feat(training): track and display CER per training run

After each training run, the Character Error Rate (CER = 1 - accuracy),
loss, accuracy, and epoch count are now stored on the OcrTrainingRun
record and shown in the training history table.

Also adds the missing POST /api/ocr/segtrain endpoint and the
triggerSegTraining service method so the segmentation training card
can actually trigger training.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

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

@@ -120,6 +120,14 @@ public class OcrController {
         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() {

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

@@ -41,6 +41,18 @@ public class OcrTrainingRun {
     @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;
 

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

@@ -26,7 +26,7 @@ public interface OcrClient {
      */
     TrainingResult trainModel(byte[] trainingDataZip);
 
-    record TrainingResult(Double loss, Double accuracy, Integer epochs) {}
+    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.

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

@@ -80,12 +80,16 @@ public class OcrTrainingService {
             byte[] zipBytes = baos.toByteArray();
 
             log.info("[trainingRun={}] Sending {} bytes to OCR service", runId, zipBytes.length);
-            ocrClient.trainModel(zipBytes);
+            OcrClient.TrainingResult result = ocrClient.trainModel(zipBytes);
 
             run.setStatus(TrainingStatus.DONE);
             run.setCompletedAt(Instant.now());
+            run.setCer(result.cer());
+            run.setLoss(result.loss());
+            run.setAccuracy(result.accuracy());
+            run.setEpochs(result.epochs());
             run = trainingRunRepository.save(run);
-            log.info("[trainingRun={}] Training completed successfully", runId);
+            log.info("[trainingRun={}] Training completed — cer={} epochs={}", runId, result.cer(), result.epochs());
         } catch (Exception e) {
             run.setStatus(TrainingStatus.FAILED);
             run.setErrorMessage(e.getMessage());
@@ -99,6 +103,66 @@ public class OcrTrainingService {
         return run;
     }
 
+    @Transactional
+    public OcrTrainingRun triggerSegTraining(UUID triggeredBy) {
+        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 run = OcrTrainingRun.builder()
+                .status(TrainingStatus.RUNNING)
+                .blockCount(segBlocks.size())
+                .documentCount((int) documentCount)
+                .modelName("blla")
+                .triggeredBy(triggeredBy)
+                .build();
+        run = trainingRunRepository.save(run);
+
+        String runId = run.getId().toString();
+        MDC.put("trainingRunId", runId);
+        log.info("Started segmentation training run {} with {} segments from {} documents",
+                runId, segBlocks.size(), documentCount);
+
+        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);
+
+            run.setStatus(TrainingStatus.DONE);
+            run.setCompletedAt(Instant.now());
+            run.setLoss(result.loss());
+            run.setAccuracy(result.accuracy());
+            run.setEpochs(result.epochs());
+            run = trainingRunRepository.save(run);
+            log.info("[trainingRun={}] Segmentation training completed — epochs={}", runId, result.epochs());
+        } catch (Exception e) {
+            run.setStatus(TrainingStatus.FAILED);
+            run.setErrorMessage(e.getMessage());
+            run.setCompletedAt(Instant.now());
+            run = trainingRunRepository.save(run);
+            log.error("[trainingRun={}] Segmentation training failed: {}", runId, e.getMessage(), e);
+        } finally {
+            MDC.remove("trainingRunId");
+        }
+
+        return run;
+    }
+
     public TrainingInfoResponse getTrainingInfo() {
         var eligibleBlocks = trainingDataExportService.queryEligibleBlocks();
         int availableDocuments = (int) eligibleBlocks.stream()

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

@@ -127,8 +127,8 @@ public class RestClientOcrClient implements OcrClient, OcrHealthClient {
                 .retrieve()
                 .body(TrainingResultJson.class);
 
-        if (result == null) return new OcrClient.TrainingResult(null, null, null);
+        if (result == null) return new OcrClient.TrainingResult(null, null, null, null);
-        return new OcrClient.TrainingResult(result.loss(), result.accuracy(), result.epochs());
+        return new OcrClient.TrainingResult(result.loss(), result.accuracy(), result.cer(), result.epochs());
     }
 
     @Override
@@ -156,8 +156,8 @@ public class RestClientOcrClient implements OcrClient, OcrHealthClient {
                 .retrieve()
                 .body(TrainingResultJson.class);
 
-        if (result == null) return new OcrClient.TrainingResult(null, null, null);
+        if (result == null) return new OcrClient.TrainingResult(null, null, null, null);
-        return new OcrClient.TrainingResult(result.loss(), result.accuracy(), result.epochs());
+        return new OcrClient.TrainingResult(result.loss(), result.accuracy(), result.cer(), result.epochs());
     }
 
     @Override
@@ -255,7 +255,7 @@ public class RestClientOcrClient implements OcrClient, OcrHealthClient {
         }
     }
 
-    record TrainingResultJson(Double loss, Double accuracy, Integer epochs) {}
+    record TrainingResultJson(Double loss, Double accuracy, Double cer, Integer epochs) {}
 
     record OcrBlockJson(
             @JsonProperty("pageNumber") int pageNumber,

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/service/OcrTrainingServiceTest.java

@@ -93,7 +93,7 @@ class OcrTrainingServiceTest {
         );
         when(exportService.queryEligibleBlocks()).thenReturn(blocks);
         when(exportService.exportToZip()).thenReturn(out -> {});
-        when(ocrClient.trainModel(any())).thenReturn(new OcrClient.TrainingResult(0.05, 0.95, 3));
+        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)

frontend/messages/de.json

@@ -550,6 +550,7 @@
 	"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…",

frontend/messages/en.json

@@ -550,6 +550,7 @@
 	"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…",

frontend/messages/es.json

@@ -550,6 +550,7 @@
 	"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…",

frontend/src/lib/components/TrainingHistory.svelte

@@ -11,6 +11,7 @@ interface Run {
 	triggeredBy?: string;
 	createdAt: string;
 	completedAt?: string;
+	cer?: number;
 }
 
 interface Props {
@@ -28,6 +29,11 @@ const dateFormatter = new Intl.DateTimeFormat('de-DE', {
 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">
@@ -37,12 +43,13 @@ function formatDate(iso: string): string {
 			<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="4" class="py-4 text-center text-sm text-ink-2">
+				<td colspan="5" class="py-4 text-center text-sm text-ink-2">
 					{m.training_history_empty()}
 				</td>
 			</tr>
@@ -71,6 +78,9 @@ function formatDate(iso: string): string {
 					</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}

ocr-service/main.py

@@ -333,8 +333,10 @@ async def train_model(
             epochs = getattr(result, "epochs", None) or 0
             loss = getattr(result, "best_loss", None)
             accuracy = getattr(result, "best_accuracy", None)
+            # CER = 1 - character accuracy (Kraken reports character-level accuracy)
+            cer = round(1.0 - accuracy, 4) if accuracy is not None else None
 
-            log.info("Training complete — epochs=%s loss=%s accuracy=%s", epochs, loss, accuracy)
+            log.info("Training complete — epochs=%s loss=%s accuracy=%s cer=%s", epochs, loss, accuracy, cer)
 
             # Backup existing model and replace
             if os.path.exists(KRAKEN_MODEL_PATH):
@@ -351,7 +353,7 @@ async def train_model(
             kraken_engine.load_models()
             log.info("Reloaded Kraken model in-process")
 
-            return {"loss": loss, "accuracy": accuracy, "epochs": epochs}
+            return {"loss": loss, "accuracy": accuracy, "cer": cer, "epochs": epochs}
 
     result = await asyncio.to_thread(_run_training)
     return result
@@ -409,7 +411,7 @@ async def segtrain_model(
             loss = getattr(result, "best_loss", None)
             accuracy = getattr(result, "best_accuracy", None)
 
-            log.info("Segmentation training complete — epochs=%s loss=%s", epochs, loss)
+            log.info("Segmentation training complete — epochs=%s loss=%s accuracy=%s", epochs, loss, accuracy)
 
             if os.path.exists(blla_model_path):
                 timestamp = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ")
@@ -420,7 +422,7 @@ async def segtrain_model(
             shutil.move(output_model_path, blla_model_path)
             log.info("Replaced blla model at %s", blla_model_path)
 
-            return {"loss": loss, "accuracy": accuracy, "epochs": epochs}
+            return {"loss": loss, "accuracy": accuracy, "cer": None, "epochs": epochs}
 
     result = await asyncio.to_thread(_run_segtrain)
     return result