -90,14 +84,4 @@ public class AnnotationService {
         });
     }
 
-    // ─── private helpers ──────────────────────────────────────────────────────
-
-    private boolean overlaps(DocumentAnnotation existing, CreateAnnotationDTO dto) {
-        double ex2 = existing.getX() + existing.getWidth();
-        double ey2 = existing.getY() + existing.getHeight();
-        double dx2 = dto.getX() + dto.getWidth();
-        double dy2 = dto.getY() + dto.getHeight();
-        return existing.getX() < dx2 && ex2 > dto.getX()
-                && existing.getY() < dy2 && ey2 > dto.getY();
-    }
 }

backend/src/main/java/org/raddatz/familienarchiv/service/DocumentService.java

@@ -9,6 +9,7 @@ import org.raddatz.familienarchiv.model.Document;
 import org.raddatz.familienarchiv.dto.DocumentSort;
 import org.raddatz.familienarchiv.model.DocumentStatus;
 import org.raddatz.familienarchiv.model.ScriptType;
+import org.raddatz.familienarchiv.model.TrainingLabel;
 import org.raddatz.familienarchiv.model.Person;
 import org.raddatz.familienarchiv.model.Tag;
 import org.raddatz.familienarchiv.repository.DocumentRepository;
@@ -385,6 +386,20 @@ public class DocumentService {
         documentRepository.save(doc);
     }
 
+    @Transactional
+    public void addTrainingLabel(UUID documentId, TrainingLabel label) {
+        Document doc = getDocumentById(documentId);
+        doc.getTrainingLabels().add(label);
+        documentRepository.save(doc);
+    }
+
+    @Transactional
+    public void removeTrainingLabel(UUID documentId, TrainingLabel label) {
+        Document doc = getDocumentById(documentId);
+        doc.getTrainingLabels().remove(label);
+        documentRepository.save(doc);
+    }
+
     public Document getDocumentById(UUID id) {
         return documentRepository.findById(id)
                 .orElseThrow(() -> DomainException.notFound(ErrorCode.DOCUMENT_NOT_FOUND, "Document not found: " + id));

backend/src/main/java/org/raddatz/familienarchiv/service/FileService.java

@@ -114,7 +114,8 @@ public class FileService {
 
     /**
      * Generates a presigned URL for downloading an object from S3/MinIO.
-     * Valid for 15 minutes — enough for OCR processing on CPU.
+     * Valid for 1 hour — covers multi-page documents on CPU-only OCR hardware
+     * (a 100-page document at 10 s/page takes ~17 min; 1 h gives ample headroom).
      */
     public String generatePresignedUrl(String s3Key) {
         GetObjectRequest getObjectRequest = GetObjectRequest.builder()
@@ -123,7 +124,7 @@ public class FileService {
                 .build();
 
         GetObjectPresignRequest presignRequest = GetObjectPresignRequest.builder()
-                .signatureDuration(Duration.ofMinutes(15))
+                .signatureDuration(Duration.ofHours(1))
                 .getObjectRequest(getObjectRequest)
                 .build();
 

backend/src/main/java/org/raddatz/familienarchiv/service/OcrAsyncRunner.java

@@ -32,6 +32,11 @@ public class OcrAsyncRunner {
 
     @Async
     public void runSingleDocument(UUID jobId, UUID documentId, UUID userId) {
+        runSingleDocument(jobId, documentId, userId, false);
+    }
+
+    @Async
+    public void runSingleDocument(UUID jobId, UUID documentId, UUID userId, boolean useExistingAnnotations) {
         OcrJob job = ocrJobRepository.findById(jobId).orElse(null);
         if (job == null) return;
 
@@ -49,7 +54,18 @@ public class OcrAsyncRunner {
 
         try {
             updateProgress(job, "LOADING");
-            clearExistingBlocks(documentId);
+
+            List<OcrClient.OcrRegion> regions = null;
+            if (useExistingAnnotations) {
+                regions = annotationService.listAnnotations(documentId).stream()
+                        .map(a -> new OcrClient.OcrRegion(
+                                a.getId().toString(), a.getPageNumber(),
+                                a.getX(), a.getY(), a.getWidth(), a.getHeight()))
+                        .toList();
+            } else {
+                clearExistingBlocks(documentId);
+            }
+
             String pdfUrl = fileService.generatePresignedUrl(doc.getFilePath());
 
             AtomicInteger blockCounter = new AtomicInteger(0);
@@ -57,7 +73,7 @@ public class OcrAsyncRunner {
             AtomicInteger skippedPages = new AtomicInteger(0);
             AtomicInteger totalPages = new AtomicInteger(0);
 
-            ocrClient.streamBlocks(pdfUrl, doc.getScriptType(), event -> {
+            ocrClient.streamBlocks(pdfUrl, doc.getScriptType(), regions, event -> {
                 switch (event) {
                     case OcrStreamEvent.Start start -> {
                         totalPages.set(start.totalPages());
@@ -204,14 +220,21 @@ public class OcrAsyncRunner {
 
     void createSingleBlock(UUID documentId, OcrBlockResult block,
                            UUID userId, String fileHash, int sortOrder) {
-        CreateAnnotationDTO annotationDTO = new CreateAnnotationDTO(
+        if (block.annotationId() != null) {
-                block.pageNumber(), block.x(), block.y(),
+            // Guided mode — annotation already exists; upsert the text block only
-                block.width(), block.height(), OCR_ANNOTATION_COLOR);
+            transcriptionService.upsertGuidedBlock(
+                    documentId, UUID.fromString(block.annotationId()), block.text(), userId);
+        } else {
+            // Normal mode — create a new annotation and a new OCR block
+            CreateAnnotationDTO annotationDTO = new CreateAnnotationDTO(
+                    block.pageNumber(), block.x(), block.y(),
+                    block.width(), block.height(), OCR_ANNOTATION_COLOR);
 
-        DocumentAnnotation annotation = annotationService.createOcrAnnotation(
+            DocumentAnnotation annotation = annotationService.createOcrAnnotation(
-                documentId, annotationDTO, userId, fileHash, block.polygon());
+                    documentId, annotationDTO, userId, fileHash, block.polygon());
 
-        transcriptionService.createOcrBlock(documentId, annotation.getId(),
+            transcriptionService.createOcrBlock(documentId, annotation.getId(),
-                block.text(), sortOrder, userId);
+                    block.text(), sortOrder, userId);
+        }
     }
 }

backend/src/main/java/org/raddatz/familienarchiv/service/OcrBlockResult.java

@@ -12,5 +12,6 @@ public record OcrBlockResult(
         double width,
         double height,
         List<List<Double>> polygon,
-        String text
+        String text,
+        String annotationId   // null in normal mode; set in guided mode to link back to existing annotation
 ) {}

backend/src/main/java/org/raddatz/familienarchiv/service/OcrClient.java

@@ -10,12 +10,42 @@ import java.util.function.Consumer;
 public interface OcrClient {
     List<OcrBlockResult> extractBlocks(String pdfUrl, ScriptType scriptType);
 
+    /**
+     * A pre-drawn annotation region to use as guidance for OCR.
+     * When regions are provided, the OCR engine crops to each region and
+     * runs recognition only within that area, skipping full-page layout detection.
+     */
+    record OcrRegion(String annotationId, int pageNumber,
+                     double x, double y, double width, double height) {}
+
+    /**
+     * Send a training ZIP to the OCR service for fine-tuning the Kurrent model.
+     *
+     * @param trainingDataZip raw ZIP bytes produced by TrainingDataExportService
+     * @return training result metrics (loss, accuracy, epochs)
+     */
+    TrainingResult trainModel(byte[] trainingDataZip);
+
+    record TrainingResult(Double loss, Double accuracy, Double cer, Integer epochs) {}
+
+    /**
+     * Send a segmentation training ZIP to the OCR service for fine-tuning the blla model.
+     *
+     * @param trainingDataZip raw ZIP bytes produced by SegmentationTrainingExportService
+     * @return training result metrics
+     */
+    TrainingResult segtrainModel(byte[] trainingDataZip);
+
     /**
      * Stream OCR results page-by-page via NDJSON. Implementations should override
      * this method. The default exists only for backward compatibility during migration
      * — it calls extractBlocks() and synthesizes events from the collected result.
+     *
+     * @param regions optional list of pre-drawn annotation regions; when non-null,
+     *                the OCR service runs in guided mode (crop + recognize per region)
      */
-    default void streamBlocks(String pdfUrl, ScriptType scriptType, Consumer<OcrStreamEvent> handler) {
+    default void streamBlocks(String pdfUrl, ScriptType scriptType,
+                               List<OcrRegion> regions, Consumer<OcrStreamEvent> handler) {
         List<OcrBlockResult> allBlocks = extractBlocks(pdfUrl, scriptType);
 
         LinkedHashMap<Integer, List<OcrBlockResult>> byPage = new LinkedHashMap<>();

backend/src/main/java/org/raddatz/familienarchiv/service/OcrService.java

@@ -52,6 +52,11 @@ public class OcrService {
     }
 
     public UUID startOcr(UUID documentId, ScriptType scriptTypeOverride, UUID userId) {
+        return startOcr(documentId, scriptTypeOverride, userId, false);
+    }
+
+    public UUID startOcr(UUID documentId, ScriptType scriptTypeOverride, UUID userId,
+                          boolean useExistingAnnotations) {
         Document doc = documentService.getDocumentById(documentId);
 
         if (doc.getStatus() == DocumentStatus.PLACEHOLDER) {
@@ -66,6 +71,9 @@ public class OcrService {
 
         if (scriptTypeOverride != null) {
             documentService.updateScriptType(documentId, scriptTypeOverride);
+            if (scriptTypeOverride == ScriptType.HANDWRITING_KURRENT) {
+                documentService.addTrainingLabel(documentId, TrainingLabel.KURRENT_RECOGNITION);
+            }
         }
 
         OcrJob job = OcrJob.builder()
@@ -82,7 +90,7 @@ public class OcrService {
                 .build();
         ocrJobDocumentRepository.save(jobDoc);
 
-        ocrAsyncRunner.runSingleDocument(job.getId(), documentId, userId);
+        ocrAsyncRunner.runSingleDocument(job.getId(), documentId, userId, useExistingAnnotations);
         return job.getId();
     }
 }

backend/src/main/java/org/raddatz/familienarchiv/service/OcrTrainingService.java

@@ -0,0 +1,236 @@
+package org.raddatz.familienarchiv.service;
+
+import lombok.RequiredArgsConstructor;
+import lombok.extern.slf4j.Slf4j;
+import org.raddatz.familienarchiv.exception.DomainException;
+import org.raddatz.familienarchiv.exception.ErrorCode;
+import org.raddatz.familienarchiv.model.OcrTrainingRun;
+import org.raddatz.familienarchiv.model.TrainingStatus;
+import org.raddatz.familienarchiv.repository.OcrTrainingRunRepository;
+import org.raddatz.familienarchiv.repository.TranscriptionBlockRepository;
+import org.slf4j.MDC;
+import org.springframework.boot.context.event.ApplicationReadyEvent;
+import org.springframework.context.event.EventListener;
+import org.springframework.stereotype.Service;
+import org.springframework.transaction.annotation.Transactional;
+import org.springframework.transaction.support.TransactionTemplate;
+
+import java.io.ByteArrayOutputStream;
+import java.time.Instant;
+import java.util.List;
+import java.util.Map;
+import java.util.Objects;
+import java.util.UUID;
+
+@Service
+@RequiredArgsConstructor
+@Slf4j
+public class OcrTrainingService {
+
+    private final OcrTrainingRunRepository trainingRunRepository;
+    private final TrainingDataExportService trainingDataExportService;
+    private final SegmentationTrainingExportService segmentationTrainingExportService;
+    private final OcrClient ocrClient;
+    private final OcrHealthClient ocrHealthClient;
+    private final TranscriptionBlockRepository blockRepository;
+    private final TransactionTemplate txTemplate;
+
+    public record TrainingInfoResponse(
+            int availableBlocks,
+            int totalOcrBlocks,
+            int availableDocuments,
+            int availableSegBlocks,
+            boolean ocrServiceAvailable,
+            OcrTrainingRun lastRun,
+            List<OcrTrainingRun> runs
+    ) {}
+
+    // Not safe for horizontal scaling: training reloads the Kraken model in-process on the
+    // Python OCR service after each run. The DB-level RUNNING constraint (V30 partial unique
+    // index) prevents concurrent training API calls, but cannot prevent two OCR service replicas
+    // from diverging on model state. Deploy as a single instance only. See ADR-001.
+    public OcrTrainingRun triggerTraining(UUID triggeredBy) {
+        // Short transaction: guard check + create RUNNING row, then commit immediately.
+        // The DB connection is released before the OCR HTTP call, which can take several minutes.
+        OcrTrainingRun run = Objects.requireNonNull(txTemplate.execute(status -> {
+            if (trainingRunRepository.findFirstByStatus(TrainingStatus.RUNNING).isPresent()) {
+                throw DomainException.conflict(ErrorCode.TRAINING_ALREADY_RUNNING,
+                        "A training run is already in progress");
+            }
+
+            var eligibleBlocks = trainingDataExportService.queryEligibleBlocks();
+            if (eligibleBlocks.size() < 5) {
+                throw DomainException.badRequest(ErrorCode.VALIDATION_ERROR,
+                        "At least 5 eligible blocks are required to start training (found " + eligibleBlocks.size() + ")");
+            }
+
+            long documentCount = eligibleBlocks.stream()
+                    .map(b -> b.getDocumentId())
+                    .distinct()
+                    .count();
+
+            OcrTrainingRun newRun = OcrTrainingRun.builder()
+                    .status(TrainingStatus.RUNNING)
+                    .blockCount(eligibleBlocks.size())
+                    .documentCount((int) documentCount)
+                    .modelName("german_kurrent")
+                    .triggeredBy(triggeredBy)
+                    .build();
+            return trainingRunRepository.save(newRun);
+        }));
+
+        String runId = run.getId().toString();
+        MDC.put("trainingRunId", runId);
+        log.info("Started training run {} with {} blocks from {} documents",
+                runId, run.getBlockCount(), run.getDocumentCount());
+
+        try {
+            ByteArrayOutputStream baos = new ByteArrayOutputStream();
+            trainingDataExportService.exportToZip().writeTo(baos);
+            byte[] zipBytes = baos.toByteArray();
+
+            log.info("[trainingRun={}] Sending {} bytes to OCR service", runId, zipBytes.length);
+            OcrClient.TrainingResult result = ocrClient.trainModel(zipBytes);
+
+            return Objects.requireNonNull(txTemplate.execute(status -> {
+                run.setStatus(TrainingStatus.DONE);
+                run.setCompletedAt(Instant.now());
+                run.setCer(result.cer());
+                run.setLoss(result.loss());
+                run.setAccuracy(result.accuracy());
+                run.setEpochs(result.epochs());
+                OcrTrainingRun updated = trainingRunRepository.save(run);
+                log.info("[trainingRun={}] Training completed — cer={} epochs={}", runId, result.cer(), result.epochs());
+                return updated;
+            }));
+        } catch (Exception e) {
+            return Objects.requireNonNull(txTemplate.execute(status -> {
+                run.setStatus(TrainingStatus.FAILED);
+                run.setErrorMessage(e.getMessage());
+                run.setCompletedAt(Instant.now());
+                OcrTrainingRun failed = trainingRunRepository.save(run);
+                log.error("[trainingRun={}] Training failed: {}", runId, e.getMessage(), e);
+                return failed;
+            }));
+        } finally {
+            MDC.remove("trainingRunId");
+        }
+    }
+
+    public OcrTrainingRun triggerSegTraining(UUID triggeredBy) {
+        // Same pattern as triggerTraining: narrow transactions around DB writes only.
+        OcrTrainingRun run = Objects.requireNonNull(txTemplate.execute(status -> {
+            if (trainingRunRepository.findFirstByStatus(TrainingStatus.RUNNING).isPresent()) {
+                throw DomainException.conflict(ErrorCode.TRAINING_ALREADY_RUNNING,
+                        "A training run is already in progress");
+            }
+
+            var segBlocks = segmentationTrainingExportService.querySegmentationBlocks();
+            if (segBlocks.size() < 5) {
+                throw DomainException.badRequest(ErrorCode.VALIDATION_ERROR,
+                        "At least 5 eligible segments are required to start training (found " + segBlocks.size() + ")");
+            }
+
+            long documentCount = segBlocks.stream()
+                    .map(b -> b.getDocumentId())
+                    .distinct()
+                    .count();
+
+            OcrTrainingRun newRun = OcrTrainingRun.builder()
+                    .status(TrainingStatus.RUNNING)
+                    .blockCount(segBlocks.size())
+                    .documentCount((int) documentCount)
+                    .modelName("blla")
+                    .triggeredBy(triggeredBy)
+                    .build();
+            return trainingRunRepository.save(newRun);
+        }));
+
+        String runId = run.getId().toString();
+        MDC.put("trainingRunId", runId);
+        log.info("Started segmentation training run {} with {} segments from {} documents",
+                runId, run.getBlockCount(), run.getDocumentCount());
+
+        try {
+            ByteArrayOutputStream baos = new ByteArrayOutputStream();
+            segmentationTrainingExportService.exportToZip().writeTo(baos);
+            byte[] zipBytes = baos.toByteArray();
+
+            log.info("[trainingRun={}] Sending {} bytes to OCR service for segtrain", runId, zipBytes.length);
+            OcrClient.TrainingResult result = ocrClient.segtrainModel(zipBytes);
+
+            return Objects.requireNonNull(txTemplate.execute(status -> {
+                run.setStatus(TrainingStatus.DONE);
+                run.setCompletedAt(Instant.now());
+                run.setLoss(result.loss());
+                run.setAccuracy(result.accuracy());
+                run.setEpochs(result.epochs());
+                OcrTrainingRun updated = trainingRunRepository.save(run);
+                log.info("[trainingRun={}] Segmentation training completed — epochs={}", runId, result.epochs());
+                return updated;
+            }));
+        } catch (Exception e) {
+            return Objects.requireNonNull(txTemplate.execute(status -> {
+                run.setStatus(TrainingStatus.FAILED);
+                run.setErrorMessage(e.getMessage());
+                run.setCompletedAt(Instant.now());
+                OcrTrainingRun failed = trainingRunRepository.save(run);
+                log.error("[trainingRun={}] Segmentation training failed: {}", runId, e.getMessage(), e);
+                return failed;
+            }));
+        } finally {
+            MDC.remove("trainingRunId");
+        }
+    }
+
+    public TrainingInfoResponse getTrainingInfo() {
+        var eligibleBlocks = trainingDataExportService.queryEligibleBlocks();
+        int availableDocuments = (int) eligibleBlocks.stream()
+                .map(b -> b.getDocumentId())
+                .distinct()
+                .count();
+
+        int totalOcrBlocks = (int) blockRepository.count();
+        int availableSegBlocks = segmentationTrainingExportService.querySegmentationBlocks().size();
+
+        List<OcrTrainingRun> recentRuns = trainingRunRepository.findTop5ByOrderByCreatedAtDesc();
+        OcrTrainingRun lastRun = recentRuns.isEmpty() ? null : recentRuns.get(0);
+
+        return new TrainingInfoResponse(
+                eligibleBlocks.size(),
+                totalOcrBlocks,
+                availableDocuments,
+                availableSegBlocks,
+                ocrHealthClient.isHealthy(),
+                lastRun,
+                recentRuns
+        );
+    }
+
+    @EventListener(ApplicationReadyEvent.class)
+    @Transactional
+    public void recoverOrphanedRuns() {
+        var cutoff = Instant.now().minusSeconds(3600);
+        trainingRunRepository.findFirstByStatus(TrainingStatus.RUNNING).ifPresent(run -> {
+            if (run.getCreatedAt().isBefore(cutoff)) {
+                run.setStatus(TrainingStatus.FAILED);
+                run.setErrorMessage("Abgebrochen: Dienst wurde neugestartet");
+                run.setCompletedAt(Instant.now());
+                trainingRunRepository.save(run);
+                log.warn("Recovered orphaned training run {} (marked FAILED on startup)", run.getId());
+            }
+        });
+    }
+
+    public Map<String, Object> buildTrainingInfoMap(TrainingInfoResponse info) {
+        return Map.of(
+                "availableBlocks", info.availableBlocks(),
+                "totalOcrBlocks", info.totalOcrBlocks(),
+                "availableDocuments", info.availableDocuments(),
+                "availableSegBlocks", info.availableSegBlocks(),
+                "ocrServiceAvailable", info.ocrServiceAvailable(),
+                "lastRun", info.lastRun() != null ? info.lastRun() : Map.of(),
+                "runs", info.runs()
+        );
+    }
+}

backend/src/main/java/org/raddatz/familienarchiv/service/RestClientOcrClient.java

@@ -9,9 +9,14 @@ import lombok.extern.slf4j.Slf4j;
 import org.raddatz.familienarchiv.model.ScriptType;
 import org.springframework.beans.factory.annotation.Value;
 import org.springframework.core.ParameterizedTypeReference;
+import org.springframework.core.io.ByteArrayResource;
+import org.springframework.http.HttpEntity;
+import org.springframework.http.HttpHeaders;
 import org.springframework.http.MediaType;
 import org.springframework.http.client.JdkClientHttpRequestFactory;
 import org.springframework.stereotype.Component;
+import org.springframework.util.LinkedMultiValueMap;
+import org.springframework.util.MultiValueMap;
 import org.springframework.web.client.RestClient;
 
 import java.io.BufferedReader;
@@ -36,11 +41,16 @@ public class RestClientOcrClient implements OcrClient, OcrHealthClient {
             .configure(DeserializationFeature.FAIL_ON_UNKNOWN_PROPERTIES, true);
 
     private final RestClient restClient;
+    private final RestClient trainingRestClient;
     private final HttpClient streamingHttpClient;
     private final String baseUrl;
+    private final String trainingToken;
 
-    public RestClientOcrClient(@Value("${app.ocr.base-url:http://ocr-service:8000}") String baseUrl) {
+    public RestClientOcrClient(
+            @Value("${app.ocr.base-url:http://ocr-service:8000}") String baseUrl,
+            @Value("${app.ocr.training-token:}") String trainingToken) {
         this.baseUrl = baseUrl;
+        this.trainingToken = trainingToken;
 
         HttpClient httpClient = HttpClient.newBuilder()
                 .version(HttpClient.Version.HTTP_1_1)
@@ -54,6 +64,17 @@ public class RestClientOcrClient implements OcrClient, OcrHealthClient {
                 .requestFactory(requestFactory)
                 .build();
 
+        HttpClient trainingHttpClient = HttpClient.newBuilder()
+                .version(HttpClient.Version.HTTP_1_1)
+                .connectTimeout(Duration.ofSeconds(10))
+                .build();
+        JdkClientHttpRequestFactory trainingRequestFactory = new JdkClientHttpRequestFactory(trainingHttpClient);
+        trainingRequestFactory.setReadTimeout(Duration.ofMinutes(10));
+        this.trainingRestClient = RestClient.builder()
+                .baseUrl(baseUrl)
+                .requestFactory(trainingRequestFactory)
+                .build();
+
         this.streamingHttpClient = HttpClient.newBuilder()
                 .version(HttpClient.Version.HTTP_1_1)
                 .connectTimeout(Duration.ofSeconds(10))
@@ -81,6 +102,64 @@ public class RestClientOcrClient implements OcrClient, OcrHealthClient {
                 .toList();
     }
 
+    @Override
+    public OcrClient.TrainingResult trainModel(byte[] trainingDataZip) {
+        ByteArrayResource zipResource = new ByteArrayResource(trainingDataZip) {
+            @Override
+            public String getFilename() { return "training-data.zip"; }
+        };
+
+        MultiValueMap<String, Object> body = new LinkedMultiValueMap<>();
+        HttpHeaders partHeaders = new HttpHeaders();
+        partHeaders.setContentType(MediaType.parseMediaType("application/zip"));
+        body.add("file", new HttpEntity<>(zipResource, partHeaders));
+
+        var spec = trainingRestClient.post()
+                .uri("/train")
+                .contentType(MediaType.MULTIPART_FORM_DATA);
+
+        if (trainingToken != null && !trainingToken.isBlank()) {
+            spec = spec.header("X-Training-Token", trainingToken);
+        }
+
+        TrainingResultJson result = spec
+                .body(body)
+                .retrieve()
+                .body(TrainingResultJson.class);
+
+        if (result == null) return new OcrClient.TrainingResult(null, null, null, null);
+        return new OcrClient.TrainingResult(result.loss(), result.accuracy(), result.cer(), result.epochs());
+    }
+
+    @Override
+    public OcrClient.TrainingResult segtrainModel(byte[] trainingDataZip) {
+        ByteArrayResource zipResource = new ByteArrayResource(trainingDataZip) {
+            @Override
+            public String getFilename() { return "segmentation-data.zip"; }
+        };
+
+        MultiValueMap<String, Object> body = new LinkedMultiValueMap<>();
+        HttpHeaders partHeaders = new HttpHeaders();
+        partHeaders.setContentType(MediaType.parseMediaType("application/zip"));
+        body.add("file", new HttpEntity<>(zipResource, partHeaders));
+
+        var spec = trainingRestClient.post()
+                .uri("/segtrain")
+                .contentType(MediaType.MULTIPART_FORM_DATA);
+
+        if (trainingToken != null && !trainingToken.isBlank()) {
+            spec = spec.header("X-Training-Token", trainingToken);
+        }
+
+        TrainingResultJson result = spec
+                .body(body)
+                .retrieve()
+                .body(TrainingResultJson.class);
+
+        if (result == null) return new OcrClient.TrainingResult(null, null, null, null);
+        return new OcrClient.TrainingResult(result.loss(), result.accuracy(), result.cer(), result.epochs());
+    }
+
     @Override
     public boolean isHealthy() {
         try {
@@ -96,13 +175,18 @@ public class RestClientOcrClient implements OcrClient, OcrHealthClient {
     }
 
     @Override
-    public void streamBlocks(String pdfUrl, ScriptType scriptType, Consumer<OcrStreamEvent> handler) {
+    public void streamBlocks(String pdfUrl, ScriptType scriptType,
+                              List<OcrRegion> regions, Consumer<OcrStreamEvent> handler) {
         String body;
         try {
-            body = NDJSON_MAPPER.writeValueAsString(Map.of(
+            var requestMap = new java.util.LinkedHashMap<String, Object>();
-                    "pdfUrl", pdfUrl,
+            requestMap.put("pdfUrl", pdfUrl);
-                    "scriptType", scriptType.name(),
+            requestMap.put("scriptType", scriptType.name());
-                    "language", "de"));
+            requestMap.put("language", "de");
+            if (regions != null && !regions.isEmpty()) {
+                requestMap.put("regions", regions);
+            }
+            body = NDJSON_MAPPER.writeValueAsString(requestMap);
         } catch (IOException e) {
             throw new RuntimeException("Failed to serialize OCR request", e);
         }
@@ -120,7 +204,7 @@ public class RestClientOcrClient implements OcrClient, OcrHealthClient {
 
             if (response.statusCode() == 404) {
                 log.info("OCR service does not support /ocr/stream (404), falling back to /ocr");
-                OcrClient.super.streamBlocks(pdfUrl, scriptType, handler);
+                OcrClient.super.streamBlocks(pdfUrl, scriptType, regions, handler);
                 return;
             }
 
@@ -171,6 +255,8 @@ public class RestClientOcrClient implements OcrClient, OcrHealthClient {
         }
     }
 
+    record TrainingResultJson(Double loss, Double accuracy, Double cer, Integer epochs) {}
+
     record OcrBlockJson(
             @JsonProperty("pageNumber") int pageNumber,
             double x,
@@ -178,10 +264,11 @@ public class RestClientOcrClient implements OcrClient, OcrHealthClient {
             double width,
             double height,
             List<List<Double>> polygon,
-            String text
+            String text,
+            String annotationId
     ) {
         OcrBlockResult toResult() {
-            return new OcrBlockResult(pageNumber, x, y, width, height, polygon, text);
+            return new OcrBlockResult(pageNumber, x, y, width, height, polygon, text, annotationId);
         }
     }
 }

backend/src/main/java/org/raddatz/familienarchiv/service/SegmentationTrainingExportService.java

@@ -0,0 +1,174 @@
+package org.raddatz.familienarchiv.service;
+
+import lombok.RequiredArgsConstructor;
+import lombok.extern.slf4j.Slf4j;
+import org.apache.pdfbox.Loader;
+import org.apache.pdfbox.pdmodel.PDDocument;
+import org.apache.pdfbox.rendering.PDFRenderer;
+import org.raddatz.familienarchiv.model.Document;
+import org.raddatz.familienarchiv.model.DocumentAnnotation;
+import org.raddatz.familienarchiv.model.TranscriptionBlock;
+import org.raddatz.familienarchiv.repository.AnnotationRepository;
+import org.raddatz.familienarchiv.repository.DocumentRepository;
+import org.raddatz.familienarchiv.repository.TranscriptionBlockRepository;
+import org.springframework.stereotype.Service;
+import org.springframework.web.servlet.mvc.method.annotation.StreamingResponseBody;
+
+import javax.imageio.ImageIO;
+import java.awt.image.BufferedImage;
+import java.io.IOException;
+import java.nio.charset.StandardCharsets;
+import java.util.*;
+import java.util.zip.ZipEntry;
+import java.util.zip.ZipOutputStream;
+
+@Service
+@RequiredArgsConstructor
+@Slf4j
+public class SegmentationTrainingExportService {
+
+    private final TranscriptionBlockRepository blockRepository;
+    private final AnnotationRepository annotationRepository;
+    private final DocumentRepository documentRepository;
+    private final FileService fileService;
+
+    public List<TranscriptionBlock> querySegmentationBlocks() {
+        return blockRepository.findSegmentationBlocks();
+    }
+
+    public StreamingResponseBody exportToZip() {
+        List<TranscriptionBlock> blocks = querySegmentationBlocks();
+        if (blocks.isEmpty()) {
+            return out -> {};
+        }
+
+        // Group by documentId so we download each PDF only once
+        Map<UUID, List<TranscriptionBlock>> byDoc = new LinkedHashMap<>();
+        for (TranscriptionBlock b : blocks) {
+            byDoc.computeIfAbsent(b.getDocumentId(), k -> new ArrayList<>()).add(b);
+        }
+
+        // Pre-fetch annotations keyed by id
+        Map<UUID, DocumentAnnotation> annotations = new HashMap<>();
+        for (TranscriptionBlock b : blocks) {
+            annotationRepository.findById(b.getAnnotationId())
+                    .ifPresent(a -> annotations.put(a.getId(), a));
+        }
+
+        // Pre-fetch documents keyed by id
+        Map<UUID, Document> documents = new HashMap<>();
+        for (UUID docId : byDoc.keySet()) {
+            documentRepository.findById(docId).ifPresent(d -> documents.put(d.getId(), d));
+        }
+
+        return out -> {
+            try (ZipOutputStream zip = new ZipOutputStream(out)) {
+                for (Map.Entry<UUID, List<TranscriptionBlock>> entry : byDoc.entrySet()) {
+                    UUID docId = entry.getKey();
+                    Document doc = documents.get(docId);
+                    if (doc == null || doc.getFilePath() == null) {
+                        log.warn("Skipping document {} — no file path", docId);
+                        continue;
+                    }
+
+                    byte[] pdfBytes;
+                    try {
+                        pdfBytes = fileService.downloadFileBytes(doc.getFilePath());
+                    } catch (FileService.StorageFileNotFoundException | IOException e) {
+                        log.warn("Skipping document {} — S3 download failed: {}", docId, e.getMessage());
+                        continue;
+                    }
+
+                    // Group blocks by page number for this document
+                    Map<Integer, List<TranscriptionBlock>> byPage = new LinkedHashMap<>();
+                    for (TranscriptionBlock b : entry.getValue()) {
+                        DocumentAnnotation ann = annotations.get(b.getAnnotationId());
+                        if (ann != null) {
+                            byPage.computeIfAbsent(ann.getPageNumber(), k -> new ArrayList<>()).add(b);
+                        }
+                    }
+
+                    try (PDDocument pdf = Loader.loadPDF(pdfBytes)) {
+                        PDFRenderer renderer = new PDFRenderer(pdf);
+                        for (Map.Entry<Integer, List<TranscriptionBlock>> pageEntry : byPage.entrySet()) {
+                            int pageNumber = pageEntry.getKey();
+                            int pageIdx = pageNumber - 1;
+                            if (pageIdx < 0 || pageIdx >= pdf.getNumberOfPages()) continue;
+
+                            BufferedImage pageImage = renderer.renderImageWithDPI(pageIdx, 300);
+                            String basename = "page-" + docId + "-" + pageNumber;
+
+                            // Collect annotations for this page
+                            List<DocumentAnnotation> pageAnnotations = new ArrayList<>();
+                            for (TranscriptionBlock b : pageEntry.getValue()) {
+                                DocumentAnnotation ann = annotations.get(b.getAnnotationId());
+                                if (ann != null) pageAnnotations.add(ann);
+                            }
+
+                            writePngEntry(zip, basename, pageImage);
+                            writePageXmlEntry(zip, basename, pageImage, pageAnnotations);
+                        }
+                    } catch (Exception e) {
+                        log.warn("Skipping document {} — rendering failed: {}", docId, e.getMessage());
+                    }
+                }
+            }
+        };
+    }
+
+    private void writePngEntry(ZipOutputStream zip, String basename, BufferedImage image) throws IOException {
+        zip.putNextEntry(new ZipEntry(basename + ".png"));
+        ImageIO.write(image, "PNG", zip);
+        zip.closeEntry();
+    }
+
+    private void writePageXmlEntry(ZipOutputStream zip, String basename,
+                                   BufferedImage pageImage,
+                                   List<DocumentAnnotation> annotations) throws IOException {
+        int imgW = pageImage.getWidth();
+        int imgH = pageImage.getHeight();
+
+        StringBuilder regions = new StringBuilder();
+        for (DocumentAnnotation ann : annotations) {
+            String coords = buildPolygonCoords(ann, imgW, imgH);
+            String regionId = ann.getId().toString();
+            regions.append("      <TextRegion id=\"").append(regionId).append("\">\n");
+            regions.append("        <Coords points=\"").append(coords).append("\"/>\n");
+            regions.append("      </TextRegion>\n");
+        }
+
+        String xml = "<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n"
+                + "<PcGts xmlns=\"http://schema.primaresearch.org/PAGE/gts/pagecontent/2019-07-15\">\n"
+                + "  <Page imageFilename=\"" + basename + ".png\""
+                + " imageWidth=\"" + imgW + "\""
+                + " imageHeight=\"" + imgH + "\">\n"
+                + regions
+                + "  </Page>\n"
+                + "</PcGts>\n";
+
+        zip.putNextEntry(new ZipEntry(basename + ".xml"));
+        zip.write(xml.getBytes(StandardCharsets.UTF_8));
+        zip.closeEntry();
+    }
+
+    String buildPolygonCoords(DocumentAnnotation ann, int imgW, int imgH) {
+        List<List<Double>> polygon = ann.getPolygon();
+        if (polygon != null && !polygon.isEmpty()) {
+            // Use explicit polygon — de-normalize to pixel coordinates
+            StringBuilder sb = new StringBuilder();
+            for (List<Double> pt : polygon) {
+                if (sb.length() > 0) sb.append(' ');
+                int px = (int) (pt.get(0) * imgW);
+                int py = (int) (pt.get(1) * imgH);
+                sb.append(px).append(',').append(py);
+            }
+            return sb.toString();
+        }
+        // Fall back to bounding box from x/y/width/height
+        int x = (int) (ann.getX() * imgW);
+        int y = (int) (ann.getY() * imgH);
+        int w = (int) (ann.getWidth() * imgW);
+        int h = (int) (ann.getHeight() * imgH);
+        return x + "," + y + " " + (x + w) + "," + y + " " + (x + w) + "," + (y + h) + " " + x + "," + (y + h);
+    }
+}

backend/src/main/java/org/raddatz/familienarchiv/service/TrainingDataExportService.java

@@ -0,0 +1,173 @@
+package org.raddatz.familienarchiv.service;
+
+import lombok.RequiredArgsConstructor;
+import lombok.extern.slf4j.Slf4j;
+import org.apache.pdfbox.Loader;
+import org.apache.pdfbox.pdmodel.PDDocument;
+import org.apache.pdfbox.rendering.PDFRenderer;
+import org.raddatz.familienarchiv.model.Document;
+import org.raddatz.familienarchiv.model.DocumentAnnotation;
+import org.raddatz.familienarchiv.model.TranscriptionBlock;
+import org.raddatz.familienarchiv.repository.AnnotationRepository;
+import org.raddatz.familienarchiv.repository.DocumentRepository;
+import org.raddatz.familienarchiv.repository.TranscriptionBlockRepository;
+import org.springframework.stereotype.Service;
+import org.springframework.web.servlet.mvc.method.annotation.StreamingResponseBody;
+
+import javax.imageio.ImageIO;
+import java.awt.image.BufferedImage;
+import java.io.ByteArrayInputStream;
+import java.io.IOException;
+import java.nio.charset.StandardCharsets;
+import java.util.*;
+import java.util.zip.ZipEntry;
+import java.util.zip.ZipOutputStream;
+
+@Service
+@RequiredArgsConstructor
+@Slf4j
+public class TrainingDataExportService {
+
+    private final TranscriptionBlockRepository blockRepository;
+    private final AnnotationRepository annotationRepository;
+    private final DocumentRepository documentRepository;
+    private final FileService fileService;
+
+    public List<TranscriptionBlock> queryEligibleBlocks() {
+        return blockRepository.findEligibleKurrentBlocks();
+    }
+
+    public StreamingResponseBody exportToZip() {
+        // Collect all data before entering the lambda — no open DB txn during streaming
+        List<TranscriptionBlock> blocks = queryEligibleBlocks();
+        if (blocks.isEmpty()) {
+            return out -> {}; // caller checks isEmpty() for 204 response
+        }
+
+        // Group blocks by documentId so we only download each PDF once
+        Map<UUID, List<TranscriptionBlock>> byDoc = new LinkedHashMap<>();
+        for (TranscriptionBlock b : blocks) {
+            byDoc.computeIfAbsent(b.getDocumentId(), k -> new ArrayList<>()).add(b);
+        }
+
+        // Pre-fetch annotations keyed by id
+        Map<UUID, DocumentAnnotation> annotations = new HashMap<>();
+        for (TranscriptionBlock b : blocks) {
+            annotationRepository.findById(b.getAnnotationId())
+                    .ifPresent(a -> annotations.put(a.getId(), a));
+        }
+
+        // Pre-fetch documents keyed by id
+        Map<UUID, Document> documents = new HashMap<>();
+        for (UUID docId : byDoc.keySet()) {
+            documentRepository.findById(docId).ifPresent(d -> documents.put(d.getId(), d));
+        }
+
+        return out -> {
+            try (ZipOutputStream zip = new ZipOutputStream(out)) {
+                for (Map.Entry<UUID, List<TranscriptionBlock>> entry : byDoc.entrySet()) {
+                    UUID docId = entry.getKey();
+                    Document doc = documents.get(docId);
+                    if (doc == null || doc.getFilePath() == null) {
+                        log.warn("Skipping document {} — no file path", docId);
+                        continue;
+                    }
+
+                    byte[] pdfBytes;
+                    try {
+                        pdfBytes = fileService.downloadFileBytes(doc.getFilePath());
+                    } catch (FileService.StorageFileNotFoundException | IOException e) {
+                        log.warn("Skipping document {} — S3 download failed: {}", docId, e.getMessage());
+                        continue;
+                    }
+
+                    try (PDDocument pdf = Loader.loadPDF(pdfBytes)) {
+                        PDFRenderer renderer = new PDFRenderer(pdf);
+                        for (TranscriptionBlock block : entry.getValue()) {
+                            DocumentAnnotation ann = annotations.get(block.getAnnotationId());
+                            if (ann == null) continue;
+
+                            int pageIdx = ann.getPageNumber() - 1; // pageNumber is 1-based
+                            if (pageIdx < 0 || pageIdx >= pdf.getNumberOfPages()) continue;
+
+                            BufferedImage pageImage = renderPageImage(renderer, pageIdx);
+                            BufferedImage cropped = cropBlockImage(pageImage, ann);
+
+                            writeTrainingPair(zip, block.getId(), cropped, block.getText());
+                        }
+                    } catch (Exception e) {
+                        log.warn("Skipping document {} — rendering failed: {}", docId, e.getMessage());
+                    }
+                }
+            }
+        };
+    }
+
+    BufferedImage renderPageImage(PDFRenderer renderer, int pageIdx) throws IOException {
+        return renderer.renderImageWithDPI(pageIdx, 300);
+    }
+
+    BufferedImage cropBlockImage(BufferedImage page, DocumentAnnotation ann) {
+        int imgW = page.getWidth();
+        int imgH = page.getHeight();
+
+        int x = (int) (ann.getX() * imgW);
+        int y = (int) (ann.getY() * imgH);
+        int w = (int) (ann.getWidth() * imgW);
+        int h = (int) (ann.getHeight() * imgH);
+
+        // Clamp to image bounds
+        x = Math.max(0, Math.min(x, imgW - 1));
+        y = Math.max(0, Math.min(y, imgH - 1));
+        w = Math.max(1, Math.min(w, imgW - x));
+        h = Math.max(1, Math.min(h, imgH - y));
+
+        return page.getSubimage(x, y, w, h);
+    }
+
+    void writeTrainingPair(ZipOutputStream zip, UUID blockId, BufferedImage image, String text) throws IOException {
+        String base = blockId.toString();
+        int w = image.getWidth();
+        int h = image.getHeight();
+        // Baseline at 75 % height — typical text baseline position in a cropped line image
+        int baselineY = (h * 3) / 4;
+
+        // Write PNG
+        zip.putNextEntry(new ZipEntry(base + ".png"));
+        ImageIO.write(image, "PNG", zip);
+        zip.closeEntry();
+
+        // Write PAGE XML (Kraken 7+ dropped the legacy "path" format)
+        String safeText = escapeXml(text != null ? text : "");
+        String xml = String.format(
+                "<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n" +
+                "<PcGts xmlns=\"http://schema.primaresearch.org/PAGE/gts/pagecontent/2019-07-15\">\n" +
+                "  <Metadata><Creator>familienarchiv</Creator></Metadata>\n" +
+                "  <Page imageFilename=\"%s.png\" imageWidth=\"%d\" imageHeight=\"%d\">\n" +
+                "    <TextRegion id=\"r0\" type=\"paragraph\">\n" +
+                "      <Coords points=\"0,0 %d,0 %d,%d 0,%d\"/>\n" +
+                "      <TextLine id=\"l0\">\n" +
+                "        <Coords points=\"0,0 %d,0 %d,%d 0,%d\"/>\n" +
+                "        <Baseline points=\"0,%d %d,%d\"/>\n" +
+                "        <TextEquiv><Unicode>%s</Unicode></TextEquiv>\n" +
+                "      </TextLine>\n" +
+                "    </TextRegion>\n" +
+                "  </Page>\n" +
+                "</PcGts>\n",
+                base, w, h,
+                w - 1, w - 1, h - 1, h - 1,
+                w - 1, w - 1, h - 1, h - 1,
+                baselineY, w - 1, baselineY,
+                safeText);
+
+        zip.putNextEntry(new ZipEntry(base + ".xml"));
+        zip.write(xml.getBytes(StandardCharsets.UTF_8));
+        zip.closeEntry();
+    }
+
+    private static String escapeXml(String text) {
+        return text.replace("&", "&amp;")
+                   .replace("<", "&lt;")
+                   .replace(">", "&gt;");
+    }
+}

backend/src/main/java/org/raddatz/familienarchiv/service/TranscriptionService.java

@@ -94,6 +94,27 @@ public class TranscriptionService {
         return saved;
     }
 
+    /**
+     * Upsert an OCR transcription block for a pre-existing annotation (guided OCR mode).
+     * If the annotation already has a MANUAL block, it is left unchanged.
+     * If it has an OCR block, the text is updated in-place.
+     * If it has no block yet, a new OCR block is created.
+     */
+    @Transactional
+    public TranscriptionBlock upsertGuidedBlock(UUID documentId, UUID annotationId,
+                                                 String text, UUID userId) {
+        return blockRepository.findByAnnotationId(annotationId).map(existing -> {
+            if (existing.getSource() == BlockSource.MANUAL && !existing.getText().isBlank()) {
+                return existing; // never overwrite non-empty manual transcription
+            }
+            existing.setText(sanitizeText(text));
+            existing.setUpdatedBy(userId);
+            TranscriptionBlock saved = blockRepository.save(existing);
+            saveVersion(saved, userId);
+            return saved;
+        }).orElseGet(() -> createOcrBlock(documentId, annotationId, text, 0, userId));
+    }
+
     @Transactional
     public TranscriptionBlock updateBlock(UUID documentId, UUID blockId,
                                           UpdateTranscriptionBlockDTO dto, UUID userId) {

backend/src/main/resources/db/migration/V29__add_document_training_labels.sql

@@ -0,0 +1,5 @@
+CREATE TABLE document_training_labels (
+    document_id UUID NOT NULL REFERENCES documents(id) ON DELETE CASCADE,
+    label       VARCHAR(50) NOT NULL,
+    PRIMARY KEY (document_id, label)
+);

backend/src/main/resources/db/migration/V30__add_ocr_training_runs.sql

@@ -0,0 +1,16 @@
+CREATE TABLE ocr_training_runs (
+    id             UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    status         VARCHAR(20)  NOT NULL DEFAULT 'RUNNING',
+    block_count    INT          NOT NULL,
+    document_count INT          NOT NULL,
+    model_name     VARCHAR(100) NOT NULL,
+    error_message  TEXT,
+    triggered_by   UUID REFERENCES users(id) ON DELETE SET NULL,
+    created_at     TIMESTAMPTZ  NOT NULL DEFAULT now(),
+    completed_at   TIMESTAMPTZ
+);
+
+-- Enforce single active run at the DB layer (application check is the UX layer)
+CREATE UNIQUE INDEX idx_ocr_training_runs_one_running
+    ON ocr_training_runs (status)
+    WHERE status = 'RUNNING';

backend/src/main/resources/db/migration/V31__make_transcription_block_text_nullable.sql

@@ -0,0 +1,5 @@
+-- Intentional: segmentation-only blocks have no text.
+-- This migration is irreversible without a data cleanup step
+-- (cannot re-add NOT NULL if null rows exist).
+ALTER TABLE transcription_blocks ALTER COLUMN text DROP NOT NULL;
+ALTER TABLE transcription_blocks ALTER COLUMN text SET DEFAULT '';

backend/src/main/resources/db/migration/V32__add_training_metrics.sql

@@ -0,0 +1,5 @@
+ALTER TABLE ocr_training_runs
+    ADD COLUMN cer      DOUBLE PRECISION,
+    ADD COLUMN loss     DOUBLE PRECISION,
+    ADD COLUMN accuracy DOUBLE PRECISION,
+    ADD COLUMN epochs   INT;

backend/src/test/java/org/raddatz/familienarchiv/controller/AnnotationControllerTest.java

@@ -123,15 +123,19 @@ class AnnotationControllerTest {
 
     @Test
     @WithMockUser(authorities = "ANNOTATE_ALL")
-    void createAnnotation_returns409_whenOverlap() throws Exception {
+    void createAnnotation_returns201_whenAnnotationsOverlap() throws Exception {
+        // Overlapping annotations are allowed — historical letter lines often intersect
+        UUID docId = UUID.randomUUID();
+        DocumentAnnotation saved = DocumentAnnotation.builder()
+                .id(UUID.randomUUID()).documentId(docId).pageNumber(1)
+                .x(0.1).y(0.1).width(0.3).height(0.3).color("#ff0000").build();
         when(documentService.getDocumentById(any())).thenReturn(Document.builder().build());
-        when(annotationService.createAnnotation(any(), any(), any(), any()))
+        when(annotationService.createAnnotation(any(), any(), any(), any())).thenReturn(saved);
-                .thenThrow(DomainException.conflict(ErrorCode.ANNOTATION_OVERLAP, "Overlap"));
 
-        mockMvc.perform(post("/api/documents/" + UUID.randomUUID() + "/annotations")
+        mockMvc.perform(post("/api/documents/" + docId + "/annotations")
                         .contentType(MediaType.APPLICATION_JSON)
                         .content(ANNOTATION_JSON))
-                .andExpect(status().isConflict());
+                .andExpect(status().isCreated());
     }
 
     // ─── DELETE /api/documents/{documentId}/annotations/{annotationId} ─────────

backend/src/test/java/org/raddatz/familienarchiv/controller/DocumentControllerTest.java

@@ -14,6 +14,7 @@ import org.raddatz.familienarchiv.service.FileService;
 import org.springframework.beans.factory.annotation.Autowired;
 import org.springframework.boot.webmvc.test.autoconfigure.WebMvcTest;
 import org.raddatz.familienarchiv.config.SecurityConfig;
+import org.springframework.http.MediaType;
 import org.springframework.boot.autoconfigure.aop.AopAutoConfiguration;
 import org.springframework.context.annotation.Import;
 import org.springframework.security.test.context.support.WithMockUser;
@@ -33,6 +34,7 @@ import static org.mockito.Mockito.verify;
 import static org.mockito.Mockito.when;
 import static org.springframework.test.web.servlet.request.MockMvcRequestBuilders.get;
 import static org.springframework.test.web.servlet.request.MockMvcRequestBuilders.multipart;
+import static org.springframework.test.web.servlet.request.MockMvcRequestBuilders.patch;
 import static org.springframework.test.web.servlet.result.MockMvcResultMatchers.jsonPath;
 import static org.springframework.test.web.servlet.result.MockMvcResultMatchers.status;
 
@@ -485,6 +487,58 @@ class DocumentControllerTest {
                 .andExpect(jsonPath("$[0].editorName").value("Emma Müller"));
     }
 
+    // ─── PATCH /api/documents/{id}/training-labels ───────────────────────────
+
+    @Test
+    void patchTrainingLabels_returns401_whenUnauthenticated() throws Exception {
+        mockMvc.perform(patch("/api/documents/" + UUID.randomUUID() + "/training-labels")
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"label\":\"KURRENT_RECOGNITION\",\"enrolled\":true}"))
+                .andExpect(status().isUnauthorized());
+    }
+
+    @Test
+    @WithMockUser
+    void patchTrainingLabels_returns403_whenMissingWritePermission() throws Exception {
+        mockMvc.perform(patch("/api/documents/" + UUID.randomUUID() + "/training-labels")
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"label\":\"KURRENT_RECOGNITION\",\"enrolled\":true}"))
+                .andExpect(status().isForbidden());
+    }
+
+    @Test
+    @WithMockUser(authorities = "WRITE_ALL")
+    void patchTrainingLabels_returns204_whenAddingLabel() throws Exception {
+        UUID id = UUID.randomUUID();
+        mockMvc.perform(patch("/api/documents/" + id + "/training-labels")
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"label\":\"KURRENT_RECOGNITION\",\"enrolled\":true}"))
+                .andExpect(status().isNoContent());
+
+        verify(documentService).addTrainingLabel(eq(id), any());
+    }
+
+    @Test
+    @WithMockUser(authorities = "WRITE_ALL")
+    void patchTrainingLabels_returns204_whenRemovingLabel() throws Exception {
+        UUID id = UUID.randomUUID();
+        mockMvc.perform(patch("/api/documents/" + id + "/training-labels")
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"label\":\"KURRENT_SEGMENTATION\",\"enrolled\":false}"))
+                .andExpect(status().isNoContent());
+
+        verify(documentService).removeTrainingLabel(eq(id), any());
+    }
+
+    @Test
+    @WithMockUser(authorities = "WRITE_ALL")
+    void patchTrainingLabels_returns400_whenUnknownLabel() throws Exception {
+        mockMvc.perform(patch("/api/documents/" + UUID.randomUUID() + "/training-labels")
+                        .contentType(MediaType.APPLICATION_JSON)
+                        .content("{\"label\":\"UNKNOWN_GARBAGE\",\"enrolled\":true}"))
+                .andExpect(status().isBadRequest());
+    }
+
     // ─── GET /api/documents/{id}/versions/{versionId} ────────────────────────
 
     @Test

backend/src/test/java/org/raddatz/familienarchiv/controller/OcrControllerTest.java

@@ -24,6 +24,7 @@ import java.util.List;
 import java.util.UUID;
 
 import static org.mockito.ArgumentMatchers.any;
+import static org.mockito.ArgumentMatchers.anyBoolean;
 import static org.mockito.ArgumentMatchers.eq;
 import static org.mockito.Mockito.when;
 import static org.springframework.test.web.servlet.request.MockMvcRequestBuilders.get;
@@ -43,15 +44,18 @@ class OcrControllerTest {
     @MockitoBean OcrProgressService ocrProgressService;
     @MockitoBean UserService userService;
     @MockitoBean CustomUserDetailsService customUserDetailsService;
+    @MockitoBean TrainingDataExportService trainingDataExportService;
+    @MockitoBean SegmentationTrainingExportService segmentationTrainingExportService;
+    @MockitoBean OcrTrainingService ocrTrainingService;
 
     @Test
     @WithMockUser(authorities = "WRITE_ALL")
     void triggerOcr_returns202_withJobId() throws Exception {
         UUID docId = UUID.randomUUID();
         UUID jobId = UUID.randomUUID();
-        TriggerOcrDTO dto = new TriggerOcrDTO(ScriptType.TYPEWRITER);
+        TriggerOcrDTO dto = new TriggerOcrDTO(ScriptType.TYPEWRITER, false);
 
-        when(ocrService.startOcr(eq(docId), eq(ScriptType.TYPEWRITER), any())).thenReturn(jobId);
+        when(ocrService.startOcr(eq(docId), eq(ScriptType.TYPEWRITER), any(), anyBoolean())).thenReturn(jobId);
 
         mockMvc.perform(post("/api/documents/{id}/ocr", docId)
                         .contentType(MediaType.APPLICATION_JSON)
@@ -64,7 +68,7 @@ class OcrControllerTest {
     @WithMockUser(authorities = "WRITE_ALL")
     void triggerOcr_returns400_whenDocumentNotUploaded() throws Exception {
         UUID docId = UUID.randomUUID();
-        when(ocrService.startOcr(eq(docId), any(), any()))
+        when(ocrService.startOcr(eq(docId), any(), any(), anyBoolean()))
                 .thenThrow(DomainException.badRequest(ErrorCode.OCR_DOCUMENT_NOT_UPLOADED, "Not uploaded"));
 
         mockMvc.perform(post("/api/documents/{id}/ocr", docId)
@@ -121,6 +125,109 @@ class OcrControllerTest {
                 .andExpect(jsonPath("$.jobId").value(jobId.toString()));
     }
 
+    // ─── GET /api/ocr/training-data/export ───────────────────────────────────
+
+    @Test
+    void exportTrainingData_returns401_whenUnauthenticated() throws Exception {
+        mockMvc.perform(get("/api/ocr/training-data/export"))
+                .andExpect(status().isUnauthorized());
+    }
+
+    @Test
+    @WithMockUser(authorities = "READ_ALL")
+    void exportTrainingData_returns403_whenNotAdmin() throws Exception {
+        mockMvc.perform(get("/api/ocr/training-data/export"))
+                .andExpect(status().isForbidden());
+    }
+
+    @Test
+    @WithMockUser(authorities = "ADMIN")
+    void exportTrainingData_returns204_whenNoEligibleBlocks() throws Exception {
+        when(trainingDataExportService.queryEligibleBlocks()).thenReturn(List.of());
+
+        mockMvc.perform(get("/api/ocr/training-data/export"))
+                .andExpect(status().isNoContent());
+    }
+
+    @Test
+    @WithMockUser(authorities = "ADMIN")
+    void exportTrainingData_returns200_withZipContentType_whenBlocksExist() throws Exception {
+        org.raddatz.familienarchiv.model.TranscriptionBlock block =
+                org.raddatz.familienarchiv.model.TranscriptionBlock.builder()
+                        .id(UUID.randomUUID()).documentId(UUID.randomUUID())
+                        .annotationId(UUID.randomUUID()).text("x").sortOrder(0).build();
+        when(trainingDataExportService.queryEligibleBlocks()).thenReturn(List.of(block));
+        when(trainingDataExportService.exportToZip()).thenReturn(out -> {});
+
+        mockMvc.perform(get("/api/ocr/training-data/export"))
+                .andExpect(status().isOk())
+                .andExpect(result ->
+                    org.assertj.core.api.Assertions.assertThat(
+                        result.getResponse().getContentType()).contains("application/zip"));
+    }
+
+    // ─── POST /api/ocr/train ───────────────────────────────────────────────────
+
+    @Test
+    void triggerTraining_returns401_whenUnauthenticated() throws Exception {
+        mockMvc.perform(post("/api/ocr/train"))
+                .andExpect(status().isUnauthorized());
+    }
+
+    @Test
+    @WithMockUser(authorities = "READ_ALL")
+    void triggerTraining_returns403_whenNotAdmin() throws Exception {
+        mockMvc.perform(post("/api/ocr/train"))
+                .andExpect(status().isForbidden());
+    }
+
+    @Test
+    @WithMockUser(authorities = "ADMIN")
+    void triggerTraining_returns409_whenRunAlreadyRunning() throws Exception {
+        when(ocrTrainingService.triggerTraining(any()))
+                .thenThrow(DomainException.conflict(ErrorCode.TRAINING_ALREADY_RUNNING, "Already running"));
+
+        mockMvc.perform(post("/api/ocr/train"))
+                .andExpect(status().isConflict());
+    }
+
+    @Test
+    @WithMockUser(authorities = "ADMIN")
+    void triggerTraining_returns201_withRunInfo() throws Exception {
+        UUID runId = UUID.randomUUID();
+        OcrTrainingRun run = OcrTrainingRun.builder()
+                .id(runId).status(TrainingStatus.DONE)
+                .blockCount(10).documentCount(3).modelName("german_kurrent").build();
+        when(ocrTrainingService.triggerTraining(any())).thenReturn(run);
+
+        mockMvc.perform(post("/api/ocr/train"))
+                .andExpect(status().isCreated())
+                .andExpect(jsonPath("$.status").value("DONE"))
+                .andExpect(jsonPath("$.blockCount").value(10));
+    }
+
+    // ─── GET /api/ocr/training-info ───────────────────────────────────────────
+
+    @Test
+    @WithMockUser(authorities = "READ_ALL")
+    void getTrainingInfo_returns403_whenNotAdmin() throws Exception {
+        mockMvc.perform(get("/api/ocr/training-info"))
+                .andExpect(status().isForbidden());
+    }
+
+    @Test
+    @WithMockUser(authorities = "ADMIN")
+    void getTrainingInfo_returns200_withInfo() throws Exception {
+        OcrTrainingService.TrainingInfoResponse info =
+                new OcrTrainingService.TrainingInfoResponse(5, 20, 2, 3, true, null, List.of());
+        when(ocrTrainingService.getTrainingInfo()).thenReturn(info);
+
+        mockMvc.perform(get("/api/ocr/training-info"))
+                .andExpect(status().isOk())
+                .andExpect(jsonPath("$.availableBlocks").value(5))
+                .andExpect(jsonPath("$.ocrServiceAvailable").value(true));
+    }
+
     @Test
     @WithMockUser(authorities = "READ_ALL")
     void getDocumentOcrStatus_returnsNone_whenNoOcrJobExists() throws Exception {

backend/src/test/java/org/raddatz/familienarchiv/repository/MigrationIntegrationTest.java

@@ -0,0 +1,136 @@
+package org.raddatz.familienarchiv.repository;
+
+import jakarta.persistence.EntityManager;
+import org.junit.jupiter.api.Test;
+import org.raddatz.familienarchiv.PostgresContainerConfig;
+import org.raddatz.familienarchiv.config.FlywayConfig;
+import org.raddatz.familienarchiv.model.Document;
+import org.raddatz.familienarchiv.model.DocumentStatus;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.boot.data.jpa.test.autoconfigure.DataJpaTest;
+import org.springframework.boot.jdbc.test.autoconfigure.AutoConfigureTestDatabase;
+import org.springframework.context.annotation.Import;
+import org.springframework.dao.DataIntegrityViolationException;
+import org.springframework.jdbc.core.JdbcTemplate;
+import org.springframework.transaction.annotation.Propagation;
+import org.springframework.transaction.annotation.Transactional;
+
+import java.util.UUID;
+
+import static org.assertj.core.api.Assertions.assertThat;
+import static org.assertj.core.api.Assertions.assertThatThrownBy;
+
+/**
+ * Integration tests that verify DB-level constraints introduced in the OCR pipeline migrations
+ * are actually enforced by PostgreSQL. These tests exercise constraints that cannot be verified
+ * by unit tests alone.
+ */
+@DataJpaTest
+@AutoConfigureTestDatabase(replace = AutoConfigureTestDatabase.Replace.NONE)
+@Import({PostgresContainerConfig.class, FlywayConfig.class})
+class MigrationIntegrationTest {
+
+    @Autowired JdbcTemplate jdbc;
+    @Autowired DocumentRepository documentRepository;
+    @Autowired EntityManager em;
+
+    // ─── V23: chk_annotation_polygon_quad CHECK constraint ───────────────────
+
+    @Test
+    void v23_polygonCheckConstraint_rejectsNonQuadrilateral() {
+        UUID docId = createDocument();
+
+        // A 3-point polygon violates chk_annotation_polygon_quad (must be exactly 4 points or NULL)
+        assertThatThrownBy(() ->
+                jdbc.update(
+                        """
+                        INSERT INTO document_annotations
+                            (id, document_id, page_number, x, y, width, height, color, polygon)
+                        VALUES (gen_random_uuid(), ?, 1, 0.1, 0.1, 0.5, 0.1, '#00C7B1',
+                                '[[0.1,0.1],[0.9,0.1],[0.9,0.2]]'::jsonb)
+                        """,
+                        docId)
+        ).isInstanceOf(DataIntegrityViolationException.class);
+    }
+
+    @Test
+    void v23_polygonCheckConstraint_allowsNullPolygon() {
+        UUID docId = createDocument();
+
+        int rows = jdbc.update(
+                """
+                INSERT INTO document_annotations
+                    (id, document_id, page_number, x, y, width, height, color, polygon)
+                VALUES (gen_random_uuid(), ?, 1, 0.1, 0.1, 0.5, 0.1, '#00C7B1', NULL)
+                """,
+                docId);
+
+        assertThat(rows).isEqualTo(1);
+    }
+
+    @Test
+    void v23_polygonCheckConstraint_allowsQuadrilateral() {
+        UUID docId = createDocument();
+
+        int rows = jdbc.update(
+                """
+                INSERT INTO document_annotations
+                    (id, document_id, page_number, x, y, width, height, color, polygon)
+                VALUES (gen_random_uuid(), ?, 1, 0.1, 0.1, 0.5, 0.1, '#00C7B1',
+                        '[[0.1,0.1],[0.9,0.1],[0.9,0.2],[0.1,0.2]]'::jsonb)
+                """,
+                docId);
+
+        assertThat(rows).isEqualTo(1);
+    }
+
+    // ─── V30: idx_ocr_training_runs_one_running partial unique index ──────────
+
+    @Test
+    @Transactional(propagation = Propagation.NOT_SUPPORTED)
+    void v30_partialUniqueIndex_preventsTwoRunningTrainingRuns() {
+        jdbc.update("""
+                INSERT INTO ocr_training_runs (id, status, block_count, document_count, model_name)
+                VALUES (gen_random_uuid(), 'RUNNING', 10, 2, 'kurrent_v1')
+                """);
+
+        // A second RUNNING row violates the partial unique index
+        assertThatThrownBy(() ->
+                jdbc.update("""
+                        INSERT INTO ocr_training_runs (id, status, block_count, document_count, model_name)
+                        VALUES (gen_random_uuid(), 'RUNNING', 5, 1, 'kurrent_v1')
+                        """)
+        ).isInstanceOf(DataIntegrityViolationException.class);
+
+        // Clean up — runs outside the DataJpaTest transaction, so must be explicit
+        jdbc.update("DELETE FROM ocr_training_runs WHERE status = 'RUNNING'");
+    }
+
+    @Test
+    void v30_partialUniqueIndex_allowsMultipleDoneRuns() {
+        int rows1 = jdbc.update("""
+                INSERT INTO ocr_training_runs (id, status, block_count, document_count, model_name)
+                VALUES (gen_random_uuid(), 'DONE', 10, 2, 'kurrent_v1')
+                """);
+        int rows2 = jdbc.update("""
+                INSERT INTO ocr_training_runs (id, status, block_count, document_count, model_name)
+                VALUES (gen_random_uuid(), 'DONE', 15, 3, 'kurrent_v2')
+                """);
+
+        assertThat(rows1).isEqualTo(1);
+        assertThat(rows2).isEqualTo(1);
+    }
+
+    // ─── helpers ─────────────────────────────────────────────────────────────
+
+    private UUID createDocument() {
+        Document doc = documentRepository.save(Document.builder()
+                .title("Testdokument")
+                .originalFilename("test.pdf")
+                .status(DocumentStatus.UPLOADED)
+                .build());
+        // Flush so the row is visible to subsequent JdbcTemplate queries within the same transaction
+        em.flush();
+        return doc.getId();
+    }
+}

backend/src/test/java/org/raddatz/familienarchiv/repository/TrainingBlockQueryTest.java

@@ -0,0 +1,135 @@
+package org.raddatz.familienarchiv.repository;
+
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Test;
+import org.raddatz.familienarchiv.PostgresContainerConfig;
+import org.raddatz.familienarchiv.config.FlywayConfig;
+import org.raddatz.familienarchiv.model.*;
+import java.util.Set;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.boot.jdbc.test.autoconfigure.AutoConfigureTestDatabase;
+import org.springframework.boot.data.jpa.test.autoconfigure.DataJpaTest;
+import org.springframework.context.annotation.Import;
+
+import java.util.List;
+import java.util.UUID;
+
+import static org.assertj.core.api.Assertions.assertThat;
+
+@DataJpaTest
+@AutoConfigureTestDatabase(replace = AutoConfigureTestDatabase.Replace.NONE)
+@Import({PostgresContainerConfig.class, FlywayConfig.class})
+class TrainingBlockQueryTest {
+
+    @Autowired TranscriptionBlockRepository blockRepository;
+    @Autowired DocumentRepository documentRepository;
+    @Autowired AnnotationRepository annotationRepository;
+
+    private UUID kurrentDocId;
+    private UUID typewriterDocId;
+    private UUID kurrentAnnotationId;
+    private UUID typewriterAnnotationId;
+
+    @BeforeEach
+    void setUp() {
+        Document kurrentDoc = documentRepository.save(Document.builder()
+                .title("Kurrent Brief")
+                .originalFilename("kurrent.pdf")
+                .status(DocumentStatus.UPLOADED)
+                .trainingLabels(new java.util.HashSet<>(Set.of(TrainingLabel.KURRENT_RECOGNITION)))
+                .build());
+        kurrentDocId = kurrentDoc.getId();
+
+        Document typewriterDoc = documentRepository.save(Document.builder()
+                .title("Getippter Brief")
+                .originalFilename("typed.pdf")
+                .status(DocumentStatus.UPLOADED)
+                .build());
+        typewriterDocId = typewriterDoc.getId();
+
+        kurrentAnnotationId = annotationRepository.save(annotation(kurrentDocId)).getId();
+        typewriterAnnotationId = annotationRepository.save(annotation(typewriterDocId)).getId();
+    }
+
+    @Test
+    void findEligibleKurrentBlocks_includesReviewedManualBlock() {
+        blockRepository.save(block(kurrentDocId, kurrentAnnotationId, BlockSource.MANUAL, true));
+
+        List<TranscriptionBlock> result = blockRepository.findEligibleKurrentBlocks();
+
+        assertThat(result).hasSize(1);
+        assertThat(result.get(0).getSource()).isEqualTo(BlockSource.MANUAL);
+    }
+
+    @Test
+    void findEligibleKurrentBlocks_excludesUnreviewedManualBlock() {
+        // MANUAL blocks filled by guided OCR but not yet reviewed should not count as checked text
+        blockRepository.save(block(kurrentDocId, kurrentAnnotationId, BlockSource.MANUAL, false));
+
+        List<TranscriptionBlock> result = blockRepository.findEligibleKurrentBlocks();
+
+        assertThat(result).isEmpty();
+    }
+
+    @Test
+    void findEligibleKurrentBlocks_includesReviewedOcrBlock() {
+        blockRepository.save(block(kurrentDocId, kurrentAnnotationId, BlockSource.OCR, true));
+
+        List<TranscriptionBlock> result = blockRepository.findEligibleKurrentBlocks();
+
+        assertThat(result).hasSize(1);
+        assertThat(result.get(0).isReviewed()).isTrue();
+    }
+
+    @Test
+    void findEligibleKurrentBlocks_excludesUnreviewedOcrBlock() {
+        blockRepository.save(block(kurrentDocId, kurrentAnnotationId, BlockSource.OCR, false));
+
+        List<TranscriptionBlock> result = blockRepository.findEligibleKurrentBlocks();
+
+        assertThat(result).isEmpty();
+    }
+
+    @Test
+    void findEligibleKurrentBlocks_excludesNonEnrolledDocument() {
+        blockRepository.save(block(typewriterDocId, typewriterAnnotationId, BlockSource.MANUAL, false));
+
+        List<TranscriptionBlock> result = blockRepository.findEligibleKurrentBlocks();
+
+        assertThat(result).isEmpty();
+    }
+
+    @Test
+    void findEligibleKurrentBlocks_returnsAllReviewedBlocksAcrossBothSources() {
+        blockRepository.save(block(kurrentDocId, kurrentAnnotationId, BlockSource.MANUAL, true));
+        blockRepository.save(block(kurrentDocId, kurrentAnnotationId, BlockSource.OCR, true));
+        blockRepository.save(block(kurrentDocId, kurrentAnnotationId, BlockSource.MANUAL, false)); // excluded
+        blockRepository.save(block(kurrentDocId, kurrentAnnotationId, BlockSource.OCR, false));   // excluded
+
+        List<TranscriptionBlock> result = blockRepository.findEligibleKurrentBlocks();
+
+        assertThat(result).hasSize(2);
+    }
+
+    // ─── helpers ─────────────────────────────────────────────────────────────
+
+    private DocumentAnnotation annotation(UUID docId) {
+        return DocumentAnnotation.builder()
+                .documentId(docId)
+                .pageNumber(1)
+                .x(0.1).y(0.2).width(0.3).height(0.4)
+                .color("#00C7B1")
+                .build();
+    }
+
+    private TranscriptionBlock block(UUID docId, UUID annotId, BlockSource source, boolean reviewed) {
+        return TranscriptionBlock.builder()
+                .annotationId(annotId)
+                .documentId(docId)
+                .text("Liebe Tante")
+                .sortOrder(0)
+                .source(source)
+                .reviewed(reviewed)
+                .build();
+    }
+}

backend/src/test/java/org/raddatz/familienarchiv/service/AnnotationServiceTest.java

@@ -9,6 +9,7 @@ import org.raddatz.familienarchiv.dto.CreateAnnotationDTO;
 import org.raddatz.familienarchiv.exception.DomainException;
 import org.raddatz.familienarchiv.model.DocumentAnnotation;
 import org.raddatz.familienarchiv.repository.AnnotationRepository;
+import org.raddatz.familienarchiv.repository.TranscriptionBlockRepository;
 
 import java.util.List;
 import java.util.Optional;
@@ -20,7 +21,6 @@ import static org.mockito.ArgumentMatchers.any;
 import static org.mockito.Mockito.never;
 import static org.mockito.Mockito.verify;
 import static org.mockito.Mockito.when;
-import static org.springframework.http.HttpStatus.CONFLICT;
 import static org.springframework.http.HttpStatus.FORBIDDEN;
 import static org.springframework.http.HttpStatus.NOT_FOUND;
 
@@ -28,39 +28,19 @@ import static org.springframework.http.HttpStatus.NOT_FOUND;
 class AnnotationServiceTest {
 
     @Mock AnnotationRepository annotationRepository;
+    @Mock TranscriptionBlockRepository blockRepository;
     @InjectMocks AnnotationService annotationService;
 
     // ─── createAnnotation ─────────────────────────────────────────────────────
 
     @Test
-    void createAnnotation_throwsConflict_whenAnnotationOverlapsExisting() {
+    void createAnnotation_savesAnnotation() {
         UUID docId = UUID.randomUUID();
         UUID userId = UUID.randomUUID();
         CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.1, 0.1, 0.3, 0.3, "#ff0000");
-
-        DocumentAnnotation existing = DocumentAnnotation.builder()
-                .id(UUID.randomUUID()).documentId(docId).pageNumber(1)
-                .x(0.2).y(0.2).width(0.3).height(0.3).color("#00ff00").build();
-        when(annotationRepository.findByDocumentIdAndPageNumber(docId, 1))
-                .thenReturn(List.of(existing));
-
-        assertThatThrownBy(() -> annotationService.createAnnotation(docId, dto, userId, null))
-                .isInstanceOf(DomainException.class)
-                .satisfies(e -> assertThat(((DomainException) e).getStatus()).isEqualTo(CONFLICT));
-
-        verify(annotationRepository, never()).save(any());
-    }
-
-    @Test
-    void createAnnotation_savesAndReturns_whenNoOverlap() {
-        UUID docId = UUID.randomUUID();
-        UUID userId = UUID.randomUUID();
-        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.0, 0.0, 0.05, 0.05, "#ff0000");
-
-        when(annotationRepository.findByDocumentIdAndPageNumber(docId, 1)).thenReturn(List.of());
         DocumentAnnotation saved = DocumentAnnotation.builder()
                 .id(UUID.randomUUID()).documentId(docId).pageNumber(1)
-                .x(0.0).y(0.0).width(0.05).height(0.05).color("#ff0000").createdBy(userId).build();
+                .x(0.1).y(0.1).width(0.3).height(0.3).color("#ff0000").createdBy(userId).build();
         when(annotationRepository.save(any())).thenReturn(saved);
 
         DocumentAnnotation result = annotationService.createAnnotation(docId, dto, userId, null);
@@ -69,6 +49,77 @@ class AnnotationServiceTest {
         verify(annotationRepository).save(any());
     }
 
+    @Test
+    void createAnnotation_allowsOverlappingAnnotations() {
+        UUID docId = UUID.randomUUID();
+        UUID userId = UUID.randomUUID();
+        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.1, 0.1, 0.3, 0.3, "#ff0000");
+        when(annotationRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+
+        // Should not throw even when overlapping annotations exist on the same page
+        annotationService.createAnnotation(docId, dto, userId, null);
+
+        verify(annotationRepository).save(any());
+    }
+
+    @Test
+    void createAnnotation_setsFileHash_whenProvided() {
+        UUID docId = UUID.randomUUID();
+        UUID userId = UUID.randomUUID();
+        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.0, 0.0, 0.05, 0.05, "#ff0000");
+        when(annotationRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+
+        DocumentAnnotation result = annotationService.createAnnotation(docId, dto, userId, "abc123");
+
+        assertThat(result.getFileHash()).isEqualTo("abc123");
+    }
+
+    @Test
+    void createAnnotation_setsNullFileHash_whenNoneProvided() {
+        UUID docId = UUID.randomUUID();
+        UUID userId = UUID.randomUUID();
+        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.0, 0.0, 0.05, 0.05, "#ff0000");
+        when(annotationRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+
+        DocumentAnnotation result = annotationService.createAnnotation(docId, dto, userId, null);
+
+        assertThat(result.getFileHash()).isNull();
+    }
+
+    // ─── createOcrAnnotation ──────────────────────────────────────────────────
+
+    @Test
+    void createOcrAnnotation_savesWithPolygon() {
+        UUID docId = UUID.randomUUID();
+        UUID userId = UUID.randomUUID();
+        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.1, 0.1, 0.8, 0.04, "#00C7B1");
+        List<List<Double>> polygon = List.of(
+                List.of(0.1, 0.1), List.of(0.9, 0.11),
+                List.of(0.89, 0.14), List.of(0.11, 0.13));
+        when(annotationRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+
+        DocumentAnnotation result = annotationService.createOcrAnnotation(
+                docId, dto, userId, "filehash", polygon);
+
+        assertThat(result.getPolygon()).isEqualTo(polygon);
+        assertThat(result.getDocumentId()).isEqualTo(docId);
+        verify(annotationRepository).save(any());
+    }
+
+    @Test
+    void createOcrAnnotation_savesWithNullPolygon_whenPolygonNotProvided() {
+        UUID docId = UUID.randomUUID();
+        UUID userId = UUID.randomUUID();
+        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.1, 0.1, 0.8, 0.04, "#00C7B1");
+        when(annotationRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+
+        DocumentAnnotation result = annotationService.createOcrAnnotation(
+                docId, dto, userId, "filehash", null);
+
+        assertThat(result.getPolygon()).isNull();
+        verify(annotationRepository).save(any());
+    }
+
     // ─── deleteAnnotation ─────────────────────────────────────────────────────
 
     @Test
@@ -114,36 +165,42 @@ class AnnotationServiceTest {
 
         annotationService.deleteAnnotation(docId, annotId, ownerId);
 
+        verify(blockRepository).deleteByAnnotationId(annotId);
         verify(annotationRepository).delete(annotation);
     }
 
     @Test
-    void createAnnotation_setsFileHash_whenProvided() {
+    void deleteAnnotation_deletesLinkedTranscriptionBlock_beforeAnnotation() {
         UUID docId = UUID.randomUUID();
-        UUID userId = UUID.randomUUID();
+        UUID annotId = UUID.randomUUID();
-        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.0, 0.0, 0.05, 0.05, "#ff0000");
+        UUID ownerId = UUID.randomUUID();
-        String fileHash = "abc123";
 
-        when(annotationRepository.findByDocumentIdAndPageNumber(docId, 1)).thenReturn(List.of());
+        DocumentAnnotation annotation = DocumentAnnotation.builder()
-        when(annotationRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+                .id(annotId).documentId(docId).createdBy(ownerId).build();
+        when(annotationRepository.findByIdAndDocumentId(annotId, docId))
+                .thenReturn(Optional.of(annotation));
 
-        DocumentAnnotation result = annotationService.createAnnotation(docId, dto, userId, fileHash);
+        annotationService.deleteAnnotation(docId, annotId, ownerId);
 
-        assertThat(result.getFileHash()).isEqualTo(fileHash);
+        var inOrder = org.mockito.Mockito.inOrder(blockRepository, annotationRepository);
+        inOrder.verify(blockRepository).deleteByAnnotationId(annotId);
+        inOrder.verify(annotationRepository).delete(annotation);
     }
 
     @Test
-    void createAnnotation_setsNullFileHash_whenNoneProvided() {
+    void deleteAnnotation_throwsForbidden_whenUserIdIsNull() {
         UUID docId = UUID.randomUUID();
-        UUID userId = UUID.randomUUID();
+        UUID annotId = UUID.randomUUID();
-        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.0, 0.0, 0.05, 0.05, "#ff0000");
+        UUID ownerId = UUID.randomUUID();
 
-        when(annotationRepository.findByDocumentIdAndPageNumber(docId, 1)).thenReturn(List.of());
+        DocumentAnnotation annotation = DocumentAnnotation.builder()
-        when(annotationRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+                .id(annotId).documentId(docId).createdBy(ownerId).build();
+        when(annotationRepository.findByIdAndDocumentId(annotId, docId))
+                .thenReturn(Optional.of(annotation));
 
-        DocumentAnnotation result = annotationService.createAnnotation(docId, dto, userId, null);
+        assertThatThrownBy(() -> annotationService.deleteAnnotation(docId, annotId, null))
-
+                .isInstanceOf(DomainException.class)
-        assertThat(result.getFileHash()).isNull();
+                .satisfies(e -> assertThat(((DomainException) e).getStatus()).isEqualTo(FORBIDDEN));
     }
 
     // ─── listAnnotations ──────────────────────────────────────────────────────
@@ -183,149 +240,4 @@ class AnnotationServiceTest {
 
         verify(annotationRepository, never()).save(any());
     }
-
-    // ─── deleteAnnotation — null userId ───────────────────────────────────────
-
-    @Test
-    void deleteAnnotation_throwsForbidden_whenUserIdIsNull() {
-        UUID docId = UUID.randomUUID();
-        UUID annotId = UUID.randomUUID();
-        UUID ownerId = UUID.randomUUID();
-
-        DocumentAnnotation annotation = DocumentAnnotation.builder()
-                .id(annotId).documentId(docId).createdBy(ownerId).build();
-        when(annotationRepository.findByIdAndDocumentId(annotId, docId))
-                .thenReturn(Optional.of(annotation));
-
-        assertThatThrownBy(() -> annotationService.deleteAnnotation(docId, annotId, null))
-                .isInstanceOf(DomainException.class)
-                .satisfies(e -> assertThat(((DomainException) e).getStatus()).isEqualTo(FORBIDDEN));
-    }
-
-    // ─── overlaps — partial overlap cases ────────────────────────────────────
-
-    @Test
-    void createAnnotation_noConflict_whenAnnotationIsToTheLeft() {
-        // existing: x=0.5, w=0.3 (x2=0.8); dto: x=0.0, w=0.4 (dx2=0.4)
-        // existing.getX() < dx2 → 0.5 < 0.4 → FALSE → no overlap (first && fails)
-        UUID docId = UUID.randomUUID();
-        DocumentAnnotation existing = DocumentAnnotation.builder()
-                .id(UUID.randomUUID()).documentId(docId).pageNumber(1)
-                .x(0.5).y(0.0).width(0.3).height(0.5).color("#ff0000").build();
-        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.0, 0.0, 0.4, 0.5, "#0000ff");
-        when(annotationRepository.findByDocumentIdAndPageNumber(docId, 1)).thenReturn(List.of(existing));
-        when(annotationRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
-
-        annotationService.createAnnotation(docId, dto, UUID.randomUUID(), null);
-
-        verify(annotationRepository).save(any());
-    }
-
-    @Test
-    void createAnnotation_noConflict_whenAnnotationIsToTheRight() {
-        // existing: x=0.0, w=0.1 (ex2=0.1); dto: x=0.2, w=0.3 (dx2=0.5)
-        // existing.getX() < dx2 → 0.0 < 0.5 → TRUE
-        // ex2 > dto.getX() → 0.1 > 0.2 → FALSE → no overlap (second && fails)
-        UUID docId = UUID.randomUUID();
-        DocumentAnnotation existing = DocumentAnnotation.builder()
-                .id(UUID.randomUUID()).documentId(docId).pageNumber(1)
-                .x(0.0).y(0.0).width(0.1).height(0.5).color("#ff0000").build();
-        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.2, 0.0, 0.3, 0.5, "#0000ff");
-        when(annotationRepository.findByDocumentIdAndPageNumber(docId, 1)).thenReturn(List.of(existing));
-        when(annotationRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
-
-        annotationService.createAnnotation(docId, dto, UUID.randomUUID(), null);
-
-        verify(annotationRepository).save(any());
-    }
-
-    @Test
-    void createAnnotation_noConflict_whenAnnotationIsBelow() {
-        // x ranges overlap, but y ranges don't
-        // existing: x=0.0, w=0.5, y=0.5, h=0.2 (ey2=0.7)
-        // dto: x=0.1, w=0.3 (dx2=0.4), y=0.0, h=0.4 (dy2=0.4)
-        // existing.getX() < dx2 → 0.0 < 0.4 → TRUE
-        // ex2 > dto.getX() → 0.5 > 0.1 → TRUE
-        // existing.getY() < dy2 → 0.5 < 0.4 → FALSE → no overlap (third && fails)
-        UUID docId = UUID.randomUUID();
-        DocumentAnnotation existing = DocumentAnnotation.builder()
-                .id(UUID.randomUUID()).documentId(docId).pageNumber(1)
-                .x(0.0).y(0.5).width(0.5).height(0.2).color("#ff0000").build();
-        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.1, 0.0, 0.3, 0.4, "#0000ff");
-        when(annotationRepository.findByDocumentIdAndPageNumber(docId, 1)).thenReturn(List.of(existing));
-        when(annotationRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
-
-        annotationService.createAnnotation(docId, dto, UUID.randomUUID(), null);
-
-        verify(annotationRepository).save(any());
-    }
-
-    // ─── createOcrAnnotation ──────────────────────────────────────────────────
-
-    @Test
-    void createOcrAnnotation_skipsOverlapCheck_andSavesWithPolygon() {
-        UUID docId = UUID.randomUUID();
-        UUID userId = UUID.randomUUID();
-        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.1, 0.1, 0.8, 0.04, "#00C7B1");
-        List<List<Double>> polygon = List.of(
-                List.of(0.1, 0.1), List.of(0.9, 0.11),
-                List.of(0.89, 0.14), List.of(0.11, 0.13));
-        when(annotationRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
-
-        DocumentAnnotation result = annotationService.createOcrAnnotation(
-                docId, dto, userId, "filehash", polygon);
-
-        assertThat(result.getPolygon()).isEqualTo(polygon);
-        assertThat(result.getDocumentId()).isEqualTo(docId);
-        verify(annotationRepository).save(any());
-        verify(annotationRepository, never()).findByDocumentIdAndPageNumber(any(), any(int.class));
-    }
-
-    @Test
-    void createOcrAnnotation_savesWithNullPolygon_whenPolygonNotProvided() {
-        UUID docId = UUID.randomUUID();
-        UUID userId = UUID.randomUUID();
-        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.1, 0.1, 0.8, 0.04, "#00C7B1");
-        when(annotationRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
-
-        DocumentAnnotation result = annotationService.createOcrAnnotation(
-                docId, dto, userId, "filehash", null);
-
-        assertThat(result.getPolygon()).isNull();
-        verify(annotationRepository).save(any());
-    }
-
-    @Test
-    void createOcrAnnotation_doesNotCheckOverlap_evenWhenOverlappingAnnotationExists() {
-        UUID docId = UUID.randomUUID();
-        UUID userId = UUID.randomUUID();
-        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.1, 0.1, 0.3, 0.3, "#00C7B1");
-        when(annotationRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
-
-        annotationService.createOcrAnnotation(docId, dto, userId, "hash", null);
-
-        verify(annotationRepository, never()).findByDocumentIdAndPageNumber(any(), any(int.class));
-    }
-
-    // ─── overlaps — partial overlap cases ────────────────────────────────────
-
-    @Test
-    void createAnnotation_noConflict_whenAnnotationIsAbove() {
-        // x ranges overlap, y ranges don't — existing is ABOVE the new annotation
-        // existing: x=0.0, w=0.5, y=0.0, h=0.1 (ey2=0.1)
-        // dto: x=0.1, w=0.3 (dx2=0.4), y=0.2, h=0.3 (dy2=0.5)
-        // A: 0.0 < 0.4 → TRUE, B: 0.5 > 0.1 → TRUE, C: 0.0 < 0.5 → TRUE
-        // D: ey2 > dto.getY() → 0.1 > 0.2 → FALSE → no overlap (fourth && fails)
-        UUID docId = UUID.randomUUID();
-        DocumentAnnotation existing = DocumentAnnotation.builder()
-                .id(UUID.randomUUID()).documentId(docId).pageNumber(1)
-                .x(0.0).y(0.0).width(0.5).height(0.1).color("#ff0000").build();
-        CreateAnnotationDTO dto = new CreateAnnotationDTO(1, 0.1, 0.2, 0.3, 0.3, "#0000ff");
-        when(annotationRepository.findByDocumentIdAndPageNumber(docId, 1)).thenReturn(List.of(existing));
-        when(annotationRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
-
-        annotationService.createAnnotation(docId, dto, UUID.randomUUID(), null);
-
-        verify(annotationRepository).save(any());
-    }
 }

backend/src/test/java/org/raddatz/familienarchiv/service/OcrAsyncRunnerTest.java

@@ -60,8 +60,8 @@ class OcrAsyncRunnerTest {
 
         when(fileService.generatePresignedUrl(any())).thenReturn("http://presigned");
         when(ocrClient.extractBlocks(any(), any())).thenReturn(List.of(
-                new OcrBlockResult(0, 0.1, 0.1, 0.8, 0.04, null, "Line 1"),
+                new OcrBlockResult(0, 0.1, 0.1, 0.8, 0.04, null, "Line 1", null),
-                new OcrBlockResult(0, 0.1, 0.2, 0.8, 0.04, null, "Line 2")));
+                new OcrBlockResult(0, 0.1, 0.2, 0.8, 0.04, null, "Line 2", null)));
         DocumentAnnotation ann = DocumentAnnotation.builder().id(annId).build();
         when(annotationService.createOcrAnnotation(any(), any(), any(), any(), any())).thenReturn(ann);
 
@@ -84,7 +84,7 @@ class OcrAsyncRunnerTest {
 
         when(fileService.generatePresignedUrl(any())).thenReturn("http://presigned");
         when(ocrClient.extractBlocks(any(), any())).thenReturn(List.of(
-                new OcrBlockResult(0, 0.1, 0.1, 0.8, 0.04, null, "Test")));
+                new OcrBlockResult(0, 0.1, 0.1, 0.8, 0.04, null, "Test", null)));
         DocumentAnnotation ann = DocumentAnnotation.builder().id(annId).build();
         when(annotationService.createOcrAnnotation(any(), any(), any(), any(), any())).thenReturn(ann);
 
@@ -112,12 +112,12 @@ class OcrAsyncRunnerTest {
 
         when(fileService.generatePresignedUrl(any())).thenReturn("http://presigned");
         doAnswer(inv -> {
-            Consumer<OcrStreamEvent> handler = inv.getArgument(2);
+            Consumer<OcrStreamEvent> handler = inv.getArgument(3);
             handler.accept(new OcrStreamEvent.Start(1));
             handler.accept(new OcrStreamEvent.Page(0, List.of()));
             handler.accept(new OcrStreamEvent.Done(0, 0));
             return null;
-        }).when(ocrClient).streamBlocks(any(), any(), any());
+        }).when(ocrClient).streamBlocks(any(), any(), any(), any());
 
         ocrAsyncRunner.runSingleDocument(jobId, docId, userId);
 
@@ -142,7 +142,7 @@ class OcrAsyncRunnerTest {
         when(documentService.getDocumentById(docId)).thenReturn(doc);
 
         when(fileService.generatePresignedUrl(any())).thenReturn("http://presigned");
-        doThrow(new RuntimeException("OCR failed")).when(ocrClient).streamBlocks(any(), any(), any());
+        doThrow(new RuntimeException("OCR failed")).when(ocrClient).streamBlocks(any(), any(), any(), any());
 
         ocrAsyncRunner.runSingleDocument(jobId, docId, userId);
 
@@ -174,18 +174,18 @@ class OcrAsyncRunnerTest {
 
         List<String> progressMessages = new ArrayList<>();
         doAnswer(inv -> {
-            Consumer<OcrStreamEvent> handler = inv.getArgument(2);
+            Consumer<OcrStreamEvent> handler = inv.getArgument(3);
             handler.accept(new OcrStreamEvent.Start(3));
             handler.accept(new OcrStreamEvent.Page(0, List.of(
-                    new OcrBlockResult(0, 0.1, 0.1, 0.8, 0.04, null, "L1"),
+                    new OcrBlockResult(0, 0.1, 0.1, 0.8, 0.04, null, "L1", null),
-                    new OcrBlockResult(0, 0.1, 0.2, 0.8, 0.04, null, "L2"))));
+                    new OcrBlockResult(0, 0.1, 0.2, 0.8, 0.04, null, "L2", null))));
             progressMessages.add(job.getProgressMessage());
             handler.accept(new OcrStreamEvent.Page(1, List.of(
-                    new OcrBlockResult(1, 0.1, 0.1, 0.8, 0.04, null, "L3"))));
+                    new OcrBlockResult(1, 0.1, 0.1, 0.8, 0.04, null, "L3", null))));
             progressMessages.add(job.getProgressMessage());
             handler.accept(new OcrStreamEvent.Done(3, 0));
             return null;
-        }).when(ocrClient).streamBlocks(any(), any(), any());
+        }).when(ocrClient).streamBlocks(any(), any(), any(), any());
 
         ocrAsyncRunner.runSingleDocument(jobId, docId, userId);
 
@@ -215,14 +215,14 @@ class OcrAsyncRunnerTest {
         when(fileService.generatePresignedUrl(any())).thenReturn("http://presigned");
 
         doAnswer(inv -> {
-            Consumer<OcrStreamEvent> handler = inv.getArgument(2);
+            Consumer<OcrStreamEvent> handler = inv.getArgument(3);
             handler.accept(new OcrStreamEvent.Start(3));
             handler.accept(new OcrStreamEvent.Page(0, List.of()));
             handler.accept(new OcrStreamEvent.Error(1, "failed"));
             handler.accept(new OcrStreamEvent.Page(2, List.of()));
             handler.accept(new OcrStreamEvent.Done(0, 1));
             return null;
-        }).when(ocrClient).streamBlocks(any(), any(), any());
+        }).when(ocrClient).streamBlocks(any(), any(), any(), any());
 
         ocrAsyncRunner.runSingleDocument(jobId, docId, userId);
 
@@ -251,13 +251,13 @@ class OcrAsyncRunnerTest {
         when(fileService.generatePresignedUrl(any())).thenReturn("http://presigned");
 
         doAnswer(inv -> {
-            Consumer<OcrStreamEvent> handler = inv.getArgument(2);
+            Consumer<OcrStreamEvent> handler = inv.getArgument(3);
             handler.accept(new OcrStreamEvent.Start(2));
             handler.accept(new OcrStreamEvent.Error(0, "some python traceback details"));
             handler.accept(new OcrStreamEvent.Page(1, List.of()));
             handler.accept(new OcrStreamEvent.Done(0, 1));
             return null;
-        }).when(ocrClient).streamBlocks(any(), any(), any());
+        }).when(ocrClient).streamBlocks(any(), any(), any(), any());
 
         ocrAsyncRunner.runSingleDocument(jobId, docId, userId);
 

backend/src/test/java/org/raddatz/familienarchiv/service/OcrClientDefaultStreamTest.java

@@ -12,13 +12,19 @@ class OcrClientDefaultStreamTest {
 
     @Test
     void defaultStreamBlocksSynthesizesEventsFromExtractBlocks() {
-        OcrClient client = (pdfUrl, scriptType) -> List.of(
+        OcrClient client = new OcrClient() {
-                new OcrBlockResult(0, 0.1, 0.1, 0.8, 0.04, null, "Line 1"),
+            @Override public List<OcrBlockResult> extractBlocks(String pdfUrl, ScriptType scriptType) {
-                new OcrBlockResult(0, 0.1, 0.2, 0.8, 0.04, null, "Line 2"),
+                return List.of(
-                new OcrBlockResult(1, 0.1, 0.1, 0.8, 0.04, null, "Line 3"));
+                    new OcrBlockResult(0, 0.1, 0.1, 0.8, 0.04, null, "Line 1", null),
+                    new OcrBlockResult(0, 0.1, 0.2, 0.8, 0.04, null, "Line 2", null),
+                    new OcrBlockResult(1, 0.1, 0.1, 0.8, 0.04, null, "Line 3", null));
+            }
+            @Override public TrainingResult trainModel(byte[] zip) { return null; }
+            @Override public TrainingResult segtrainModel(byte[] zip) { return null; }
+        };
 
         List<OcrStreamEvent> events = new ArrayList<>();
-        client.streamBlocks("http://test", ScriptType.TYPEWRITER, events::add);
+        client.streamBlocks("http://test", ScriptType.TYPEWRITER, null, events::add);
 
         assertThat(events).hasSize(4);
         assertThat(events.get(0)).isInstanceOf(OcrStreamEvent.Start.class);
@@ -42,10 +48,14 @@ class OcrClientDefaultStreamTest {
 
     @Test
     void defaultStreamBlocksHandlesEmptyResults() {
-        OcrClient client = (pdfUrl, scriptType) -> List.of();
+        OcrClient client = new OcrClient() {
+            @Override public List<OcrBlockResult> extractBlocks(String pdfUrl, ScriptType scriptType) { return List.of(); }
+            @Override public TrainingResult trainModel(byte[] zip) { return null; }
+            @Override public TrainingResult segtrainModel(byte[] zip) { return null; }
+        };
 
         List<OcrStreamEvent> events = new ArrayList<>();
-        client.streamBlocks("http://test", ScriptType.TYPEWRITER, events::add);
+        client.streamBlocks("http://test", ScriptType.TYPEWRITER, null, events::add);
 
         assertThat(events).hasSize(2);
         assertThat(events.get(0)).isInstanceOf(OcrStreamEvent.Start.class);

backend/src/test/java/org/raddatz/familienarchiv/service/OcrServiceTest.java

@@ -142,7 +142,7 @@ class OcrServiceTest {
         UUID result = ocrService.startOcr(docId, null, userId);
 
         assertThat(result).isEqualTo(jobId);
-        verify(ocrAsyncRunner).runSingleDocument(jobId, docId, userId);
+        verify(ocrAsyncRunner).runSingleDocument(jobId, docId, userId, false);
     }
 
     @Test

backend/src/test/java/org/raddatz/familienarchiv/service/OcrStreamEventTest.java

@@ -17,7 +17,7 @@ class OcrStreamEventTest {
 
     @Test
     void pageRecordHoldsBlocksAndPageNumber() {
-        var block = new OcrBlockResult(0, 0.1, 0.2, 0.8, 0.1, null, "Test");
+        var block = new OcrBlockResult(0, 0.1, 0.2, 0.8, 0.1, null, "Test", null);
         var page = new OcrStreamEvent.Page(0, List.of(block));
         assertThat(page.pageNumber()).isEqualTo(0);
         assertThat(page.blocks()).hasSize(1);

backend/src/test/java/org/raddatz/familienarchiv/service/OcrTrainingServiceTest.java

@@ -0,0 +1,181 @@
+package org.raddatz.familienarchiv.service;
+
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Test;
+import org.raddatz.familienarchiv.exception.DomainException;
+import org.raddatz.familienarchiv.model.OcrTrainingRun;
+import org.raddatz.familienarchiv.model.TrainingLabel;
+import org.raddatz.familienarchiv.model.TrainingStatus;
+import org.raddatz.familienarchiv.model.TranscriptionBlock;
+import org.raddatz.familienarchiv.repository.OcrTrainingRunRepository;
+import org.raddatz.familienarchiv.repository.TranscriptionBlockRepository;
+import org.springframework.transaction.support.TransactionCallback;
+import org.springframework.transaction.support.TransactionTemplate;
+
+import java.time.Instant;
+import java.util.List;
+import java.util.Optional;
+import java.util.UUID;
+
+import static org.assertj.core.api.Assertions.assertThat;
+import static org.assertj.core.api.Assertions.assertThatThrownBy;
+import static org.mockito.ArgumentMatchers.any;
+import static org.mockito.ArgumentMatchers.argThat;
+import static org.mockito.Mockito.*;
+
+class OcrTrainingServiceTest {
+
+    OcrTrainingRunRepository runRepository;
+    TrainingDataExportService exportService;
+    SegmentationTrainingExportService segExportService;
+    OcrClient ocrClient;
+    OcrHealthClient healthClient;
+    TranscriptionBlockRepository blockRepository;
+    TransactionTemplate txTemplate;
+    OcrTrainingService service;
+
+    @BeforeEach
+    void setUp() {
+        runRepository = mock(OcrTrainingRunRepository.class);
+        exportService = mock(TrainingDataExportService.class);
+        segExportService = mock(SegmentationTrainingExportService.class);
+        ocrClient = mock(OcrClient.class);
+        healthClient = mock(OcrHealthClient.class);
+        blockRepository = mock(TranscriptionBlockRepository.class);
+        txTemplate = mock(TransactionTemplate.class);
+
+        // Execute transaction callbacks inline so unit tests run without a real DataSource
+        when(txTemplate.execute(any())).thenAnswer(inv -> {
+            TransactionCallback<?> callback = inv.getArgument(0);
+            return callback.doInTransaction(null);
+        });
+
+        service = new OcrTrainingService(runRepository, exportService, segExportService, ocrClient, healthClient, blockRepository, txTemplate);
+
+        when(blockRepository.count()).thenReturn(0L);
+        when(runRepository.findTop5ByOrderByCreatedAtDesc()).thenReturn(List.of());
+        when(segExportService.querySegmentationBlocks()).thenReturn(List.of());
+    }
+
+    // ─── Concurrent guard ─────────────────────────────────────────────────────
+
+    @Test
+    void triggerTraining_throws409_whenRunningRunExists() {
+        when(runRepository.findFirstByStatus(TrainingStatus.RUNNING))
+                .thenReturn(Optional.of(OcrTrainingRun.builder()
+                        .id(UUID.randomUUID()).status(TrainingStatus.RUNNING)
+                        .blockCount(5).documentCount(2).modelName("german_kurrent").build()));
+
+        assertThatThrownBy(() -> service.triggerTraining(null))
+                .isInstanceOf(DomainException.class)
+                .extracting("status")
+                .satisfies(s -> assertThat(s.toString()).contains("409"));
+    }
+
+    // ─── Threshold guard ──────────────────────────────────────────────────────
+
+    @Test
+    void triggerTraining_throws422_whenFewerThan5Blocks() {
+        when(runRepository.findFirstByStatus(TrainingStatus.RUNNING)).thenReturn(Optional.empty());
+        when(exportService.queryEligibleBlocks()).thenReturn(List.of(
+                TranscriptionBlock.builder().id(UUID.randomUUID()).documentId(UUID.randomUUID()).build(),
+                TranscriptionBlock.builder().id(UUID.randomUUID()).documentId(UUID.randomUUID()).build()
+        ));
+
+        assertThatThrownBy(() -> service.triggerTraining(null))
+                .isInstanceOf(DomainException.class);
+    }
+
+    // ─── Happy path ───────────────────────────────────────────────────────────
+
+    @Test
+    void triggerTraining_createsRunWithCorrectCounts_andMarksDone() throws Exception {
+        when(runRepository.findFirstByStatus(TrainingStatus.RUNNING)).thenReturn(Optional.empty());
+
+        UUID docA = UUID.randomUUID();
+        UUID docB = UUID.randomUUID();
+        List<TranscriptionBlock> blocks = List.of(
+                TranscriptionBlock.builder().id(UUID.randomUUID()).documentId(docA).build(),
+                TranscriptionBlock.builder().id(UUID.randomUUID()).documentId(docA).build(),
+                TranscriptionBlock.builder().id(UUID.randomUUID()).documentId(docA).build(),
+                TranscriptionBlock.builder().id(UUID.randomUUID()).documentId(docA).build(),
+                TranscriptionBlock.builder().id(UUID.randomUUID()).documentId(docB).build()
+        );
+        when(exportService.queryEligibleBlocks()).thenReturn(blocks);
+        when(exportService.exportToZip()).thenReturn(out -> {});
+        when(ocrClient.trainModel(any())).thenReturn(new OcrClient.TrainingResult(0.05, 0.95, 0.05, 3));
+
+        OcrTrainingRun saved = OcrTrainingRun.builder()
+                .id(UUID.randomUUID()).status(TrainingStatus.RUNNING)
+                .blockCount(5).documentCount(2).modelName("german_kurrent").build();
+        when(runRepository.save(any())).thenReturn(saved);
+
+        service.triggerTraining(null);
+
+        // Verify run created with correct counts and then updated to DONE
+        verify(runRepository, times(2)).save(argThat(run ->
+                run.getBlockCount() == 5 || run.getStatus() == TrainingStatus.DONE));
+    }
+
+    // ─── Failure path ─────────────────────────────────────────────────────────
+
+    @Test
+    void triggerTraining_marksRunFailed_whenOcrClientThrows() throws Exception {
+        when(runRepository.findFirstByStatus(TrainingStatus.RUNNING)).thenReturn(Optional.empty());
+
+        UUID docA = UUID.randomUUID();
+        List<TranscriptionBlock> blocks = List.of(
+                TranscriptionBlock.builder().id(UUID.randomUUID()).documentId(docA).build(),
+                TranscriptionBlock.builder().id(UUID.randomUUID()).documentId(docA).build(),
+                TranscriptionBlock.builder().id(UUID.randomUUID()).documentId(docA).build(),
+                TranscriptionBlock.builder().id(UUID.randomUUID()).documentId(docA).build(),
+                TranscriptionBlock.builder().id(UUID.randomUUID()).documentId(docA).build()
+        );
+        when(exportService.queryEligibleBlocks()).thenReturn(blocks);
+        when(exportService.exportToZip()).thenReturn(out -> {});
+        when(ocrClient.trainModel(any())).thenThrow(new RuntimeException("OCR service timeout"));
+
+        OcrTrainingRun saved = OcrTrainingRun.builder()
+                .id(UUID.randomUUID()).status(TrainingStatus.RUNNING)
+                .blockCount(5).documentCount(1).modelName("german_kurrent").build();
+        when(runRepository.save(any())).thenReturn(saved);
+
+        service.triggerTraining(null);
+
+        verify(runRepository, atLeastOnce()).save(argThat(run ->
+                run.getStatus() == TrainingStatus.FAILED && run.getErrorMessage() != null));
+    }
+
+    // ─── Orphan recovery ──────────────────────────────────────────────────────
+
+    @Test
+    void recoverOrphanedRuns_marksRunFailed_whenOlderThanOneHour() {
+        OcrTrainingRun orphan = OcrTrainingRun.builder()
+                .id(UUID.randomUUID()).status(TrainingStatus.RUNNING)
+                .blockCount(5).documentCount(1).modelName("german_kurrent")
+                .createdAt(Instant.now().minusSeconds(7200))
+                .build();
+        when(runRepository.findFirstByStatus(TrainingStatus.RUNNING)).thenReturn(Optional.of(orphan));
+        when(runRepository.save(any())).thenReturn(orphan);
+
+        service.recoverOrphanedRuns();
+
+        verify(runRepository).save(argThat(run ->
+                run.getStatus() == TrainingStatus.FAILED
+                        && run.getErrorMessage().contains("Abgebrochen")));
+    }
+
+    @Test
+    void recoverOrphanedRuns_doesNothing_whenRunIsRecent() {
+        OcrTrainingRun recent = OcrTrainingRun.builder()
+                .id(UUID.randomUUID()).status(TrainingStatus.RUNNING)
+                .blockCount(5).documentCount(1).modelName("german_kurrent")
+                .createdAt(Instant.now().minusSeconds(60))
+                .build();
+        when(runRepository.findFirstByStatus(TrainingStatus.RUNNING)).thenReturn(Optional.of(recent));
+
+        service.recoverOrphanedRuns();
+
+        verify(runRepository, never()).save(any());
+    }
+}

backend/src/test/java/org/raddatz/familienarchiv/service/TrainingDataExportServiceTest.java

@@ -0,0 +1,290 @@
+package org.raddatz.familienarchiv.service;
+
+import org.apache.pdfbox.Loader;
+import org.apache.pdfbox.pdmodel.PDDocument;
+import org.apache.pdfbox.pdmodel.PDPage;
+import org.apache.pdfbox.pdmodel.common.PDRectangle;
+import org.junit.jupiter.api.BeforeAll;
+import org.junit.jupiter.api.Test;
+import org.raddatz.familienarchiv.PostgresContainerConfig;
+import org.raddatz.familienarchiv.config.FlywayConfig;
+import org.raddatz.familienarchiv.model.*;
+import org.raddatz.familienarchiv.repository.AnnotationRepository;
+import org.raddatz.familienarchiv.repository.DocumentRepository;
+import org.raddatz.familienarchiv.repository.TranscriptionBlockRepository;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.boot.jdbc.test.autoconfigure.AutoConfigureTestDatabase;
+import org.springframework.boot.data.jpa.test.autoconfigure.DataJpaTest;
+import org.springframework.context.annotation.Import;
+import org.springframework.web.servlet.mvc.method.annotation.StreamingResponseBody;
+
+import java.io.ByteArrayInputStream;
+import java.io.ByteArrayOutputStream;
+import java.util.HashSet;
+import java.util.Set;
+import java.util.UUID;
+import java.util.zip.ZipEntry;
+import java.util.zip.ZipInputStream;
+
+import static org.assertj.core.api.Assertions.assertThat;
+import static org.mockito.ArgumentMatchers.anyString;
+import static org.mockito.Mockito.*;
+
+@DataJpaTest
+@AutoConfigureTestDatabase(replace = AutoConfigureTestDatabase.Replace.NONE)
+@Import({PostgresContainerConfig.class, FlywayConfig.class})
+class TrainingDataExportServiceTest {
+
+    @Autowired TranscriptionBlockRepository blockRepository;
+    @Autowired DocumentRepository documentRepository;
+    @Autowired AnnotationRepository annotationRepository;
+
+    static byte[] minimalPdfBytes;
+
+    @BeforeAll
+    static void createMinimalPdf() throws Exception {
+        try (PDDocument doc = new PDDocument()) {
+            doc.addPage(new PDPage(PDRectangle.A4));
+            ByteArrayOutputStream out = new ByteArrayOutputStream();
+            doc.save(out);
+            minimalPdfBytes = out.toByteArray();
+        }
+    }
+
+    // ─── Query: enrollment filter ─────────────────────────────────────────────
+
+    @Test
+    void export_includesManualBlockFromEnrolledDocument() throws Exception {
+        UUID docId = enrolledDoc("enrolled.pdf");
+        UUID annotId = annotation(docId);
+        blockRepository.save(manualBlock(docId, annotId, "Liebe Mutter"));
+
+        FileService fileService = mockFileService();
+        TrainingDataExportService service = new TrainingDataExportService(blockRepository, annotationRepository, documentRepository, fileService);
+
+        StreamingResponseBody body = service.exportToZip();
+        byte[] zipBytes = stream(body);
+        assertThat(zipEntryNames(zipBytes)).isNotEmpty();
+    }
+
+    @Test
+    void export_excludesUnreviewedManualBlockFromEnrolledDocument() throws Exception {
+        // MANUAL blocks whose text hasn't been verified yet should not count as training data
+        UUID docId = enrolledDoc("unreviewed-manual.pdf");
+        UUID annotId = annotation(docId);
+        TranscriptionBlock block = TranscriptionBlock.builder()
+                .annotationId(annotId).documentId(docId)
+                .text("Liebe Tante").sortOrder(0)
+                .source(BlockSource.MANUAL).reviewed(false).build();
+        blockRepository.save(block);
+
+        FileService fileService = mockFileService();
+        TrainingDataExportService service = new TrainingDataExportService(blockRepository, annotationRepository, documentRepository, fileService);
+
+        StreamingResponseBody body = service.exportToZip();
+        assertThat(zipEntryNames(stream(body))).isEmpty();
+    }
+
+    @Test
+    void export_excludesManualBlockFromNonEnrolledDocument() throws Exception {
+        UUID docId = nonEnrolledDoc("notenrolled.pdf");
+        UUID annotId = annotation(docId);
+        blockRepository.save(manualBlock(docId, annotId, "Liebe Tante"));
+
+        FileService fileService = mockFileService();
+        TrainingDataExportService service = new TrainingDataExportService(blockRepository, annotationRepository, documentRepository, fileService);
+
+        StreamingResponseBody body = service.exportToZip();
+        byte[] zipBytes = stream(body);
+        assertThat(zipEntryNames(zipBytes)).isEmpty();
+    }
+
+    @Test
+    void export_includesReviewedOcrBlockFromEnrolledDocument() throws Exception {
+        UUID docId = enrolledDoc("ocr-reviewed.pdf");
+        UUID annotId = annotation(docId);
+        TranscriptionBlock block = TranscriptionBlock.builder()
+                .annotationId(annotId).documentId(docId)
+                .text("OCR text").sortOrder(0)
+                .source(BlockSource.OCR).reviewed(true).build();
+        blockRepository.save(block);
+
+        FileService fileService = mockFileService();
+        TrainingDataExportService service = new TrainingDataExportService(blockRepository, annotationRepository, documentRepository, fileService);
+
+        StreamingResponseBody body = service.exportToZip();
+        assertThat(zipEntryNames(stream(body))).isNotEmpty();
+    }
+
+    @Test
+    void export_excludesUnreviewedOcrBlockFromEnrolledDocument() throws Exception {
+        UUID docId = enrolledDoc("ocr-unreviewed.pdf");
+        UUID annotId = annotation(docId);
+        TranscriptionBlock block = TranscriptionBlock.builder()
+                .annotationId(annotId).documentId(docId)
+                .text("Raw OCR").sortOrder(0)
+                .source(BlockSource.OCR).reviewed(false).build();
+        blockRepository.save(block);
+
+        FileService fileService = mockFileService();
+        TrainingDataExportService service = new TrainingDataExportService(blockRepository, annotationRepository, documentRepository, fileService);
+
+        StreamingResponseBody body = service.exportToZip();
+        assertThat(zipEntryNames(stream(body))).isEmpty();
+    }
+
+    // ─── ZIP structure ────────────────────────────────────────────────────────
+
+    @Test
+    void export_producesExactly2EntriesPerBlock_pngAndTxt() throws Exception {
+        UUID docId = enrolledDoc("zip-struct.pdf");
+        UUID annotId = annotation(docId);
+        blockRepository.save(manualBlock(docId, annotId, "Erste Zeile"));
+        blockRepository.save(manualBlock(docId, annotId, "Zweite Zeile"));
+
+        FileService fileService = mockFileService();
+        TrainingDataExportService service = new TrainingDataExportService(blockRepository, annotationRepository, documentRepository, fileService);
+
+        byte[] zipBytes = stream(service.exportToZip());
+        var names = zipEntryNames(zipBytes);
+        assertThat(names).hasSize(4); // 2 blocks × 2 entries each
+        assertThat(names.stream().filter(n -> n.endsWith(".png")).count()).isEqualTo(2);
+        assertThat(names.stream().filter(n -> n.endsWith(".xml")).count()).isEqualTo(2);
+    }
+
+    @Test
+    void export_pageXmlContainsBlockText() throws Exception {
+        UUID docId = enrolledDoc("txt-content.pdf");
+        UUID annotId = annotation(docId);
+        String expectedText = "Sehr geehrte Frau";
+        blockRepository.save(manualBlock(docId, annotId, expectedText));
+
+        FileService fileService = mockFileService();
+        TrainingDataExportService service = new TrainingDataExportService(blockRepository, annotationRepository, documentRepository, fileService);
+
+        byte[] zipBytes = stream(service.exportToZip());
+        String xmlContent = readZipEntry(zipBytes, ".xml");
+        assertThat(xmlContent).contains("<Unicode>" + expectedText + "</Unicode>");
+    }
+
+    @Test
+    void export_pageXmlEscapesSpecialCharacters() throws Exception {
+        UUID docId = enrolledDoc("special-chars.pdf");
+        UUID annotId = annotation(docId);
+        blockRepository.save(manualBlock(docId, annotId, "A & B < C > D"));
+
+        FileService fileService = mockFileService();
+        TrainingDataExportService service = new TrainingDataExportService(blockRepository, annotationRepository, documentRepository, fileService);
+
+        byte[] zipBytes = stream(service.exportToZip());
+        String xmlContent = readZipEntry(zipBytes, ".xml");
+        assertThat(xmlContent).contains("<Unicode>A &amp; B &lt; C &gt; D</Unicode>");
+    }
+
+    // ─── S3 failure resilience ────────────────────────────────────────────────
+
+    @Test
+    void export_skipsDocumentWhenS3DownloadFails_andStillIncludesOtherDocuments() throws Exception {
+        UUID failDocId = enrolledDoc("fail.pdf");
+        UUID okDocId = enrolledDoc("ok.pdf");
+        UUID failAnnotId = annotation(failDocId);
+        UUID okAnnotId = annotation(okDocId);
+        blockRepository.save(manualBlock(failDocId, failAnnotId, "Will fail"));
+        blockRepository.save(manualBlock(okDocId, okAnnotId, "Will succeed"));
+
+        FileService fileService = mock(FileService.class);
+        when(fileService.downloadFileBytes("fail.pdf")).thenThrow(new FileService.StorageFileNotFoundException("missing"));
+        when(fileService.downloadFileBytes("ok.pdf")).thenReturn(minimalPdfBytes);
+
+        TrainingDataExportService service = new TrainingDataExportService(blockRepository, annotationRepository, documentRepository, fileService);
+
+        byte[] zipBytes = stream(service.exportToZip());
+        var names = zipEntryNames(zipBytes);
+        // ok.pdf block produces 2 entries; fail.pdf block is skipped
+        assertThat(names).hasSize(2);
+    }
+
+    // ─── Empty export ─────────────────────────────────────────────────────────
+
+    @Test
+    void queryEligibleBlocks_returnsEmpty_whenNoEnrolledDocuments() {
+        FileService fileService = mockFileService();
+        TrainingDataExportService service = new TrainingDataExportService(blockRepository, annotationRepository, documentRepository, fileService);
+
+        assertThat(service.queryEligibleBlocks()).isEmpty();
+    }
+
+    // ─── helpers ─────────────────────────────────────────────────────────────
+
+    private UUID enrolledDoc(String filename) {
+        Document doc = documentRepository.save(Document.builder()
+                .title(filename).originalFilename(filename).filePath(filename)
+                .status(DocumentStatus.UPLOADED)
+                .trainingLabels(new HashSet<>(Set.of(TrainingLabel.KURRENT_RECOGNITION)))
+                .build());
+        return doc.getId();
+    }
+
+    private UUID nonEnrolledDoc(String filename) {
+        Document doc = documentRepository.save(Document.builder()
+                .title(filename).originalFilename(filename).filePath(filename)
+                .status(DocumentStatus.UPLOADED)
+                .build());
+        return doc.getId();
+    }
+
+    private UUID annotation(UUID docId) {
+        return annotationRepository.save(DocumentAnnotation.builder()
+                .documentId(docId).pageNumber(1)
+                .x(0.1).y(0.1).width(0.8).height(0.1).color("#00C7B1")
+                .build()).getId();
+    }
+
+    private TranscriptionBlock manualBlock(UUID docId, UUID annotId, String text) {
+        return TranscriptionBlock.builder()
+                .annotationId(annotId).documentId(docId)
+                .text(text).sortOrder(0)
+                .source(BlockSource.MANUAL).reviewed(true).build();
+    }
+
+    private FileService mockFileService() {
+        FileService fs = mock(FileService.class);
+        try {
+            when(fs.downloadFileBytes(anyString())).thenReturn(minimalPdfBytes);
+        } catch (Exception e) {
+            throw new RuntimeException(e);
+        }
+        return fs;
+    }
+
+    private static byte[] stream(StreamingResponseBody body) throws Exception {
+        ByteArrayOutputStream out = new ByteArrayOutputStream();
+        body.writeTo(out);
+        return out.toByteArray();
+    }
+
+    private static java.util.List<String> zipEntryNames(byte[] zipBytes) throws Exception {
+        var names = new java.util.ArrayList<String>();
+        try (ZipInputStream zis = new ZipInputStream(new ByteArrayInputStream(zipBytes))) {
+            ZipEntry entry;
+            while ((entry = zis.getNextEntry()) != null) {
+                names.add(entry.getName());
+                zis.closeEntry();
+            }
+        }
+        return names;
+    }
+
+    private static String readZipEntry(byte[] zipBytes, String suffix) throws Exception {
+        try (ZipInputStream zis = new ZipInputStream(new ByteArrayInputStream(zipBytes))) {
+            ZipEntry entry;
+            while ((entry = zis.getNextEntry()) != null) {
+                if (entry.getName().endsWith(suffix)) {
+                    return new String(zis.readAllBytes(), java.nio.charset.StandardCharsets.UTF_8);
+                }
+                zis.closeEntry();
+            }
+        }
+        return null;
+    }
+}

backend/src/test/java/org/raddatz/familienarchiv/service/TranscriptionServiceGuidedTest.java

@@ -0,0 +1,111 @@
+package org.raddatz.familienarchiv.service;
+
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Test;
+import org.raddatz.familienarchiv.model.BlockSource;
+import org.raddatz.familienarchiv.model.TranscriptionBlock;
+import org.raddatz.familienarchiv.repository.AnnotationRepository;
+import org.raddatz.familienarchiv.repository.TranscriptionBlockRepository;
+import org.raddatz.familienarchiv.repository.TranscriptionBlockVersionRepository;
+
+import java.util.Optional;
+import java.util.UUID;
+
+import static org.assertj.core.api.Assertions.assertThat;
+import static org.mockito.ArgumentMatchers.any;
+import static org.mockito.Mockito.*;
+
+class TranscriptionServiceGuidedTest {
+
+    TranscriptionBlockRepository blockRepository;
+    TranscriptionBlockVersionRepository versionRepository;
+    AnnotationRepository annotationRepository;
+    AnnotationService annotationService;
+    DocumentService documentService;
+    TranscriptionService service;
+
+    UUID docId = UUID.randomUUID();
+    UUID annId = UUID.randomUUID();
+    UUID userId = UUID.randomUUID();
+
+    @BeforeEach
+    void setUp() {
+        blockRepository = mock(TranscriptionBlockRepository.class);
+        versionRepository = mock(TranscriptionBlockVersionRepository.class);
+        annotationRepository = mock(AnnotationRepository.class);
+        annotationService = mock(AnnotationService.class);
+        documentService = mock(DocumentService.class);
+
+        service = new TranscriptionService(blockRepository, versionRepository,
+                annotationRepository, annotationService, documentService);
+
+        when(blockRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+        when(versionRepository.save(any())).thenAnswer(inv -> inv.getArgument(0));
+    }
+
+    @Test
+    void upsertGuidedBlock_createsNewBlock_whenAnnotationHasNoBlock() {
+        when(blockRepository.findByAnnotationId(annId)).thenReturn(Optional.empty());
+
+        TranscriptionBlock result = service.upsertGuidedBlock(docId, annId, "Hallo", userId);
+
+        assertThat(result.getText()).isEqualTo("Hallo");
+        assertThat(result.getSource()).isEqualTo(BlockSource.OCR);
+        assertThat(result.getAnnotationId()).isEqualTo(annId);
+        verify(blockRepository).save(any());
+    }
+
+    @Test
+    void upsertGuidedBlock_updatesExistingOcrBlock_whenAnnotationHasOcrBlock() {
+        TranscriptionBlock existing = TranscriptionBlock.builder()
+                .id(UUID.randomUUID())
+                .annotationId(annId)
+                .documentId(docId)
+                .text("old text")
+                .source(BlockSource.OCR)
+                .sortOrder(0)
+                .build();
+        when(blockRepository.findByAnnotationId(annId)).thenReturn(Optional.of(existing));
+
+        TranscriptionBlock result = service.upsertGuidedBlock(docId, annId, "new text", userId);
+
+        assertThat(result.getText()).isEqualTo("new text");
+        verify(blockRepository).save(any());
+    }
+
+    @Test
+    void upsertGuidedBlock_doesNotOverwriteNonEmptyManualBlock() {
+        TranscriptionBlock manual = TranscriptionBlock.builder()
+                .id(UUID.randomUUID())
+                .annotationId(annId)
+                .documentId(docId)
+                .text("manually written")
+                .source(BlockSource.MANUAL)
+                .sortOrder(0)
+                .build();
+        when(blockRepository.findByAnnotationId(annId)).thenReturn(Optional.of(manual));
+
+        TranscriptionBlock result = service.upsertGuidedBlock(docId, annId, "ocr result", userId);
+
+        assertThat(result.getText()).isEqualTo("manually written");
+        verify(blockRepository, never()).save(any());
+    }
+
+    @Test
+    void upsertGuidedBlock_fillsEmptyManualBlock_withOcrText() {
+        TranscriptionBlock emptyManual = TranscriptionBlock.builder()
+                .id(UUID.randomUUID())
+                .annotationId(annId)
+                .documentId(docId)
+                .text("")
+                .source(BlockSource.MANUAL)
+                .sortOrder(0)
+                .build();
+        when(blockRepository.findByAnnotationId(annId)).thenReturn(Optional.of(emptyManual));
+
+        TranscriptionBlock result = service.upsertGuidedBlock(docId, annId, "ocr result", userId);
+
+        assertThat(result.getText()).isEqualTo("ocr result");
+        verify(blockRepository).save(any());
+    }
+}

docker-compose.yml

@@ -72,6 +72,9 @@ services:
       - archive-net
 
   # --- OCR: Python microservice (Surya + Kraken) ---
+  # Single-node only: OCR training reloads the model in-process after each run.
+  # Running multiple replicas would cause training conflicts and model-state divergence.
+  # See ADR-001 for the architectural rationale.
   ocr-service:
     build:
       context: ./ocr-service
@@ -87,6 +90,7 @@ services:
       - ocr_cache:/root/.cache
     environment:
       KRAKEN_MODEL_PATH: /app/models/german_kurrent.mlmodel
+      TRAINING_TOKEN: "${OCR_TRAINING_TOKEN:-}"
       OCR_CONFIDENCE_THRESHOLD: "0.3"
       OCR_CONFIDENCE_THRESHOLD_KURRENT: "0.5"
       RECOGNITION_BATCH_SIZE: "16"

frontend/messages/de.json

@@ -1,7 +1,6 @@
 {
 	"$schema": "https://inlang.com/schema/inlang-message-format",
 	"error_annotation_not_found": "Die Annotation wurde nicht gefunden.",
-	"error_annotation_overlap": "Die Annotation überschneidet sich mit einer vorhandenen.",
 	"annotation_outdated_notice": "Einige Annotationen beziehen sich auf eine frühere Dateiversion und werden nicht angezeigt.",
 	"error_document_not_found": "Das Dokument wurde nicht gefunden.",
 	"error_document_no_file": "Diesem Dokument ist noch keine Datei zugeordnet.",
@@ -480,6 +479,7 @@
 	"scan_collapse": "Scan verkleinern",
 	"transcription_empty_title": "Noch keine Transkription",
 	"transcription_empty_desc": "Zeichne Bereiche auf dem Scan und tippe den Text ab, um eine Transkription zu erstellen.",
+	"transcription_empty_draw_hint": "Zeichnen Sie Bereiche auf dem Dokument, um mit der Transkription zu beginnen.",
 	"transcription_panel_close": "Panel schließen",
 	"person_alias_heading": "Namensverlauf",
 	"person_alias_empty": "Noch keine Namensaenderungen erfasst.",
@@ -505,6 +505,7 @@
 	"error_ocr_job_not_found": "Der OCR-Auftrag wurde nicht gefunden.",
 	"error_ocr_document_not_uploaded": "Das Dokument hat keine Datei — OCR ist nicht möglich.",
 	"error_ocr_processing_failed": "Die OCR-Verarbeitung ist fehlgeschlagen.",
+	"error_training_already_running": "Es läuft bereits ein Trainings-Vorgang.",
 	"ocr_script_type_typewriter": "Schreibmaschine",
 	"ocr_script_type_handwriting_latin": "Handschrift (lateinisch)",
 	"ocr_script_type_handwriting_kurrent": "Handschrift (Kurrent/Sütterlin)",
@@ -530,7 +531,35 @@
 	"ocr_status_analyzing_page": "Seite {current} von {total} wird analysiert…",
 	"ocr_status_done_skipped": "{count} Blöcke erstellt, {skipped} Seite(n) übersprungen",
 	"ocr_status_error": "OCR fehlgeschlagen",
+	"ocr_trigger_no_annotations": "Zeichnen Sie zuerst Bereiche auf dem Dokument ein.",
+	"ocr_section_heading": "OCR ausführen",
 	"transcription_block_review": "Als geprüft markieren",
 	"transcription_block_unreview": "Markierung aufheben",
-	"transcription_reviewed_count": "{reviewed} von {total} geprüft"
+	"transcription_reviewed_count": "{reviewed} von {total} geprüft",
+	"training_ocr_heading": "Kurrent-Erkennung trainieren",
+	"training_ocr_description": "Starte ein neues Training mit den bisher geprüften OCR-Blöcken, um die Erkennungsgenauigkeit für Kurrentschrift zu verbessern.",
+	"training_ocr_blocks_ready": "{blocks} geprüfte Blöcke bereit / {docs} Dokumente",
+	"training_ocr_blocks_total": "(von {total} OCR-Blöcken gesamt)",
+	"training_start_btn": "Training starten",
+	"training_in_progress": "…",
+	"training_success": "Training wurde gestartet und abgeschlossen.",
+	"training_too_few_blocks": "Mindestens 5 geprüfte Blöcke erforderlich (aktuell: {available}).",
+	"training_service_down": "OCR-Dienst ist nicht erreichbar.",
+	"training_history_heading": "Verlauf",
+	"training_history_empty": "Noch keine Trainings-Läufe.",
+	"training_history_col_date": "Datum",
+	"training_history_col_status": "Status",
+	"training_history_col_blocks": "Blöcke",
+	"training_history_col_docs": "Dokumente",
+	"training_history_col_cer": "Fehlerrate",
+	"training_status_done": "Fertig",
+	"training_status_failed": "Fehler",
+	"training_status_running": "Läuft…",
+	"training_seg_heading": "Segmentierung trainieren",
+	"training_seg_description": "Starte ein neues Training mit annotierten Segmentierungsbereichen, um die Texterkennung zu verbessern.",
+	"training_seg_blocks_ready": "{blocks} Segmentierungsblöcke bereit",
+	"training_seg_too_few_blocks": "Mindestens 5 Segmentierungsblöcke erforderlich (aktuell: {available}).",
+	"transcription_block_segmentation_only": "Nur Segmentierung",
+	"training_chip_kurrent": "Kurrent-Erkennung",
+	"training_chip_segmentation": "Segmentierung"
 }

frontend/messages/en.json

@@ -1,7 +1,6 @@
 {
 	"$schema": "https://inlang.com/schema/inlang-message-format",
 	"error_annotation_not_found": "Annotation not found.",
-	"error_annotation_overlap": "The annotation overlaps an existing one.",
 	"annotation_outdated_notice": "Some annotations refer to an earlier file version and are not shown.",
 	"error_document_not_found": "Document not found.",
 	"error_document_no_file": "No file is associated with this document.",
@@ -480,6 +479,7 @@
 	"scan_collapse": "Collapse scan",
 	"transcription_empty_title": "No transcription yet",
 	"transcription_empty_desc": "Draw regions on the scan and type the text to create a transcription.",
+	"transcription_empty_draw_hint": "Draw regions on the document to start transcribing.",
 	"transcription_panel_close": "Close panel",
 	"person_alias_heading": "Name history",
 	"person_alias_empty": "No name changes recorded yet.",
@@ -505,6 +505,7 @@
 	"error_ocr_job_not_found": "The OCR job was not found.",
 	"error_ocr_document_not_uploaded": "The document has no file — OCR is not possible.",
 	"error_ocr_processing_failed": "OCR processing failed.",
+	"error_training_already_running": "A training run is already in progress.",
 	"ocr_script_type_typewriter": "Typewriter",
 	"ocr_script_type_handwriting_latin": "Handwriting (Latin)",
 	"ocr_script_type_handwriting_kurrent": "Handwriting (Kurrent/Sütterlin)",
@@ -530,7 +531,35 @@
 	"ocr_status_analyzing_page": "Analyzing page {current} of {total}…",
 	"ocr_status_done_skipped": "{count} blocks created, {skipped} page(s) skipped",
 	"ocr_status_error": "OCR failed",
+	"ocr_trigger_no_annotations": "Draw regions on the document first.",
+	"ocr_section_heading": "Run OCR",
 	"transcription_block_review": "Mark as reviewed",
 	"transcription_block_unreview": "Unmark as reviewed",
-	"transcription_reviewed_count": "{reviewed} of {total} reviewed"
+	"transcription_reviewed_count": "{reviewed} of {total} reviewed",
+	"training_ocr_heading": "Train Kurrent recognition",
+	"training_ocr_description": "Start a new training run using the reviewed OCR blocks to improve recognition accuracy for Kurrent script.",
+	"training_ocr_blocks_ready": "{blocks} reviewed blocks ready / {docs} documents",
+	"training_ocr_blocks_total": "(of {total} OCR blocks total)",
+	"training_start_btn": "Start training",
+	"training_in_progress": "…",
+	"training_success": "Training started and completed.",
+	"training_too_few_blocks": "At least 5 reviewed blocks required (currently: {available}).",
+	"training_service_down": "OCR service is unavailable.",
+	"training_history_heading": "History",
+	"training_history_empty": "No training runs yet.",
+	"training_history_col_date": "Date",
+	"training_history_col_status": "Status",
+	"training_history_col_blocks": "Blocks",
+	"training_history_col_docs": "Documents",
+	"training_history_col_cer": "Error Rate",
+	"training_status_done": "Done",
+	"training_status_failed": "Failed",
+	"training_status_running": "Running…",
+	"training_seg_heading": "Train segmentation",
+	"training_seg_description": "Start a new training run using annotated segmentation regions to improve text detection.",
+	"training_seg_blocks_ready": "{blocks} segmentation blocks ready",
+	"training_seg_too_few_blocks": "At least 5 segmentation blocks required (currently: {available}).",
+	"transcription_block_segmentation_only": "Segmentation only",
+	"training_chip_kurrent": "Kurrent recognition",
+	"training_chip_segmentation": "Segmentation"
 }

frontend/messages/es.json

@@ -1,7 +1,6 @@
 {
 	"$schema": "https://inlang.com/schema/inlang-message-format",
 	"error_annotation_not_found": "Anotación no encontrada.",
-	"error_annotation_overlap": "La anotación se superpone con una existente.",
 	"annotation_outdated_notice": "Algunas anotaciones hacen referencia a una versión anterior del archivo y no se muestran.",
 	"error_document_not_found": "Documento no encontrado.",
 	"error_document_no_file": "No hay ningún archivo asociado a este documento.",
@@ -480,6 +479,7 @@
 	"scan_collapse": "Reducir escaneo",
 	"transcription_empty_title": "Sin transcripcion",
 	"transcription_empty_desc": "Dibuja regiones en el escaneo y escribe el texto para crear una transcripcion.",
+	"transcription_empty_draw_hint": "Dibuje regiones en el documento para comenzar a transcribir.",
 	"transcription_panel_close": "Cerrar panel",
 	"person_alias_heading": "Historial de nombres",
 	"person_alias_empty": "Aun no se han registrado cambios de nombre.",
@@ -505,6 +505,7 @@
 	"error_ocr_job_not_found": "No se encontró el trabajo OCR.",
 	"error_ocr_document_not_uploaded": "El documento no tiene archivo — OCR no es posible.",
 	"error_ocr_processing_failed": "El procesamiento OCR ha fallado.",
+	"error_training_already_running": "Ya hay un proceso de entrenamiento en curso.",
 	"ocr_script_type_typewriter": "Máquina de escribir",
 	"ocr_script_type_handwriting_latin": "Escritura manuscrita (latina)",
 	"ocr_script_type_handwriting_kurrent": "Escritura manuscrita (Kurrent/Sütterlin)",
@@ -530,7 +531,35 @@
 	"ocr_status_analyzing_page": "Analizando página {current} de {total}…",
 	"ocr_status_done_skipped": "{count} bloques creados, {skipped} página(s) omitida(s)",
 	"ocr_status_error": "OCR fallido",
+	"ocr_trigger_no_annotations": "Dibuje regiones en el documento primero.",
+	"ocr_section_heading": "Ejecutar OCR",
 	"transcription_block_review": "Marcar como revisado",
 	"transcription_block_unreview": "Desmarcar como revisado",
-	"transcription_reviewed_count": "{reviewed} de {total} revisados"
+	"transcription_reviewed_count": "{reviewed} de {total} revisados",
+	"training_ocr_heading": "Entrenar reconocimiento Kurrent",
+	"training_ocr_description": "Inicia un nuevo entrenamiento con los bloques OCR revisados para mejorar la precisión de reconocimiento del script Kurrent.",
+	"training_ocr_blocks_ready": "{blocks} bloques revisados listos / {docs} documentos",
+	"training_ocr_blocks_total": "(de {total} bloques OCR en total)",
+	"training_start_btn": "Iniciar entrenamiento",
+	"training_in_progress": "…",
+	"training_success": "Entrenamiento iniciado y completado.",
+	"training_too_few_blocks": "Se requieren al menos 5 bloques revisados (actualmente: {available}).",
+	"training_service_down": "El servicio OCR no está disponible.",
+	"training_history_heading": "Historial",
+	"training_history_empty": "Todavía no hay ejecuciones de entrenamiento.",
+	"training_history_col_date": "Fecha",
+	"training_history_col_status": "Estado",
+	"training_history_col_blocks": "Bloques",
+	"training_history_col_docs": "Documentos",
+	"training_history_col_cer": "Tasa de error",
+	"training_status_done": "Listo",
+	"training_status_failed": "Error",
+	"training_status_running": "Ejecutando…",
+	"training_seg_heading": "Entrenar segmentación",
+	"training_seg_description": "Inicia un nuevo entrenamiento con regiones de segmentación anotadas para mejorar la detección de texto.",
+	"training_seg_blocks_ready": "{blocks} bloques de segmentación listos",
+	"training_seg_too_few_blocks": "Se requieren al menos 5 bloques de segmentación (actualmente: {available}).",
+	"transcription_block_segmentation_only": "Solo segmentación",
+	"training_chip_kurrent": "Reconocimiento Kurrent",
+	"training_chip_segmentation": "Segmentación"
 }

frontend/src/lib/components/OcrProgress.svelte

@@ -72,7 +72,9 @@ $effect(() => {
 			></div>
 		</div>
 		<p class="mt-2 text-right text-sm text-gray-500">
-			{m.ocr_progress_page({ current: String(currentPage), total: String(totalPages) })}
+			<span aria-live="polite" aria-atomic="true">
+				{m.ocr_progress_page({ current: String(currentPage), total: String(totalPages) })}
+			</span>
 		</p>
 	</div>
 {:else if status === 'error'}

frontend/src/lib/components/OcrTrainingCard.svelte

@@ -0,0 +1,92 @@
+<script lang="ts">
+import TrainingHistory from './TrainingHistory.svelte';
+import { m } from '$lib/paraglide/messages.js';
+
+interface Run {
+	id: string;
+	status: 'RUNNING' | 'DONE' | 'FAILED';
+	blockCount: number;
+	documentCount: number;
+	modelName: string;
+	errorMessage?: string;
+	triggeredBy?: string;
+	createdAt: string;
+	completedAt?: string;
+}
+
+interface TrainingInfo {
+	availableBlocks?: number;
+	totalOcrBlocks?: number;
+	availableDocuments?: number;
+	ocrServiceAvailable?: boolean;
+	lastRun?: Run | null;
+	runs?: Run[];
+}
+
+interface Props {
+	trainingInfo: TrainingInfo | null;
+}
+
+let { trainingInfo }: Props = $props();
+
+let training = $state(false);
+let successMessage = $state<string | null>(null);
+
+const available = $derived(trainingInfo?.availableBlocks ?? 0);
+const tooFewBlocks = $derived(available < 5);
+const serviceDown = $derived(trainingInfo?.ocrServiceAvailable === false);
+const disabled = $derived(training || tooFewBlocks || serviceDown);
+
+async function startTraining() {
+	training = true;
+	successMessage = null;
+	try {
+		const res = await fetch('/api/ocr/train', { method: 'POST' });
+		if (res.ok) {
+			successMessage = m.training_success();
+			setTimeout(() => {
+				successMessage = null;
+			}, 5000);
+		}
+	} finally {
+		training = false;
+	}
+}
+</script>
+
+<div class="rounded-sm border border-line bg-surface p-6 shadow-sm">
+	<h2 class="mb-1 font-sans text-sm font-bold text-ink">{m.training_ocr_heading()}</h2>
+	<p class="mb-4 text-sm text-ink-2">{m.training_ocr_description()}</p>
+
+	<p class="mb-3 text-sm text-ink">
+		{m.training_ocr_blocks_ready({ blocks: available, docs: trainingInfo?.availableDocuments ?? 0 })}
+		<span class="text-ink-2"
+			>{m.training_ocr_blocks_total({ total: trainingInfo?.totalOcrBlocks ?? 0 })}</span
+		>
+	</p>
+
+	<button
+		onclick={startTraining}
+		disabled={disabled}
+		class="rounded-sm bg-primary px-5 py-2 font-sans text-xs font-bold tracking-widest text-primary-fg uppercase transition-opacity hover:opacity-80 focus-visible:ring-2 focus-visible:ring-brand-navy disabled:cursor-not-allowed disabled:opacity-50"
+	>
+		{training ? '…' : m.training_start_btn()}
+	</button>
+
+	{#if tooFewBlocks}
+		<p class="mt-2 text-xs text-ink-3">
+			{m.training_too_few_blocks({ available })}
+		</p>
+	{:else if serviceDown}
+		<p class="mt-2 text-xs text-orange-600">{m.training_service_down()}</p>
+	{/if}
+
+	{#if successMessage}
+		<p class="mt-2 text-xs text-green-700">{successMessage}</p>
+	{/if}
+
+	<h3 class="mt-6 mb-3 text-xs font-bold tracking-widest text-ink-3 uppercase">
+		{m.training_history_heading()}
+	</h3>
+	<TrainingHistory runs={(trainingInfo?.runs ?? []).filter((r) => r.modelName !== 'blla')} />
+</div>

frontend/src/lib/components/OcrTrainingCard.svelte.spec.ts

@@ -0,0 +1,96 @@
+import { afterEach, describe, expect, it, vi } from 'vitest';
+import { cleanup, render } from 'vitest-browser-svelte';
+import { page } from 'vitest/browser';
+import OcrTrainingCard from './OcrTrainingCard.svelte';
+
+afterEach(cleanup);
+afterEach(() => vi.restoreAllMocks());
+
+const baseInfo = {
+	availableBlocks: 10,
+	totalOcrBlocks: 20,
+	availableDocuments: 3,
+	ocrServiceAvailable: true,
+	lastRun: null,
+	runs: []
+};
+
+describe('OcrTrainingCard — disabled states', () => {
+	it('disables button and shows hint when availableBlocks is 0', async () => {
+		render(OcrTrainingCard, { trainingInfo: { ...baseInfo, availableBlocks: 0 } });
+
+		const btn = page.getByRole('button', { name: /Training starten/i });
+		await expect.element(btn).toBeDisabled();
+		await expect
+			.element(page.getByText(/Mindestens 5 geprüfte Blöcke erforderlich/i))
+			.toBeInTheDocument();
+	});
+
+	it('disables button and shows hint when availableBlocks is less than 5', async () => {
+		render(OcrTrainingCard, { trainingInfo: { ...baseInfo, availableBlocks: 3 } });
+
+		const btn = page.getByRole('button', { name: /Training starten/i });
+		await expect.element(btn).toBeDisabled();
+		await expect.element(page.getByText(/Mindestens 5/i)).toBeInTheDocument();
+	});
+
+	it('disables button and shows service-down warning when ocrServiceAvailable is false', async () => {
+		render(OcrTrainingCard, { trainingInfo: { ...baseInfo, ocrServiceAvailable: false } });
+
+		const btn = page.getByRole('button', { name: /Training starten/i });
+		await expect.element(btn).toBeDisabled();
+		await expect.element(page.getByText(/OCR-Dienst ist nicht erreichbar/i)).toBeInTheDocument();
+	});
+
+	it('does not show service-down warning when blocks are insufficient', async () => {
+		// tooFewBlocks hint takes priority over serviceDown hint
+		render(OcrTrainingCard, {
+			trainingInfo: { ...baseInfo, availableBlocks: 2, ocrServiceAvailable: false }
+		});
+
+		await expect.element(page.getByText(/Mindestens 5/i)).toBeInTheDocument();
+		// serviceDown text should NOT appear because tooFewBlocks branch hides it
+		const serviceMsg = document.querySelector('.text-orange-600');
+		expect(serviceMsg).toBeNull();
+	});
+});
+
+describe('OcrTrainingCard — enabled state', () => {
+	it('enables button when availableBlocks >= 5 and service is up', async () => {
+		render(OcrTrainingCard, { trainingInfo: baseInfo });
+
+		const btn = page.getByRole('button', { name: /Training starten/i });
+		await expect.element(btn).not.toBeDisabled();
+	});
+
+	it('shows block count info text', async () => {
+		render(OcrTrainingCard, {
+			trainingInfo: { ...baseInfo, availableBlocks: 7, totalOcrBlocks: 15 }
+		});
+
+		await expect.element(page.getByText(/7/)).toBeInTheDocument();
+		await expect.element(page.getByText(/von 15 OCR-Blöcken/i)).toBeInTheDocument();
+	});
+});
+
+describe('OcrTrainingCard — in-flight state', () => {
+	it('shows "…" while POST is in-flight', async () => {
+		let resolveFetch!: (v: unknown) => void;
+		const pendingFetch = new Promise((resolve) => {
+			resolveFetch = resolve;
+		});
+
+		vi.stubGlobal('fetch', vi.fn().mockReturnValue(pendingFetch));
+
+		render(OcrTrainingCard, { trainingInfo: baseInfo });
+
+		const btn = page.getByRole('button', { name: /Training starten/i });
+		await btn.click();
+
+		// While fetch is still pending the button label becomes "…"
+		await expect.element(page.getByRole('button', { name: '…' })).toBeInTheDocument();
+
+		// Cleanup: resolve the pending promise
+		resolveFetch({ ok: false });
+	});
+});

frontend/src/lib/components/OcrTrigger.svelte

@@ -1,37 +1,23 @@
 <script lang="ts">
-import { untrack } from 'svelte';
 import { m } from '$lib/paraglide/messages.js';
-import { getConfirmService } from '$lib/services/confirm.svelte';
 import ScriptTypeSelect from './ScriptTypeSelect.svelte';
 
 interface Props {
-	existingBlockCount: number;
+	blockCount: number;
 	storedScriptType: string;
-	onTrigger: (scriptType: string) => void;
+	onTrigger: (scriptType: string, useExistingAnnotations: boolean) => void;
 }
 
-let { existingBlockCount, storedScriptType, onTrigger }: Props = $props();
+let { blockCount, storedScriptType, onTrigger }: Props = $props();
-
-const { confirm } = getConfirmService();
 
+import { untrack } from 'svelte';
 let selectedScriptType: string = $state(
 	untrack(() => (storedScriptType && storedScriptType !== 'UNKNOWN' ? storedScriptType : ''))
 );
 
-async function handleClick() {
+function handleClick() {
 	if (!selectedScriptType) return;
-
+	onTrigger(selectedScriptType, true);
-	if (existingBlockCount > 0) {
-		const confirmed = await confirm({
-			title: m.ocr_confirm_title(),
-			body: m.ocr_confirm_body({ count: String(existingBlockCount) }),
-			confirmLabel: m.ocr_confirm_btn(),
-			destructive: true
-		});
-		if (!confirmed) return;
-	}
-
-	onTrigger(selectedScriptType);
 }
 </script>
 
@@ -39,11 +25,14 @@ async function handleClick() {
 	<ScriptTypeSelect bind:value={selectedScriptType} />
 	<button
 		type="button"
-		disabled={!selectedScriptType}
+		disabled={!selectedScriptType || blockCount === 0}
 		title={!selectedScriptType ? m.ocr_trigger_btn_disabled() : undefined}
 		onclick={handleClick}
 		class="min-h-[44px] w-full rounded-sm bg-brand-navy font-sans text-sm font-medium text-white transition-colors hover:bg-brand-navy/90 disabled:cursor-not-allowed disabled:opacity-50"
 	>
 		{m.ocr_trigger_btn()}
 	</button>
+	{#if blockCount === 0}
+		<p class="text-xs text-ink-3">{m.ocr_trigger_no_annotations()}</p>
+	{/if}
 </div>

frontend/src/lib/components/SegmentationTrainingCard.svelte

@@ -0,0 +1,86 @@
+<script lang="ts">
+import TrainingHistory from './TrainingHistory.svelte';
+import { m } from '$lib/paraglide/messages.js';
+
+interface Run {
+	id: string;
+	status: 'RUNNING' | 'DONE' | 'FAILED';
+	blockCount: number;
+	documentCount: number;
+	modelName: string;
+	errorMessage?: string;
+	triggeredBy?: string;
+	createdAt: string;
+	completedAt?: string;
+}
+
+interface TrainingInfo {
+	availableSegBlocks?: number;
+	ocrServiceAvailable?: boolean;
+	runs?: Run[];
+}
+
+interface Props {
+	trainingInfo: TrainingInfo | null;
+}
+
+let { trainingInfo }: Props = $props();
+
+let training = $state(false);
+let successMessage = $state<string | null>(null);
+
+const available = $derived(trainingInfo?.availableSegBlocks ?? 0);
+const tooFewBlocks = $derived(available < 5);
+const serviceDown = $derived(trainingInfo?.ocrServiceAvailable === false);
+const disabled = $derived(training || tooFewBlocks || serviceDown);
+
+async function startTraining() {
+	training = true;
+	successMessage = null;
+	try {
+		const res = await fetch('/api/ocr/segtrain', { method: 'POST' });
+		if (res.ok) {
+			successMessage = m.training_success();
+			setTimeout(() => {
+				successMessage = null;
+			}, 5000);
+		}
+	} finally {
+		training = false;
+	}
+}
+</script>
+
+<div class="rounded-sm border border-line bg-surface p-6 shadow-sm">
+	<h2 class="mb-1 font-sans text-sm font-bold text-ink">{m.training_seg_heading()}</h2>
+	<p class="mb-4 text-sm text-ink-2">{m.training_seg_description()}</p>
+
+	<p class="mb-3 text-sm text-ink">
+		{m.training_seg_blocks_ready({ blocks: available })}
+	</p>
+
+	<button
+		onclick={startTraining}
+		disabled={disabled}
+		class="rounded-sm bg-primary px-5 py-2 font-sans text-xs font-bold tracking-widest text-primary-fg uppercase transition-opacity hover:opacity-80 focus-visible:ring-2 focus-visible:ring-brand-navy disabled:cursor-not-allowed disabled:opacity-50"
+	>
+		{training ? '…' : m.training_start_btn()}
+	</button>
+
+	{#if tooFewBlocks}
+		<p class="mt-2 text-xs text-ink-3">
+			{m.training_seg_too_few_blocks({ available })}
+		</p>
+	{:else if serviceDown}
+		<p class="mt-2 text-xs text-orange-600">{m.training_service_down()}</p>
+	{/if}
+
+	{#if successMessage}
+		<p class="mt-2 text-xs text-green-700">{successMessage}</p>
+	{/if}
+
+	<h3 class="mt-6 mb-3 text-xs font-bold tracking-widest text-ink-3 uppercase">
+		{m.training_history_heading()}
+	</h3>
+	<TrainingHistory runs={(trainingInfo?.runs ?? []).filter((r) => r.modelName === 'blla')} />
+</div>

frontend/src/lib/components/TrainingHistory.svelte

@@ -0,0 +1,119 @@
+<script lang="ts">
+import { m } from '$lib/paraglide/messages.js';
+
+interface Run {
+	id: string;
+	status: 'RUNNING' | 'DONE' | 'FAILED';
+	blockCount: number;
+	documentCount: number;
+	modelName: string;
+	errorMessage?: string;
+	triggeredBy?: string;
+	createdAt: string;
+	completedAt?: string;
+	cer?: number;
+}
+
+interface Props {
+	runs: Run[];
+}
+
+let { runs }: Props = $props();
+
+const dateFormatter = new Intl.DateTimeFormat('de-DE', {
+	day: 'numeric',
+	month: 'short',
+	year: 'numeric'
+});
+
+function formatDate(iso: string): string {
+	return dateFormatter.format(new Date(iso));
+}
+
+function formatCer(cer: number | undefined | null): string {
+	if (cer == null) return '—';
+	return (cer * 100).toFixed(1) + ' %';
+}
+</script>
+
+<table class="w-full text-sm">
+	<thead>
+		<tr class="border-b border-line text-xs font-bold tracking-widest text-ink-3 uppercase">
+			<th class="pb-2 text-left">{m.training_history_col_date()}</th>
+			<th class="pb-2 text-left">{m.training_history_col_status()}</th>
+			<th class="pb-2 text-right">{m.training_history_col_blocks()}</th>
+			<th class="hidden pb-2 text-right md:table-cell">{m.training_history_col_docs()}</th>
+			<th class="hidden pb-2 text-right md:table-cell">{m.training_history_col_cer()}</th>
+		</tr>
+	</thead>
+	<tbody>
+		{#if runs.length === 0}
+			<tr>
+				<td colspan="5" class="py-4 text-center text-sm text-ink-2">
+					{m.training_history_empty()}
+				</td>
+			</tr>
+		{:else}
+			{#each runs as run (run.id)}
+				<tr class="border-b border-line/50 last:border-0">
+					<td class="py-2 text-ink-2">{formatDate(run.createdAt)}</td>
+					<td class="py-2">
+						{#if run.status === 'DONE'}
+							<span
+								class="inline-flex items-center gap-1 rounded-sm bg-green-100 px-1.5 py-0.5 text-xs font-medium text-green-700"
+							>
+								<svg
+									aria-hidden="true"
+									class="h-3 w-3 shrink-0"
+									viewBox="0 0 20 20"
+									fill="currentColor"
+								>
+									<path
+										fill-rule="evenodd"
+										d="M16.707 5.293a1 1 0 010 1.414l-8 8a1 1 0 01-1.414 0l-4-4a1 1 0 011.414-1.414L8 12.586l7.293-7.293a1 1 0 011.414 0z"
+										clip-rule="evenodd"
+									/>
+								</svg>
+								{m.training_status_done()}
+							</span>
+						{:else if run.status === 'FAILED'}
+							<span
+								class="inline-flex items-center gap-1 rounded-sm bg-red-100 px-1.5 py-0.5 text-xs font-medium text-red-700"
+								title={run.errorMessage}
+							>
+								<svg
+									aria-hidden="true"
+									class="h-3 w-3 shrink-0"
+									viewBox="0 0 20 20"
+									fill="currentColor"
+								>
+									<path
+										fill-rule="evenodd"
+										d="M4.293 4.293a1 1 0 011.414 0L10 8.586l4.293-4.293a1 1 0 111.414 1.414L11.414 10l4.293 4.293a1 1 0 01-1.414 1.414L10 11.414l-4.293 4.293a1 1 0 01-1.414-1.414L8.586 10 4.293 5.707a1 1 0 010-1.414z"
+										clip-rule="evenodd"
+									/>
+								</svg>
+								{m.training_status_failed()}
+							</span>
+						{:else}
+							<span
+								class="inline-flex items-center gap-1 rounded-sm bg-yellow-100 px-1.5 py-0.5 text-xs font-medium text-yellow-700"
+							>
+								<span
+									aria-hidden="true"
+									class="h-1.5 w-1.5 animate-pulse rounded-full bg-yellow-500"
+								></span>
+								{m.training_status_running()}
+							</span>
+						{/if}
+					</td>
+					<td class="py-2 text-right text-ink-2">{run.blockCount}</td>
+					<td class="hidden py-2 text-right text-ink-2 md:table-cell">{run.documentCount}</td>
+					<td class="hidden py-2 text-right md:table-cell"
+						>{run.status === 'DONE' && run.cer != null ? formatCer(run.cer) : '—'}</td
+					>
+				</tr>
+			{/each}
+		{/if}
+	</tbody>
+</table>

frontend/src/lib/components/TranscriptionBlock.svelte

@@ -27,6 +27,7 @@ type Props = {
 	onMoveDown?: () => void;
 	isFirst?: boolean;
 	isLast?: boolean;
+	source?: 'MANUAL' | 'OCR';
 };
 
 let {
@@ -48,7 +49,8 @@ let {
 	onMoveUp,
 	onMoveDown,
 	isFirst = false,
-	isLast = false
+	isLast = false,
+	source = 'MANUAL'
 }: Props = $props();
 
 let localText = $state(text);
@@ -172,6 +174,11 @@ function handleTextareaMouseUp() {
 					{label}
 				</span>
 			{/if}
+			{#if (!text || text.trim() === '') && source === 'MANUAL'}
+				<span class="rounded bg-muted px-1.5 py-0.5 text-xs font-medium text-ink-3"
+					>{m.transcription_block_segmentation_only()}</span
+				>
+			{/if}
 		</div>
 
 		<!-- Textarea -->

frontend/src/lib/components/TranscriptionEditView.svelte

@@ -19,7 +19,10 @@ type Props = {
 	onSaveBlock: (blockId: string, text: string) => Promise<void>;
 	onDeleteBlock: (blockId: string) => Promise<void>;
 	onReviewToggle: (blockId: string) => Promise<void>;
-	onTriggerOcr?: (scriptType: string) => void;
+	onTriggerOcr?: (scriptType: string, useExistingAnnotations: boolean) => void;
+	canWrite?: boolean;
+	trainingLabels?: string[];
+	onToggleTrainingLabel?: (label: string, enrolled: boolean) => Promise<void>;
 };
 
 let {
@@ -34,10 +37,14 @@ let {
 	onSaveBlock,
 	onDeleteBlock,
 	onReviewToggle,
-	onTriggerOcr
+	onTriggerOcr,
+	canWrite = false,
+	trainingLabels = [],
+	onToggleTrainingLabel
 }: Props = $props();
 
 let activeBlockId: string | null = $state(null);
+let localLabels: string[] = $derived.by(() => [...trainingLabels]);
 
 // Sync: when an annotation is clicked on the PDF, activate the corresponding block
 $effect(() => {
@@ -50,6 +57,9 @@ let debounceTimers = new SvelteMap<string, ReturnType<typeof setTimeout>>();
 let pendingTexts = new SvelteMap<string, string>();
 let sortedBlocks = $derived([...blocks].sort((a, b) => a.sortOrder - b.sortOrder));
 let hasBlocks = $derived(blocks.length > 0);
+let reviewedCount = $derived(blocks.filter((b) => b.reviewed).length);
+let totalCount = $derived(blocks.length);
+let reviewProgress = $derived(totalCount > 0 ? (reviewedCount / totalCount) * 100 : 0);
 
 function getSaveState(blockId: string): SaveState {
 	return saveStates.get(blockId) ?? 'idle';
@@ -185,7 +195,7 @@ let dropTargetIdx: number | null = $state(null);
 let dragOffsetY: number = $state(0);
 let dragStartY = 0;
 let capturedEl: HTMLElement | null = null;
-let listEl: HTMLElement | null = null;
+let listEl: HTMLElement | null = $state(null);
 
 function handleGripDown(e: PointerEvent, blockId: string) {
 	if (!(e.target as HTMLElement).closest('[data-drag-handle]')) return;
@@ -237,6 +247,23 @@ function handlePointerUp() {
 	capturedEl = null;
 }
 
+async function handleLabelToggle(label: string) {
+	if (!onToggleTrainingLabel) return;
+	const enrolled = !localLabels.includes(label);
+	// Optimistic update
+	if (enrolled) {
+		localLabels = [...localLabels, label];
+	} else {
+		localLabels = localLabels.filter((l) => l !== label);
+	}
+	try {
+		await onToggleTrainingLabel(label, enrolled);
+	} catch {
+		// Revert on failure
+		localLabels = [...trainingLabels];
+	}
+}
+
 function flushViaBeacon() {
 	for (const [blockId, text] of pendingTexts) {
 		clearDebounce(blockId);
@@ -263,78 +290,91 @@ $effect(() => {
 });
 </script>
 
-<div class="flex h-full flex-col overflow-y-auto bg-surface p-4">
+<div class="flex h-full flex-col overflow-y-auto bg-surface">
 	{#if hasBlocks}
-		<!-- svelte-ignore a11y_no_static_element_interactions -->
+		<!-- Sticky review progress header -->
-		<div
+		<div class="sticky top-0 z-10 border-b border-line bg-surface px-4 pt-3 pb-2">
-			class="flex flex-col gap-3"
+			<p class="font-sans text-xs text-ink-2">
-			bind:this={listEl}
+				<span class="font-semibold text-ink">{reviewedCount} / {totalCount}</span> geprüft
-			onpointermove={handlePointerMove}
+			</p>
-			onpointerup={handlePointerUp}
+			<div class="bg-brand-sand mt-1.5 h-0.5 w-full overflow-hidden rounded-full">
-		>
-			{#each sortedBlocks as block, i (block.id)}
-				{#if dropTargetIdx === i}
-					<div class="h-1 rounded-full bg-turquoise transition-all"></div>
-				{/if}
-				<!-- svelte-ignore a11y_no_static_element_interactions -->
 				<div
-					data-block-wrapper
+					class="h-full rounded-full bg-brand-mint transition-all duration-300"
-					onblur={handleBlur}
+					style="width: {reviewProgress}%"
-					onpointerdown={(e) => handleGripDown(e, block.id)}
+				></div>
-					class="relative transition-all duration-150 {draggedBlockId === block.id ? 'z-10 rounded-lg shadow-xl ring-2 ring-turquoise/40' : ''}"
-					style={draggedBlockId === block.id ? `transform: translateY(${dragOffsetY}px) scale(1.02); opacity: 0.9;` : ''}
-				>
-					<TranscriptionBlock
-						blockId={block.id}
-						documentId={documentId}
-						blockNumber={i + 1}
-						text={block.text}
-						label={block.label}
-						active={activeBlockId === block.id}
-						reviewed={block.reviewed ?? false}
-						saveState={getSaveState(block.id)}
-						canComment={canComment}
-						currentUserId={currentUserId}
-						onTextChange={(text) => handleTextChange(block.id, text)}
-						onFocus={() => handleFocus(block.id)}
-						onDeleteClick={() => handleDelete(block.id)}
-						onRetry={() => handleRetry(block.id)}
-						onReviewToggle={() => onReviewToggle(block.id)}
-						onMoveUp={() => handleMoveUp(block.id)}
-						onMoveDown={() => handleMoveDown(block.id)}
-						isFirst={i === 0}
-						isLast={i === sortedBlocks.length - 1}
-					/>
-				</div>
-			{/each}
-
-			{#if dropTargetIdx === sortedBlocks.length}
-				<div class="h-1 rounded-full bg-turquoise transition-all"></div>
-			{/if}
-
-			<!-- Next block CTA — dashed outline hint -->
-			<div
-				class="flex items-center justify-center rounded border border-dashed border-line px-4 py-5 text-center font-sans text-sm text-ink-3"
-			>
-				{m.transcription_next_block_cta({ number: sortedBlocks.length + 1 })}
 			</div>
-
+		</div>
-			{#if canRunOcr && onTriggerOcr}
+		<div class="p-4">
-				<details class="mt-6">
+			<!-- svelte-ignore a11y_no_static_element_interactions -->
-					<summary
+			<div
-						class="cursor-pointer font-sans text-xs font-medium text-ink-3 transition-colors hover:text-brand-navy"
+				class="flex flex-col gap-3"
+				bind:this={listEl}
+				onpointermove={handlePointerMove}
+				onpointerup={handlePointerUp}
+			>
+				{#each sortedBlocks as block, i (block.id)}
+					{#if dropTargetIdx === i}
+						<div class="h-1 rounded-full bg-turquoise transition-all"></div>
+					{/if}
+					<!-- svelte-ignore a11y_no_static_element_interactions -->
+					<div
+						data-block-wrapper
+						onblur={handleBlur}
+						onpointerdown={(e) => handleGripDown(e, block.id)}
+						class="relative transition-all duration-150 {draggedBlockId === block.id ? 'z-10 rounded-lg shadow-xl ring-2 ring-turquoise/40' : ''}"
+						style={draggedBlockId === block.id ? `transform: translateY(${dragOffsetY}px) scale(1.02); opacity: 0.9;` : ''}
 					>
-						{m.ocr_rerun_label()}
+						<TranscriptionBlock
-					</summary>
+							blockId={block.id}
-					<div class="mt-3 max-w-xs">
+							documentId={documentId}
-						<OcrTrigger
+							blockNumber={i + 1}
-							existingBlockCount={blocks.length}
+							text={block.text}
-							storedScriptType={storedScriptType}
+							label={block.label}
-							onTrigger={onTriggerOcr}
+							active={activeBlockId === block.id}
+							reviewed={block.reviewed ?? false}
+							saveState={getSaveState(block.id)}
+							canComment={canComment}
+							currentUserId={currentUserId}
+							onTextChange={(text) => handleTextChange(block.id, text)}
+							onFocus={() => handleFocus(block.id)}
+							onDeleteClick={() => handleDelete(block.id)}
+							onRetry={() => handleRetry(block.id)}
+							onReviewToggle={() => onReviewToggle(block.id)}
+							onMoveUp={() => handleMoveUp(block.id)}
+							onMoveDown={() => handleMoveDown(block.id)}
+							isFirst={i === 0}
+							isLast={i === sortedBlocks.length - 1}
+							source={block.source}
 						/>
 					</div>
-				</details>
+				{/each}
-			{/if}
+
+				{#if dropTargetIdx === sortedBlocks.length}
+					<div class="h-1 rounded-full bg-turquoise transition-all"></div>
+				{/if}
+
+				<!-- Next block CTA — dashed outline hint -->
+				<div
+					class="flex items-center justify-center rounded border border-dashed border-line px-4 py-5 text-center font-sans text-sm text-ink-3"
+				>
+					{m.transcription_next_block_cta({ number: sortedBlocks.length + 1 })}
+				</div>
+
+				{#if canRunOcr && onTriggerOcr}
+					<div class="mt-6">
+						<p class="mb-3 font-sans text-xs font-bold tracking-widest text-ink-3 uppercase">
+							{m.ocr_section_heading()}
+						</p>
+						<div class="max-w-xs">
+							<OcrTrigger
+								blockCount={blocks.length}
+								storedScriptType={storedScriptType}
+								onTrigger={onTriggerOcr}
+							/>
+						</div>
+					</div>
+				{/if}
+			</div>
 		</div>
 	{:else}
 		<div class="flex flex-1 flex-col items-center justify-center px-6 py-12 text-center">
@@ -352,25 +392,28 @@ $effect(() => {
 				/>
 			</svg>
 
-			{#if canRunOcr && onTriggerOcr}
+			<p class="max-w-xs text-sm leading-relaxed text-ink-3">
-				<p class="mb-6 max-w-xs text-sm leading-relaxed text-ink-3">
+				{m.transcription_empty_draw_hint()}
-					{m.transcription_empty_title()}
+			</p>
-				</p>
+		</div>
-				<div class="w-full max-w-xs">
+	{/if}
-					<OcrTrigger
+
-						existingBlockCount={0}
+	{#if canWrite}
-						storedScriptType={storedScriptType}
+		<div class="border-t border-line px-4 py-3">
-						onTrigger={onTriggerOcr}
+			<p class="mb-2 font-sans text-xs font-medium text-ink-2">Für Training vormerken</p>
-					/>
+			<div class="flex flex-wrap gap-2">
-				</div>
+				{#each [{ label: 'KURRENT_RECOGNITION', display: m.training_chip_kurrent() }, { label: 'KURRENT_SEGMENTATION', display: m.training_chip_segmentation() }] as chip (chip.label)}
-				<p class="mt-4 text-xs text-ink-3">
+					<button
-					{m.transcription_empty_desc()}
+						type="button"
-				</p>
+						onclick={() => handleLabelToggle(chip.label)}
-			{:else}
+						class="rounded-full border px-3 py-1 font-sans text-xs font-medium transition-colors {localLabels.includes(chip.label)
-				<p class="max-w-xs text-sm leading-relaxed text-ink-3">
+							? 'border-brand-mint bg-brand-mint text-brand-navy'
-					{m.transcription_empty_cta()}
+							: 'border-line bg-surface text-ink-3 hover:border-brand-mint hover:text-brand-navy'}"
-				</p>
+					>
-			{/if}
+						{chip.display}
+					</button>
+				{/each}
+			</div>
 		</div>
 	{/if}
 </div>

frontend/src/lib/components/TranscriptionEditView.svelte.spec.ts

@@ -13,7 +13,9 @@ const block1 = {
 	text: 'Block eins',
 	label: null,
 	sortOrder: 0,
-	version: 0
+	version: 0,
+	source: 'MANUAL' as const,
+	reviewed: false
 };
 const block2 = {
 	id: 'b2',
@@ -22,7 +24,9 @@ const block2 = {
 	text: 'Block zwei',
 	label: null,
 	sortOrder: 1,
-	version: 0
+	version: 0,
+	source: 'OCR' as const,
+	reviewed: true
 };
 
 function renderView(overrides: Record<string, unknown> = {}, service = createConfirmService()) {
@@ -36,6 +40,7 @@ function renderView(overrides: Record<string, unknown> = {}, service = createCon
 				onBlockFocus: vi.fn(),
 				onSaveBlock: vi.fn(),
 				onDeleteBlock: vi.fn(),
+				onReviewToggle: vi.fn(),
 				...overrides
 			},
 			context: new Map([[CONFIRM_KEY, service]])
@@ -58,7 +63,7 @@ describe('TranscriptionEditView — rendering', () => {
 
 	it('shows empty state when no blocks', async () => {
 		renderView({ blocks: [] });
-		await expect.element(page.getByText(/Markiere einen Bereich/)).toBeInTheDocument();
+		await expect.element(page.getByText(/Zeichnen Sie Bereiche/)).toBeInTheDocument();
 	});
 });
 
@@ -232,3 +237,23 @@ describe('TranscriptionEditView — delete block', () => {
 		expect(onDeleteBlock).not.toHaveBeenCalled();
 	});
 });
+
+// ─── Review progress counter ──────────────────────────────────────────────────
+
+describe('TranscriptionEditView — review progress counter', () => {
+	it('shows reviewed count and total when blocks exist', async () => {
+		// block1: reviewed=false, block2: reviewed=true → "1 / 2 geprüft"
+		renderView();
+		await expect.element(page.getByText(/1 \/ 2 geprüft/)).toBeInTheDocument();
+	});
+
+	it('shows 0 reviewed when no blocks are reviewed', async () => {
+		renderView({ blocks: [block1] }); // block1.reviewed = false
+		await expect.element(page.getByText(/0 \/ 1 geprüft/)).toBeInTheDocument();
+	});
+
+	it('does not show progress counter when there are no blocks', async () => {
+		renderView({ blocks: [] });
+		await expect.element(page.getByText(/geprüft/)).not.toBeInTheDocument();
+	});
+});

frontend/src/lib/errors.ts

@@ -18,7 +18,6 @@ export type ErrorCode =
 	| 'IMPORT_ALREADY_RUNNING'
 	| 'INVALID_RESET_TOKEN'
 	| 'ANNOTATION_NOT_FOUND'
-	| 'ANNOTATION_OVERLAP'
 	| 'TRANSCRIPTION_BLOCK_NOT_FOUND'
 	| 'TRANSCRIPTION_BLOCK_CONFLICT'
 	| 'COMMENT_NOT_FOUND'
@@ -26,6 +25,7 @@ export type ErrorCode =
 	| 'OCR_JOB_NOT_FOUND'
 	| 'OCR_DOCUMENT_NOT_UPLOADED'
 	| 'OCR_PROCESSING_FAILED'
+	| 'TRAINING_ALREADY_RUNNING'
 	| 'UNAUTHORIZED'
 	| 'FORBIDDEN'
 	| 'VALIDATION_ERROR'
@@ -81,8 +81,6 @@ export function getErrorMessage(code: ErrorCode | string | undefined): string {
 			return m.error_invalid_reset_token();
 		case 'ANNOTATION_NOT_FOUND':
 			return m.error_annotation_not_found();
-		case 'ANNOTATION_OVERLAP':
-			return m.error_annotation_overlap();
 		case 'TRANSCRIPTION_BLOCK_NOT_FOUND':
 			return m.error_transcription_block_not_found();
 		case 'TRANSCRIPTION_BLOCK_CONFLICT':
@@ -97,6 +95,8 @@ export function getErrorMessage(code: ErrorCode | string | undefined): string {
 			return m.error_ocr_document_not_uploaded();
 		case 'OCR_PROCESSING_FAILED':
 			return m.error_ocr_processing_failed();
+		case 'TRAINING_ALREADY_RUNNING':
+			return m.error_training_already_running();
 		case 'UNAUTHORIZED':
 			return m.error_unauthorized();
 		case 'FORBIDDEN':

frontend/src/lib/generated/api.ts

@@ -116,6 +116,22 @@ export interface paths {
         patch?: never;
         trace?: never;
     };
+    "/api/documents/{documentId}/transcription-blocks/{blockId}/review": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get?: never;
+        put: operations["reviewBlock"];
+        post?: never;
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
     "/api/documents/{documentId}/transcription-blocks/reorder": {
         parameters: {
             query?: never;
@@ -212,6 +228,54 @@ export interface paths {
         patch?: never;
         trace?: never;
     };
+    "/api/ocr/train": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get?: never;
+        put?: never;
+        post: operations["triggerTraining"];
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
+    "/api/ocr/segtrain": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get?: never;
+        put?: never;
+        post: operations["triggerSegTraining"];
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
+    "/api/ocr/batch": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get?: never;
+        put?: never;
+        post: operations["triggerBatch"];
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
     "/api/notifications/read-all": {
         parameters: {
             query?: never;
@@ -308,6 +372,22 @@ export interface paths {
         patch?: never;
         trace?: never;
     };
+    "/api/documents/{documentId}/ocr": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get?: never;
+        put?: never;
+        post: operations["triggerOcr"];
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
     "/api/documents/{documentId}/comments": {
         parameters: {
             query?: never;
@@ -516,6 +596,22 @@ export interface paths {
         patch: operations["updateGroup"];
         trace?: never;
     };
+    "/api/documents/{id}/training-labels": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get?: never;
+        put?: never;
+        post?: never;
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch: operations["patchTrainingLabel"];
+        trace?: never;
+    };
     "/api/documents/{documentId}/comments/{commentId}": {
         parameters: {
             query?: never;
@@ -628,6 +724,86 @@ export interface paths {
         patch?: never;
         trace?: never;
     };
+    "/api/ocr/training-info": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get: operations["getTrainingInfo"];
+        put?: never;
+        post?: never;
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
+    "/api/ocr/training-data/export": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get: operations["exportTrainingData"];
+        put?: never;
+        post?: never;
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
+    "/api/ocr/segmentation-training-data/export": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get: operations["exportSegmentationTrainingData"];
+        put?: never;
+        post?: never;
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
+    "/api/ocr/jobs/{jobId}": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get: operations["getJobStatus"];
+        put?: never;
+        post?: never;
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
+    "/api/ocr/jobs/{jobId}/progress": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get: operations["streamProgress"];
+        put?: never;
+        post?: never;
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
     "/api/notifications": {
         parameters: {
             query?: never;
@@ -740,6 +916,22 @@ export interface paths {
         patch?: never;
         trace?: never;
     };
+    "/api/documents/{documentId}/ocr-status": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get: operations["getDocumentOcrStatus"];
+        put?: never;
+        post?: never;
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
     "/api/documents/search": {
         parameters: {
             query?: never;
@@ -940,6 +1132,7 @@ export interface components {
             name: string;
         };
         PersonUpdateDTO: {
+            title?: string;
             firstName?: string;
             lastName?: string;
             alias?: string;
@@ -978,6 +1171,8 @@ export interface components {
             receiverIds?: string[];
             tags?: string;
             metadataComplete?: boolean;
+            /** @enum {string} */
+            scriptType?: "UNKNOWN" | "TYPEWRITER" | "HANDWRITING_LATIN" | "HANDWRITING_KURRENT";
         };
         Document: {
             /** Format: uuid */
@@ -1002,9 +1197,12 @@ export interface components {
             /** Format: date-time */
             updatedAt: string;
             metadataComplete: boolean;
+            /** @enum {string} */
+            scriptType: "UNKNOWN" | "TYPEWRITER" | "HANDWRITING_LATIN" | "HANDWRITING_KURRENT";
             receivers?: components["schemas"]["Person"][];
             sender?: components["schemas"]["Person"];
             tags?: components["schemas"]["Tag"][];
+            trainingLabels?: ("KURRENT_RECOGNITION" | "KURRENT_SEGMENTATION")[];
         };
         UpdateTranscriptionBlockDTO: {
             text?: string;
@@ -1017,10 +1215,13 @@ export interface components {
             annotationId: string;
             /** Format: uuid */
             documentId: string;
-            text: string;
+            text?: string;
             label?: string;
             /** Format: int32 */
             sortOrder: number;
+            /** @enum {string} */
+            source: "MANUAL" | "OCR";
+            reviewed: boolean;
             /** Format: int32 */
             version: number;
             /** Format: uuid */
@@ -1068,6 +1269,35 @@ export interface components {
             /** Format: date-time */
             createdAt: string;
         };
+        OcrTrainingRun: {
+            /** Format: uuid */
+            id: string;
+            /** @enum {string} */
+            status: "RUNNING" | "DONE" | "FAILED";
+            /** Format: int32 */
+            blockCount: number;
+            /** Format: int32 */
+            documentCount: number;
+            modelName: string;
+            /** Format: double */
+            cer?: number;
+            /** Format: double */
+            loss?: number;
+            /** Format: double */
+            accuracy?: number;
+            /** Format: int32 */
+            epochs?: number;
+            errorMessage?: string;
+            /** Format: uuid */
+            triggeredBy?: string;
+            /** Format: date-time */
+            createdAt: string;
+            /** Format: date-time */
+            completedAt?: string;
+        };
+        BatchOcrDTO: {
+            documentIds: string[];
+        };
         GroupDTO: {
             name?: string;
             permissions?: string[];
@@ -1118,6 +1348,11 @@ export interface components {
             firstName: string;
             lastName: string;
         };
+        TriggerOcrDTO: {
+            /** @enum {string} */
+            scriptType?: "UNKNOWN" | "TYPEWRITER" | "HANDWRITING_LATIN" | "HANDWRITING_KURRENT";
+            useExistingAnnotations?: boolean;
+        };
         CreateAnnotationDTO: {
             /** Format: int32 */
             pageNumber?: number;
@@ -1130,6 +1365,7 @@ export interface components {
             /** Format: double */
             height?: number;
             color?: string;
+            polygon?: number[][];
         };
         DocumentAnnotation: {
             /** Format: uuid */
@@ -1148,6 +1384,7 @@ export interface components {
             height: number;
             color: string;
             fileHash?: string;
+            polygon?: number[][];
             /** Format: uuid */
             createdBy?: string;
             /** Format: date-time */
@@ -1199,6 +1436,10 @@ export interface components {
             actorName?: string;
             documentTitle?: string;
         };
+        TrainingLabelRequest: {
+            label?: string;
+            enrolled?: boolean;
+        };
         StatsDTO: {
             /** Format: int64 */
             totalPersons?: number;
@@ -1218,15 +1459,54 @@ export interface components {
             deathYear?: number;
             alias?: string;
             notes?: string;
-            personType?: string;
             /** Format: int64 */
             documentCount?: number;
+            personType?: string;
+        };
+        TrainingInfoResponse: {
+            /** Format: int32 */
+            availableBlocks?: number;
+            /** Format: int32 */
+            totalOcrBlocks?: number;
+            /** Format: int32 */
+            availableDocuments?: number;
+            /** Format: int32 */
+            availableSegBlocks?: number;
+            ocrServiceAvailable?: boolean;
+            lastRun?: components["schemas"]["OcrTrainingRun"];
+            runs?: components["schemas"]["OcrTrainingRun"][];
+        };
+        StreamingResponseBody: unknown;
+        OcrJob: {
+            /** Format: uuid */
+            id: string;
+            /** @enum {string} */
+            status: "PENDING" | "RUNNING" | "DONE" | "FAILED";
+            /** Format: int32 */
+            totalDocuments: number;
+            /** Format: int32 */
+            processedDocuments: number;
+            /** Format: int32 */
+            errorCount: number;
+            /** Format: int32 */
+            skippedCount: number;
+            progressMessage?: string;
+            /** Format: uuid */
+            createdBy?: string;
+            /** Format: date-time */
+            createdAt: string;
+            /** Format: date-time */
+            updatedAt: string;
+        };
+        SseEmitter: {
+            /** Format: int64 */
+            timeout?: number;
         };
         PageNotificationDTO: {
-            /** Format: int64 */
-            totalElements?: number;
             /** Format: int32 */
             totalPages?: number;
+            /** Format: int64 */
+            totalElements?: number;
             pageable?: components["schemas"]["PageableObject"];
             /** Format: int32 */
             size?: number;
@@ -1256,10 +1536,6 @@ export interface components {
             empty?: boolean;
             unsorted?: boolean;
         };
-        SseEmitter: {
-            /** Format: int64 */
-            timeout?: number;
-        };
         DocumentVersionSummary: {
             /** Format: uuid */
             id: string;
@@ -1292,6 +1568,15 @@ export interface components {
             /** Format: date-time */
             changedAt: string;
         };
+        OcrStatusDTO: {
+            status?: string;
+            /** Format: uuid */
+            jobId?: string;
+            /** Format: int32 */
+            currentPage?: number;
+            /** Format: int32 */
+            totalPages?: number;
+        };
         DocumentSearchResult: {
             documents?: components["schemas"]["Document"][];
             /** Format: int64 */
@@ -1702,6 +1987,29 @@ export interface operations {
             };
         };
     };
+    reviewBlock: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path: {
+                documentId: string;
+                blockId: string;
+            };
+            cookie?: never;
+        };
+        requestBody?: never;
+        responses: {
+            /** @description OK */
+            200: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "*/*": components["schemas"]["TranscriptionBlock"];
+                };
+            };
+        };
+    };
     reorderBlocks: {
         parameters: {
             query?: never;
@@ -1914,6 +2222,72 @@ export interface operations {
             };
         };
     };
+    triggerTraining: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        requestBody?: never;
+        responses: {
+            /** @description Created */
+            201: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "*/*": components["schemas"]["OcrTrainingRun"];
+                };
+            };
+        };
+    };
+    triggerSegTraining: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        requestBody?: never;
+        responses: {
+            /** @description Created */
+            201: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "*/*": components["schemas"]["OcrTrainingRun"];
+                };
+            };
+        };
+    };
+    triggerBatch: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        requestBody: {
+            content: {
+                "application/json": components["schemas"]["BatchOcrDTO"];
+            };
+        };
+        responses: {
+            /** @description Accepted */
+            202: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "*/*": {
+                        [key: string]: string;
+                    };
+                };
+            };
+        };
+    };
     markAllRead: {
         parameters: {
             query?: never;
@@ -2124,6 +2498,34 @@ export interface operations {
             };
         };
     };
+    triggerOcr: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path: {
+                documentId: string;
+            };
+            cookie?: never;
+        };
+        requestBody: {
+            content: {
+                "application/json": components["schemas"]["TriggerOcrDTO"];
+            };
+        };
+        responses: {
+            /** @description Accepted */
+            202: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "*/*": {
+                        [key: string]: string;
+                    };
+                };
+            };
+        };
+    };
     getDocumentComments: {
         parameters: {
             query?: never;
@@ -2521,6 +2923,30 @@ export interface operations {
             };
         };
     };
+    patchTrainingLabel: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path: {
+                id: string;
+            };
+            cookie?: never;
+        };
+        requestBody: {
+            content: {
+                "application/json": components["schemas"]["TrainingLabelRequest"];
+            };
+        };
+        responses: {
+            /** @description OK */
+            200: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content?: never;
+            };
+        };
+    };
     deleteComment: {
         parameters: {
             query?: never;
@@ -2701,6 +3127,110 @@ export interface operations {
             };
         };
     };
+    getTrainingInfo: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        requestBody?: never;
+        responses: {
+            /** @description OK */
+            200: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "*/*": components["schemas"]["TrainingInfoResponse"];
+                };
+            };
+        };
+    };
+    exportTrainingData: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        requestBody?: never;
+        responses: {
+            /** @description OK */
+            200: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "*/*": components["schemas"]["StreamingResponseBody"];
+                };
+            };
+        };
+    };
+    exportSegmentationTrainingData: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        requestBody?: never;
+        responses: {
+            /** @description OK */
+            200: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "*/*": components["schemas"]["StreamingResponseBody"];
+                };
+            };
+        };
+    };
+    getJobStatus: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path: {
+                jobId: string;
+            };
+            cookie?: never;
+        };
+        requestBody?: never;
+        responses: {
+            /** @description OK */
+            200: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "*/*": components["schemas"]["OcrJob"];
+                };
+            };
+        };
+    };
+    streamProgress: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path: {
+                jobId: string;
+            };
+            cookie?: never;
+        };
+        requestBody?: never;
+        responses: {
+            /** @description OK */
+            200: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "text/event-stream": components["schemas"]["SseEmitter"];
+                };
+            };
+        };
+    };
     getNotifications: {
         parameters: {
             query?: {
@@ -2860,6 +3390,28 @@ export interface operations {
             };
         };
     };
+    getDocumentOcrStatus: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path: {
+                documentId: string;
+            };
+            cookie?: never;
+        };
+        requestBody?: never;
+        responses: {
+            /** @description OK */
+            200: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "*/*": components["schemas"]["OcrStatusDTO"];
+                };
+            };
+        };
+    };
     search_1: {
         parameters: {
             query?: {

frontend/src/routes/admin/system/+page.svelte

@@ -1,6 +1,13 @@
 <script lang="ts">
 import { onDestroy } from 'svelte';
 import { m } from '$lib/paraglide/messages.js';
+import OcrTrainingCard from '$lib/components/OcrTrainingCard.svelte';
+import SegmentationTrainingCard from '$lib/components/SegmentationTrainingCard.svelte';
+import type { components } from '$lib/generated/api.js';
+
+type TrainingInfo = components['schemas']['TrainingInfoResponse'];
+
+let trainingInfo: TrainingInfo | null = $state(null);
 
 let backfillResult: number | null = $state(null);
 let backfillLoading = $state(false);
@@ -51,8 +58,16 @@ async function triggerImport() {
 	}
 }
 
+async function fetchTrainingInfo() {
+	const res = await fetch('/api/ocr/training-info');
+	if (res.ok) {
+		trainingInfo = await res.json();
+	}
+}
+
 $effect(() => {
 	fetchImportStatus();
+	fetchTrainingInfo();
 });
 
 onDestroy(() => stopPolling());
@@ -88,6 +103,12 @@ async function backfillFileHashes() {
 
 <div class="flex-1 overflow-y-auto p-6">
 	<div class="mx-auto max-w-2xl space-y-5">
+		<!-- OCR Recognition Training -->
+		<OcrTrainingCard trainingInfo={trainingInfo} />
+
+		<!-- OCR Segmentation Training -->
+		<SegmentationTrainingCard trainingInfo={trainingInfo} />
+
 		<!-- Backfill versions -->
 		<div class="rounded-sm border border-line bg-surface p-6 shadow-sm">
 			<h2 class="mb-1 font-sans text-sm font-bold text-ink">{m.admin_system_backfill_heading()}</h2>

frontend/src/routes/admin/system/page.svelte.spec.ts

@@ -78,6 +78,8 @@ describe('Admin system page — mass import card', () => {
 					startedAt: null
 				})
 			})
+			// training info fetch → empty
+			.mockResolvedValueOnce({ ok: true, json: async () => ({}) })
 			// trigger POST → returns RUNNING immediately
 			.mockResolvedValueOnce({
 				ok: true,

frontend/src/routes/documents/[id]/+page.svelte

@@ -129,20 +129,29 @@ async function reviewToggle(blockId: string) {
 	transcriptionBlocks = transcriptionBlocks.map((b) => (b.id === blockId ? updated : b));
 }
 
+async function toggleTrainingLabel(label: string, enrolled: boolean) {
+	const res = await fetch(`/api/documents/${doc.id}/training-labels`, {
+		method: 'PATCH',
+		headers: { 'Content-Type': 'application/json' },
+		body: JSON.stringify({ label, enrolled })
+	});
+	if (!res.ok) throw new Error('Failed to update training label');
+}
+
 let ocrRunning = $state(false);
 let ocrProgressMessage = $state('');
 let ocrErrorMessage = $state('');
 let ocrPollTimer = $state<ReturnType<typeof setInterval> | null>(null);
 let ocrSkippedPages = $state(0);
 
-async function triggerOcr(scriptType: string) {
+async function triggerOcr(scriptType: string, useExistingAnnotations: boolean = false) {
 	ocrRunning = true;
 	ocrErrorMessage = '';
 	try {
 		const res = await fetch(`/api/documents/${doc.id}/ocr`, {
 			method: 'POST',
 			headers: { 'Content-Type': 'application/json' },
-			body: JSON.stringify({ scriptType })
+			body: JSON.stringify({ scriptType, useExistingAnnotations })
 		});
 		if (res.ok) {
 			const data = await res.json();
@@ -449,11 +458,14 @@ onMount(() => {
 							activeAnnotationId={activeAnnotationId}
 							storedScriptType={doc.scriptType ?? ''}
 							canRunOcr={canWrite && !!doc.filePath}
+							canWrite={canWrite}
+							trainingLabels={doc.trainingLabels ?? []}
 							onBlockFocus={handleBlockFocus}
 							onSaveBlock={saveBlock}
 							onDeleteBlock={deleteBlock}
 							onReviewToggle={reviewToggle}
 							onTriggerOcr={triggerOcr}
+							onToggleTrainingLabel={toggleTrainingLabel}
 						/>
 					{/if}
 				</div>

ocr-service/Dockerfile

@@ -23,4 +23,4 @@ COPY . .
 
 EXPOSE 8000
 
-CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000", "--workers", "1"]

ocr-service/engines/kraken.py

@@ -47,7 +47,7 @@ def extract_page_blocks(image, page_idx: int, language: str = "de") -> list[dict
     pred_it = rpred.rpred(_model, image, baseline_seg)
 
     for record in pred_it:
-        polygon_pts = record.cuts if hasattr(record, "cuts") else []
+        polygon_pts = record.boundary if hasattr(record, "boundary") and record.boundary else []
 
         if polygon_pts:
             xs = [p[0] for p in polygon_pts]
@@ -55,8 +55,8 @@ def extract_page_blocks(image, page_idx: int, language: str = "de") -> list[dict
             x1, y1 = min(xs), min(ys)
             x2, y2 = max(xs), max(ys)
         else:
-            xs = [p[0] for p in record.line]
+            xs = [p[0] for p in record.baseline]
-            ys = [p[1] for p in record.line]
+            ys = [p[1] for p in record.baseline]
             x1, y1 = min(xs), min(ys) - 5
             x2, y2 = max(xs), max(ys) + 5
 
@@ -79,6 +79,54 @@ def extract_page_blocks(image, page_idx: int, language: str = "de") -> list[dict
     return blocks
 
 
+def extract_region_text(image, x: float, y: float, w: float, h: float) -> str:
+    """Crop image to a normalized region and run Kraken recognition on the crop.
+
+    Used for guided OCR — skips full-page layout detection entirely.
+    A single synthetic baseline spanning the full crop width is used so that
+    blla.segment() (which crashes on small crops) is never called.
+    Coordinates are normalized to [0, 1].
+    """
+    from kraken import rpred
+    from kraken.containers import Segmentation, BaselineLine
+
+    if _model is None:
+        raise RuntimeError("Kraken model is not loaded")
+
+    pw, ph = image.size
+    x1 = max(0, int(x * pw))
+    y1 = max(0, int(y * ph))
+    x2 = min(pw, int((x + w) * pw))
+    y2 = min(ph, int((y + h) * ph))
+    crop = image.crop((x1, y1, x2, y2))
+
+    cw, ch = crop.size
+    if cw == 0 or ch == 0:
+        return ""
+
+    # Single synthetic baseline at vertical midpoint, spanning full crop width.
+    # Kraken's bounds check is >= (not >), so all coordinates must be < image
+    # dimension — use cw-1 / ch-1 to stay strictly inside.
+    mid_y = ch // 2
+    synthetic_seg = Segmentation(
+        type="baselines",
+        imagename="",
+        text_direction="horizontal-lr",
+        script_detection=False,
+        lines=[
+            BaselineLine(
+                id="line0",
+                baseline=[(0, mid_y), (cw - 1, mid_y)],
+                boundary=[(0, 0), (cw - 1, 0), (cw - 1, ch - 1), (0, ch - 1)],
+            )
+        ],
+        regions={},
+        line_orders=[],
+    )
+    pred_it = rpred.rpred(_model, crop, synthetic_seg)
+    return " ".join(r.prediction for r in pred_it)
+
+
 def extract_blocks(images: list, language: str = "de") -> list[dict]:
     """Run Kraken segmentation + recognition on a list of PIL images.
 

ocr-service/engines/surya.py

@@ -81,6 +81,25 @@ def extract_page_blocks(image, page_idx: int, language: str = "de") -> list[dict
     return blocks
 
 
+def extract_region_text(image, x: float, y: float, w: float, h: float) -> str:
+    """Crop image to a normalized region and run Surya recognition on the crop.
+
+    Used for guided OCR — skips full-page layout detection and only processes
+    the given bounding box. Coordinates are normalized to [0, 1].
+    """
+    load_models()
+
+    pw, ph = image.size
+    x1 = max(0, int(x * pw))
+    y1 = max(0, int(y * ph))
+    x2 = min(pw, int((x + w) * pw))
+    y2 = min(ph, int((y + h) * ph))
+    crop = image.crop((x1, y1, x2, y2))
+
+    predictions = _recognition_predictor([crop], det_predictor=_detection_predictor)
+    return " ".join(line.text for line in predictions[0].text_lines)
+
+
 def extract_blocks(images: list, language: str = "de") -> list[dict]:
     """Run Surya OCR on a list of PIL images (one per page).
 

ocr-service/main.py

@@ -1,16 +1,23 @@
 """OCR microservice — FastAPI app with Surya and Kraken engine support."""
 
 import asyncio
+import glob
 import io
 import json
 import logging
 import os
+import re
+import shutil
+import subprocess
+import tempfile
+import zipfile
 from contextlib import asynccontextmanager
+from datetime import datetime, timezone
 from urllib.parse import urlparse
 
 import httpx
 import pypdfium2 as pdfium
-from fastapi import FastAPI, HTTPException
+from fastapi import FastAPI, Header, HTTPException, UploadFile
 from fastapi.responses import StreamingResponse
 from PIL import Image
 
@@ -19,6 +26,9 @@ from engines import kraken as kraken_engine
 from engines import surya as surya_engine
 from models import OcrBlock, OcrRequest
 
+TRAINING_TOKEN = os.environ.get("TRAINING_TOKEN", "")
+KRAKEN_MODEL_PATH = os.environ.get("KRAKEN_MODEL_PATH", "/app/models/german_kurrent.mlmodel")
+
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
 
@@ -100,7 +110,12 @@ async def run_ocr(request: OcrRequest):
 
 @app.post("/ocr/stream")
 async def run_ocr_stream(request: OcrRequest):
-    """Run OCR on a PDF with NDJSON streaming — one JSON line per completed page."""
+    """Run OCR on a PDF with NDJSON streaming — one JSON line per completed page.
+
+    When request.regions is provided, runs in guided mode: each region is cropped
+    and recognized individually, skipping full-page layout detection. The response
+    blocks include the annotationId from the region.
+    """
     if not _models_ready:
         raise HTTPException(status_code=503, detail="Models not loaded yet")
 
@@ -115,6 +130,81 @@ async def run_ocr_stream(request: OcrRequest):
             detail="Kraken model not available — cannot process Kurrent script",
         )
 
+    engine = kraken_engine if use_kraken else surya_engine
+
+    if request.regions:
+        # Guided mode: recognize only the user-drawn annotation regions
+        regions_by_page: dict[int, list] = {}
+        for region in request.regions:
+            regions_by_page.setdefault(region.pageNumber, []).append(region)
+
+        async def generate_guided():
+            total_pages = len(images)
+            yield json.dumps({"type": "start", "totalPages": total_pages}) + "\n"
+
+            total_blocks = 0
+            skipped_pages = 0
+
+            for page_idx, image in enumerate(images, start=1):
+                page_regions = regions_by_page.get(page_idx, [])
+                if not page_regions:
+                    yield json.dumps({
+                        "type": "page",
+                        "pageNumber": page_idx,
+                        "blocks": [],
+                    }) + "\n"
+                    del image
+                    continue
+
+                try:
+                    blocks = []
+                    for region in page_regions:
+                        text = await asyncio.to_thread(
+                            engine.extract_region_text, image,
+                            region.x, region.y, region.width, region.height,
+                        )
+                        blocks.append({
+                            "pageNumber": page_idx,
+                            "x": region.x,
+                            "y": region.y,
+                            "width": region.width,
+                            "height": region.height,
+                            "polygon": None,
+                            "text": text,
+                            "annotationId": region.annotationId,
+                        })
+
+                    total_blocks += len(blocks)
+                    yield json.dumps({
+                        "type": "page",
+                        "pageNumber": page_idx,
+                        "blocks": blocks,
+                    }) + "\n"
+
+                except Exception:
+                    logger.exception("Guided OCR failed on page %d", page_idx)
+                    skipped_pages += 1
+                    yield json.dumps({
+                        "type": "error",
+                        "pageNumber": page_idx,
+                        "message": f"Guided OCR processing failed on page {page_idx}",
+                    }) + "\n"
+
+                finally:
+                    del image
+
+            yield json.dumps({
+                "type": "done",
+                "totalBlocks": total_blocks,
+                "skippedPages": skipped_pages,
+            }) + "\n"
+
+        return StreamingResponse(
+            generate_guided(),
+            media_type="application/x-ndjson",
+            headers={"X-Accel-Buffering": "no", "Cache-Control": "no-cache"},
+        )
+
     async def generate():
         total_pages = len(images)
         yield json.dumps({"type": "start", "totalPages": total_pages}) + "\n"
@@ -124,7 +214,6 @@ async def run_ocr_stream(request: OcrRequest):
 
         for page_idx, image in enumerate(images, start=1):
             try:
-                engine = kraken_engine if use_kraken else surya_engine
                 blocks = await asyncio.to_thread(
                     engine.extract_page_blocks, image, page_idx, request.language
                 )
@@ -169,6 +258,262 @@ async def run_ocr_stream(request: OcrRequest):
     )
 
 
+def _check_training_token(x_training_token: str | None) -> None:
+    """Validate training token — fails closed when TRAINING_TOKEN is not configured."""
+    if not TRAINING_TOKEN:
+        raise HTTPException(status_code=503, detail="Training not configured on this node")
+    if x_training_token != TRAINING_TOKEN:
+        raise HTTPException(status_code=403, detail="Invalid or missing X-Training-Token")
+
+
+def _validate_zip_entry(name: str, extract_dir: str) -> None:
+    """Reject ZIP Slip attacks: path traversal and absolute paths."""
+    if os.path.isabs(name) or name.startswith(".."):
+        raise HTTPException(status_code=400, detail=f"Unsafe ZIP entry: {name}")
+    resolved = os.path.realpath(os.path.join(extract_dir, name))
+    if not resolved.startswith(os.path.realpath(extract_dir)):
+        raise HTTPException(status_code=400, detail=f"ZIP Slip detected: {name}")
+
+
+def _rotate_backups(model_path: str, keep: int = 3) -> None:
+    """Keep only the last `keep` timestamped backups of the model."""
+    pattern = model_path + ".*.bak"
+    backups = sorted(glob.glob(pattern))
+    for old in backups[:-keep]:
+        try:
+            os.remove(old)
+        except OSError:
+            logger.warning("Could not remove old backup: %s", old)
+
+
+def _parse_best_checkpoint(checkpoint_dir: str) -> tuple[float | None, int]:
+    """Parse checkpoint filenames to find the best validation metric and epoch count.
+
+    Kraken saves checkpoints as e.g. ``checkpoint_03-0.9500.ckpt``.
+    Returns (best_accuracy, epoch_count).
+    """
+    pattern = re.compile(r"checkpoint_(\d+)-([0-9.]+)\.(ckpt|mlmodel)$")
+    best_acc: float | None = None
+    max_epoch = 0
+    for fname in os.listdir(checkpoint_dir):
+        m = pattern.match(fname)
+        if m:
+            epoch = int(m.group(1))
+            acc = float(m.group(2))
+            max_epoch = max(max_epoch, epoch)
+            if best_acc is None or acc > best_acc:
+                best_acc = acc
+    return best_acc, max_epoch
+
+
+def _find_best_model(checkpoint_dir: str) -> str | None:
+    """Return the best final model file produced by ketos train.
+
+    With --weights-format coreml, ketos writes ``best_<score>.mlmodel``.
+    Falls back to any .mlmodel in the directory.
+    """
+    # Prefer the named best file (e.g. best_0.8256.mlmodel or best_0.8256.safetensors)
+    best_pattern = re.compile(r"best_([0-9.]+)\.(mlmodel|safetensors)$")
+    best_acc: float | None = None
+    best_path: str | None = None
+    for fname in os.listdir(checkpoint_dir):
+        m = best_pattern.match(fname)
+        if m:
+            acc = float(m.group(1))
+            if best_acc is None or acc > best_acc:
+                best_acc = acc
+                best_path = os.path.join(checkpoint_dir, fname)
+    if best_path:
+        return best_path
+    # Fallback: any .mlmodel file
+    for fname in os.listdir(checkpoint_dir):
+        if fname.endswith(".mlmodel"):
+            return os.path.join(checkpoint_dir, fname)
+    return None
+
+
+@app.post("/train")
+async def train_model(
+    file: UploadFile,
+    x_training_token: str | None = Header(default=None),
+):
+    """Fine-tune the Kurrent recognition model with uploaded training data.
+
+    Accepts a ZIP archive containing .png/.gt.txt training pairs exported
+    by the Java backend. Training mutates in-process model state — not safe
+    if the service is replicated.
+    """
+    _check_training_token(x_training_token)
+
+    if not _models_ready:
+        raise HTTPException(status_code=503, detail="Models not loaded yet")
+
+    zip_bytes = await file.read()
+
+    training_run_id = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ")
+    log = logging.LoggerAdapter(logger, {"training_run_id": training_run_id})
+    log.info("Starting training run %s", training_run_id)
+
+    def _run_training() -> dict:
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            # Extract ZIP with safety checks
+            with zipfile.ZipFile(io.BytesIO(zip_bytes)) as zf:
+                for entry in zf.namelist():
+                    _validate_zip_entry(entry, tmp_dir)
+                zf.extractall(tmp_dir)
+
+            log.info("Extracted %d ZIP entries to %s", len(os.listdir(tmp_dir)), tmp_dir)
+
+            ground_truth = glob.glob(os.path.join(tmp_dir, "*.xml"))
+            if not ground_truth:
+                raise HTTPException(status_code=422, detail="No ground-truth files found in ZIP")
+
+            log.info("Training on %d ground-truth pairs", len(ground_truth))
+            checkpoint_dir = os.path.join(tmp_dir, "checkpoints")
+            os.makedirs(checkpoint_dir, exist_ok=True)
+
+            cmd = [
+                "ketos", "--workers", "0", "--device", "cpu", "--threads", "2",
+                "train",
+                "-f", "page",
+                "--weights-format", "coreml",
+                "-o", checkpoint_dir,
+                "-q", "fixed",
+                "-N", "10",
+                "-B", "1",
+            ]
+            if os.path.exists(KRAKEN_MODEL_PATH):
+                cmd += ["-i", KRAKEN_MODEL_PATH]
+            cmd += ground_truth
+
+            log.info("Running: %s", " ".join(cmd[:6]) + " ...")
+            proc = subprocess.run(cmd, capture_output=True, text=True)
+            log.info("ketos train stdout: %s", proc.stdout[-2000:] if proc.stdout else "")
+            if proc.stderr:
+                log.info("ketos train stderr: %s", proc.stderr[-2000:])
+            if proc.returncode != 0:
+                raise RuntimeError(f"ketos train failed (exit {proc.returncode}): {proc.stderr[-500:]}")
+
+            accuracy, epochs = _parse_best_checkpoint(checkpoint_dir)
+            cer = round(1.0 - accuracy, 4) if accuracy is not None else None
+            log.info("Training complete — epochs=%s accuracy=%s cer=%s", epochs, accuracy, cer)
+
+            # Find the best model file produced by training
+            best_model = _find_best_model(checkpoint_dir)
+            if best_model is None:
+                raise RuntimeError("Training produced no model file")
+
+            # Backup existing model and replace
+            if os.path.exists(KRAKEN_MODEL_PATH):
+                timestamp = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ")
+                backup_path = f"{KRAKEN_MODEL_PATH}.{timestamp}.bak"
+                shutil.copy2(KRAKEN_MODEL_PATH, backup_path)
+                log.info("Backed up model to %s", backup_path)
+                _rotate_backups(KRAKEN_MODEL_PATH, keep=3)
+
+            shutil.copy2(best_model, KRAKEN_MODEL_PATH)
+            log.info("Replaced model at %s", KRAKEN_MODEL_PATH)
+
+            # Reload model in-process
+            kraken_engine.load_models()
+            log.info("Reloaded Kraken model in-process")
+
+            return {"loss": None, "accuracy": accuracy, "cer": cer, "epochs": epochs}
+
+    result = await asyncio.to_thread(_run_training)
+    return result
+
+
+@app.post("/segtrain")
+async def segtrain_model(
+    file: UploadFile,
+    x_training_token: str | None = Header(default=None),
+):
+    """Fine-tune the blla segmentation model with uploaded PAGE XML training data.
+
+    Accepts a ZIP archive containing .png/.xml (PAGE XML) training pairs exported
+    by the Java backend. Training mutates in-process model state — not safe
+    if the service is replicated.
+    """
+    _check_training_token(x_training_token)
+
+    if not _models_ready:
+        raise HTTPException(status_code=503, detail="Models not loaded yet")
+
+    zip_bytes = await file.read()
+
+    training_run_id = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ")
+    log = logging.LoggerAdapter(logger, {"training_run_id": training_run_id})
+    log.info("Starting segmentation training run %s", training_run_id)
+
+    blla_model_path = os.environ.get("BLLA_MODEL_PATH", "/app/models/blla.mlmodel")
+
+    def _run_segtrain() -> dict:
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            with zipfile.ZipFile(io.BytesIO(zip_bytes)) as zf:
+                for entry in zf.namelist():
+                    _validate_zip_entry(entry, tmp_dir)
+                zf.extractall(tmp_dir)
+
+            log.info("Extracted %d ZIP entries for segmentation training", len(os.listdir(tmp_dir)))
+
+            xml_files = glob.glob(os.path.join(tmp_dir, "*.xml"))
+            if not xml_files:
+                raise HTTPException(status_code=422, detail="No PAGE XML files found in ZIP")
+
+            log.info("Training on %d PAGE XML files", len(xml_files))
+            checkpoint_dir = os.path.join(tmp_dir, "checkpoints")
+            os.makedirs(checkpoint_dir, exist_ok=True)
+
+            cmd = [
+                "ketos", "--workers", "0", "--device", "cpu", "--threads", "2",
+                "segtrain",
+                "-o", checkpoint_dir,
+                "-q", "fixed",
+                "-N", "10",
+            ]
+            if os.path.exists(blla_model_path):
+                cmd += ["-i", blla_model_path, "--resize", "both"]
+            else:
+                # No pretrained model — train from scratch with reduced height (800px)
+                # to keep peak RAM under ~200 MB on CPU (default 1800px uses ~500 MB+)
+                cmd += [
+                    "-s",
+                    "[1,800,0,3 Cr7,7,64,2,2 Gn32 Cr3,3,128,2,2 Gn32 Cr3,3,128 Gn32 "
+                    "Cr3,3,256 Gn32 Cr3,3,256 Gn32 Lbx32 Lby32 Cr1,1,32 Gn32 Lby32 Lbx32]",
+                ]
+            cmd += xml_files
+
+            log.info("Running: %s", " ".join(cmd[:5]) + " ...")
+            proc = subprocess.run(cmd, capture_output=True, text=True)
+            log.info("ketos segtrain stdout: %s", proc.stdout[-2000:] if proc.stdout else "")
+            if proc.stderr:
+                log.info("ketos segtrain stderr: %s", proc.stderr[-2000:])
+            if proc.returncode != 0:
+                raise RuntimeError(f"ketos segtrain failed (exit {proc.returncode}): {proc.stderr[-500:]}")
+
+            accuracy, epochs = _parse_best_checkpoint(checkpoint_dir)
+            log.info("Segmentation training complete — epochs=%s accuracy=%s", epochs, accuracy)
+
+            best_model = _find_best_model(checkpoint_dir)
+            if best_model is None:
+                raise RuntimeError("Segmentation training produced no model file")
+
+            if os.path.exists(blla_model_path):
+                timestamp = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ")
+                backup_path = f"{blla_model_path}.{timestamp}.bak"
+                shutil.copy2(blla_model_path, backup_path)
+                _rotate_backups(blla_model_path, keep=3)
+
+            shutil.copy2(best_model, blla_model_path)
+            log.info("Replaced blla model at %s", blla_model_path)
+
+            return {"loss": None, "accuracy": accuracy, "cer": None, "epochs": epochs}
+
+    result = await asyncio.to_thread(_run_segtrain)
+    return result
+
+
 async def _download_and_convert_pdf(url: str) -> list[Image.Image]:
     """Download a PDF from a presigned URL and convert each page to a PIL Image."""
     _validate_url(url)

ocr-service/models.py

@@ -1,12 +1,24 @@
 from pydantic import BaseModel, ConfigDict
 
 
+class OcrRegion(BaseModel):
+    model_config = ConfigDict(populate_by_name=True)
+
+    annotationId: str
+    pageNumber: int
+    x: float
+    y: float
+    width: float
+    height: float
+
+
 class OcrRequest(BaseModel):
     model_config = ConfigDict(populate_by_name=True)
 
     pdfUrl: str
     scriptType: str = "UNKNOWN"
     language: str = "de"
+    regions: list[OcrRegion] | None = None
 
 
 class OcrBlock(BaseModel):
@@ -19,3 +31,4 @@ class OcrBlock(BaseModel):
     height: float
     polygon: list[list[float]] | None = None
     text: str
+    annotationId: str | None = None

ocr-service/test_training_auth.py

@@ -0,0 +1,69 @@
+"""Tests for /train and /segtrain endpoint authentication."""
+
+import io
+import zipfile
+from unittest.mock import AsyncMock, patch
+
+import pytest
+from httpx import ASGITransport, AsyncClient
+
+from main import app
+
+
+def _minimal_zip() -> bytes:
+    """Return a ZIP with one .xml file so endpoint validation passes."""
+    buf = io.BytesIO()
+    with zipfile.ZipFile(buf, "w") as zf:
+        zf.writestr("page_01.xml", "<PcGts/>")
+    return buf.getvalue()
+
+
+# ─── Missing TRAINING_TOKEN → fail closed ─────────────────────────────────────
+
+
+@pytest.mark.asyncio
+async def test_train_returns_503_when_training_token_not_configured():
+    """POST /train must return 503 when TRAINING_TOKEN env var is empty.
+
+    An empty token means the service was started without training configured.
+    Allowing requests through would grant unauthenticated access to the
+    training endpoint, contradicting the principle of failing closed.
+    """
+    with patch("main.TRAINING_TOKEN", ""), \
+         patch("main._models_ready", True):
+        async with AsyncClient(transport=ASGITransport(app=app), base_url="http://test") as client:
+            response = await client.post(
+                "/train",
+                files={"file": ("training.zip", _minimal_zip(), "application/zip")},
+            )
+
+    assert response.status_code == 503
+
+
+@pytest.mark.asyncio
+async def test_segtrain_returns_503_when_training_token_not_configured():
+    """POST /segtrain must return 503 when TRAINING_TOKEN env var is empty."""
+    with patch("main.TRAINING_TOKEN", ""), \
+         patch("main._models_ready", True):
+        async with AsyncClient(transport=ASGITransport(app=app), base_url="http://test") as client:
+            response = await client.post(
+                "/segtrain",
+                files={"file": ("training.zip", _minimal_zip(), "application/zip")},
+            )
+
+    assert response.status_code == 503
+
+
+@pytest.mark.asyncio
+async def test_train_returns_403_when_token_wrong():
+    """POST /train must return 403 when TRAINING_TOKEN is set but header is wrong."""
+    with patch("main.TRAINING_TOKEN", "secret-token"), \
+         patch("main._models_ready", True):
+        async with AsyncClient(transport=ASGITransport(app=app), base_url="http://test") as client:
+            response = await client.post(
+                "/train",
+                files={"file": ("training.zip", _minimal_zip(), "application/zip")},
+                headers={"X-Training-Token": "wrong-token"},
+            )
+
+    assert response.status_code == 403