Add GT button to OCR overlay, prominent category picker, track pipeline

- Ground Truth button on last step of Pipeline/Kombi modes in ocr-overlay
- Prominent category picker in active session info bar (pulses when unset)
- GT badge shown when session has ground truth reference
- Backend: auto-detect pipeline from ocr_engine, store in GT snapshot
- Pipeline info shown in GT session list and regression reports
- Also pass pipeline param from ocr-pipeline StepGroundTruth

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

admin-lehrer/app/(admin)/ai/ocr-overlay/page.tsx

@@ -28,6 +28,10 @@ export default function OcrOverlayPage() {
   const [editNameValue, setEditNameValue] = useState('')
   const [editingCategory, setEditingCategory] = useState<string | null>(null)
   const [activeCategory, setActiveCategory] = useState<DocumentCategory | undefined>(undefined)
+  const [editingActiveCategory, setEditingActiveCategory] = useState(false)
+  const [isGroundTruth, setIsGroundTruth] = useState(false)
+  const [gtSaving, setGtSaving] = useState(false)
+  const [gtMessage, setGtMessage] = useState('')
   const [steps, setSteps] = useState<PipelineStep[]>(
     OVERLAY_PIPELINE_STEPS.map((s, i) => ({
       ...s,
@@ -64,6 +68,8 @@ export default function OcrOverlayPage() {
       setSessionId(sid)
       setSessionName(data.name || data.filename || '')
       setActiveCategory(data.document_category || undefined)
+      setIsGroundTruth(!!data.ground_truth?.build_grid_reference)
+      setGtMessage('')
 
       // Check if this session was processed with paddle_direct, kombi, or rapid_kombi
       const ocrEngine = data.word_result?.ocr_engine
@@ -239,6 +245,33 @@ export default function OcrOverlayPage() {
   // eslint-disable-next-line react-hooks/exhaustive-deps
   }, [sessionId, goToStep])
 
+  const handleMarkGroundTruth = async () => {
+    if (!sessionId) return
+    setGtSaving(true)
+    setGtMessage('')
+    try {
+      const resp = await fetch(
+        `${KLAUSUR_API}/api/v1/ocr-pipeline/sessions/${sessionId}/mark-ground-truth?pipeline=${mode}`,
+        { method: 'POST' }
+      )
+      if (!resp.ok) {
+        const body = await resp.text().catch(() => '')
+        throw new Error(`Ground Truth fehlgeschlagen (${resp.status}): ${body}`)
+      }
+      const data = await resp.json()
+      setIsGroundTruth(true)
+      setGtMessage(`Ground Truth gespeichert (${data.cells_saved} Zellen)`)
+      setTimeout(() => setGtMessage(''), 5000)
+    } catch (e) {
+      setGtMessage(e instanceof Error ? e.message : String(e))
+    } finally {
+      setGtSaving(false)
+    }
+  }
+
+  const isLastStep = currentStep === steps.length - 1
+  const showGtButton = isLastStep && sessionId != null
+
   const renderStep = () => {
     if (mode === 'paddle-direct' || mode === 'kombi') {
       switch (currentStep) {
@@ -472,14 +505,48 @@ export default function OcrOverlayPage() {
         )}
       </div>
 
-      {/* Active session info */}
+      {/* Active session info + category picker */}
       {sessionId && sessionName && (
-        <div className="flex items-center gap-3 text-sm text-gray-500 dark:text-gray-400">
+        <div className="relative flex items-center gap-3 text-sm text-gray-500 dark:text-gray-400">
           <span>Aktive Session: <span className="font-medium text-gray-700 dark:text-gray-300">{sessionName}</span></span>
-          {activeCategory && (() => {
-            const cat = DOCUMENT_CATEGORIES.find(c => c.value === activeCategory)
-            return cat ? <span className="text-xs px-2 py-0.5 rounded-full bg-teal-50 dark:bg-teal-900/30 border border-teal-200 dark:border-teal-700 text-teal-700 dark:text-teal-300">{cat.icon} {cat.label}</span> : null
-          })()}
+          <button
+            onClick={() => setEditingActiveCategory(!editingActiveCategory)}
+            className={`text-xs px-2.5 py-1 rounded-full border transition-colors ${
+              activeCategory
+                ? 'bg-teal-50 dark:bg-teal-900/30 border-teal-200 dark:border-teal-700 text-teal-700 dark:text-teal-300 hover:bg-teal-100 dark:hover:bg-teal-900/50'
+                : 'bg-amber-50 dark:bg-amber-900/20 border-amber-300 dark:border-amber-700 text-amber-700 dark:text-amber-300 hover:bg-amber-100 dark:hover:bg-amber-900/40 animate-pulse'
+            }`}
+          >
+            {activeCategory ? (() => {
+              const cat = DOCUMENT_CATEGORIES.find(c => c.value === activeCategory)
+              return cat ? `${cat.icon} ${cat.label}` : activeCategory
+            })() : 'Kategorie setzen'}
+          </button>
+          {isGroundTruth && (
+            <span className="text-xs px-2 py-0.5 rounded-full bg-amber-50 dark:bg-amber-900/20 border border-amber-300 dark:border-amber-700 text-amber-700 dark:text-amber-300">
+              GT
+            </span>
+          )}
+          {editingActiveCategory && (
+            <div className="absolute left-0 top-full mt-1 z-20 bg-white dark:bg-gray-800 border border-gray-200 dark:border-gray-700 rounded-lg shadow-lg p-2 grid grid-cols-2 gap-1 w-64">
+              {DOCUMENT_CATEGORIES.map((cat) => (
+                <button
+                  key={cat.value}
+                  onClick={() => {
+                    updateCategory(sessionId, cat.value)
+                    setEditingActiveCategory(false)
+                  }}
+                  className={`text-xs px-2 py-1.5 rounded-md text-left transition-colors ${
+                    activeCategory === cat.value
+                      ? 'bg-teal-100 dark:bg-teal-900/40 text-teal-700 dark:text-teal-300'
+                      : 'hover:bg-gray-100 dark:hover:bg-gray-700 text-gray-600 dark:text-gray-400'
+                  }`}
+                >
+                  {cat.icon} {cat.label}
+                </button>
+              ))}
+            </div>
+          )}
         </div>
       )}
 
@@ -543,6 +610,26 @@ export default function OcrOverlayPage() {
       />
 
       <div className="min-h-[400px]">{renderStep()}</div>
+
+      {/* Ground Truth button bar — visible on last step */}
+      {showGtButton && (
+        <div className="sticky bottom-0 bg-white dark:bg-gray-900 border-t dark:border-gray-700 py-3 px-4 -mx-1 flex items-center justify-between rounded-b-xl">
+          <div className="text-sm text-gray-500 dark:text-gray-400">
+            {gtMessage && (
+              <span className={gtMessage.includes('fehlgeschlagen') ? 'text-red-500' : 'text-amber-600 dark:text-amber-400'}>
+                {gtMessage}
+              </span>
+            )}
+          </div>
+          <button
+            onClick={handleMarkGroundTruth}
+            disabled={gtSaving}
+            className="px-4 py-2 text-sm bg-amber-600 text-white rounded hover:bg-amber-700 disabled:opacity-50"
+          >
+            {gtSaving ? 'Speichere...' : isGroundTruth ? 'Ground Truth aktualisieren' : 'Als Ground Truth markieren'}
+          </button>
+        </div>
+      )}
     </div>
   )
 }

admin-lehrer/components/ocr-pipeline/StepGroundTruth.tsx

@@ -210,7 +210,7 @@ export function StepGroundTruth({ sessionId, onNext }: StepGroundTruthProps) {
     setGtMessage('')
     try {
       const resp = await fetch(
-        `${KLAUSUR_API}/api/v1/ocr-pipeline/sessions/${sessionId}/mark-ground-truth`,
+        `${KLAUSUR_API}/api/v1/ocr-pipeline/sessions/${sessionId}/mark-ground-truth?pipeline=ocr-pipeline`,
         { method: 'POST' }
       )
       if (!resp.ok) {

klausur-service/backend/ocr_pipeline_regression.py

@@ -12,7 +12,7 @@ import logging
 from datetime import datetime, timezone
 from typing import Any, Dict, List, Optional
 
-from fastapi import APIRouter, HTTPException
+from fastapi import APIRouter, HTTPException, Query
 
 from grid_editor_api import _build_grid_core
 from ocr_pipeline_session_store import (
@@ -49,7 +49,10 @@ def _extract_cells_for_comparison(grid_result: dict) -> List[Dict[str, Any]]:
     return cells
 
 
-def _build_reference_snapshot(grid_result: dict) -> dict:
+def _build_reference_snapshot(
+    grid_result: dict,
+    pipeline: Optional[str] = None,
+) -> dict:
     """Build a ground-truth reference snapshot from a grid_editor_result."""
     cells = _extract_cells_for_comparison(grid_result)
 
@@ -57,9 +60,10 @@ def _build_reference_snapshot(grid_result: dict) -> dict:
     total_columns = sum(len(z.get("columns", [])) for z in grid_result.get("zones", []))
     total_rows = sum(len(z.get("rows", [])) for z in grid_result.get("zones", []))
 
-    return {
+    snapshot = {
         "saved_at": datetime.now(timezone.utc).isoformat(),
         "version": 1,
+        "pipeline": pipeline,
         "summary": {
             "total_zones": total_zones,
             "total_columns": total_columns,
@@ -68,6 +72,7 @@ def _build_reference_snapshot(grid_result: dict) -> dict:
         },
         "cells": cells,
     }
+    return snapshot
 
 
 def compare_grids(reference: dict, current: dict) -> dict:
@@ -160,7 +165,10 @@ def compare_grids(reference: dict, current: dict) -> dict:
 # ---------------------------------------------------------------------------
 
 @router.post("/sessions/{session_id}/mark-ground-truth")
-async def mark_ground_truth(session_id: str):
+async def mark_ground_truth(
+    session_id: str,
+    pipeline: Optional[str] = Query(None, description="Pipeline used: kombi, pipeline, paddle-direct"),
+):
     """Save the current build-grid result as ground-truth reference."""
     session = await get_session_db(session_id)
     if not session:
@@ -173,7 +181,18 @@ async def mark_ground_truth(session_id: str):
             detail="No grid_editor_result found. Run build-grid first.",
         )
 
-    reference = _build_reference_snapshot(grid_result)
+    # Auto-detect pipeline from word_result if not provided
+    if not pipeline:
+        wr = session.get("word_result") or {}
+        engine = wr.get("ocr_engine", "")
+        if engine in ("kombi", "rapid_kombi"):
+            pipeline = "kombi"
+        elif engine == "paddle_direct":
+            pipeline = "paddle-direct"
+        else:
+            pipeline = "pipeline"
+
+    reference = _build_reference_snapshot(grid_result, pipeline=pipeline)
 
     # Merge into existing ground_truth JSONB
     gt = session.get("ground_truth") or {}
@@ -224,6 +243,8 @@ async def list_ground_truth_sessions():
             "session_id": s["id"],
             "name": s.get("name", ""),
             "filename": s.get("filename", ""),
+            "document_category": s.get("document_category"),
+            "pipeline": ref.get("pipeline"),
             "saved_at": ref.get("saved_at"),
             "summary": ref.get("summary", {}),
         })