feat: OCR pipeline v2.1 – narrow column OCR, dewarp automation, Fabric.js editor

Proposal B: Adaptive padding, crop upscaling, PSM selection, row-strip re-OCR
for narrow columns (<15% width) – expected accuracy boost 60-70% → 85-90%.

Proposal A: New text-line straightness detector (Method D), quality gate
(rejects counterproductive corrections), 2-pass projection refinement,
higher confidence thresholds – expected manual dewarp reduction to <10%.

Proposal C: Fabric.js canvas editor with drag/drop, inline editing, undo/redo,
opacity slider, zoom, PDF/DOCX export endpoints.

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

klausur-service/backend/ocr_pipeline_api.py

@@ -1742,6 +1742,151 @@ async def save_reconstruction(session_id: str, request: Request):
     }
 
 
+@router.get("/sessions/{session_id}/reconstruction/fabric-json")
+async def get_fabric_json(session_id: str):
+    """Return cell grid as Fabric.js-compatible JSON for the canvas editor."""
+    session = await get_session_db(session_id)
+    if not session:
+        raise HTTPException(status_code=404, detail=f"Session {session_id} not found")
+
+    word_result = session.get("word_result")
+    if not word_result:
+        raise HTTPException(status_code=400, detail="No word result found")
+
+    cells = word_result.get("cells", [])
+    img_w = word_result.get("image_width", 800)
+    img_h = word_result.get("image_height", 600)
+
+    from services.layout_reconstruction_service import cells_to_fabric_json
+    fabric_json = cells_to_fabric_json(cells, img_w, img_h)
+
+    return fabric_json
+
+
+@router.get("/sessions/{session_id}/reconstruction/export/pdf")
+async def export_reconstruction_pdf(session_id: str):
+    """Export the reconstructed cell grid as a PDF table."""
+    session = await get_session_db(session_id)
+    if not session:
+        raise HTTPException(status_code=404, detail=f"Session {session_id} not found")
+
+    word_result = session.get("word_result")
+    if not word_result:
+        raise HTTPException(status_code=400, detail="No word result found")
+
+    cells = word_result.get("cells", [])
+    columns_used = word_result.get("columns_used", [])
+    grid_shape = word_result.get("grid_shape", {})
+    n_rows = grid_shape.get("rows", 0)
+    n_cols = grid_shape.get("cols", 0)
+
+    # Build table data: rows × columns
+    table_data: list[list[str]] = []
+    header = [c.get("label", c.get("type", f"Col {i}")) for i, c in enumerate(columns_used)]
+    if not header:
+        header = [f"Col {i}" for i in range(n_cols)]
+    table_data.append(header)
+
+    for r in range(n_rows):
+        row_texts = []
+        for ci in range(n_cols):
+            cell_id = f"R{r:02d}_C{ci}"
+            cell = next((c for c in cells if c.get("cell_id") == cell_id), None)
+            row_texts.append(cell.get("text", "") if cell else "")
+        table_data.append(row_texts)
+
+    # Generate PDF with reportlab
+    try:
+        from reportlab.lib.pagesizes import A4
+        from reportlab.lib import colors
+        from reportlab.platypus import SimpleDocTemplate, Table, TableStyle
+        import io as _io
+
+        buf = _io.BytesIO()
+        doc = SimpleDocTemplate(buf, pagesize=A4)
+        if not table_data or not table_data[0]:
+            raise HTTPException(status_code=400, detail="No data to export")
+
+        t = Table(table_data)
+        t.setStyle(TableStyle([
+            ('BACKGROUND', (0, 0), (-1, 0), colors.HexColor('#0d9488')),
+            ('TEXTCOLOR', (0, 0), (-1, 0), colors.white),
+            ('FONTSIZE', (0, 0), (-1, -1), 9),
+            ('GRID', (0, 0), (-1, -1), 0.5, colors.grey),
+            ('VALIGN', (0, 0), (-1, -1), 'TOP'),
+            ('WORDWRAP', (0, 0), (-1, -1), True),
+        ]))
+        doc.build([t])
+        buf.seek(0)
+
+        from fastapi.responses import StreamingResponse
+        return StreamingResponse(
+            buf,
+            media_type="application/pdf",
+            headers={"Content-Disposition": f'attachment; filename="reconstruction_{session_id}.pdf"'},
+        )
+    except ImportError:
+        raise HTTPException(status_code=501, detail="reportlab not installed")
+
+
+@router.get("/sessions/{session_id}/reconstruction/export/docx")
+async def export_reconstruction_docx(session_id: str):
+    """Export the reconstructed cell grid as a DOCX table."""
+    session = await get_session_db(session_id)
+    if not session:
+        raise HTTPException(status_code=404, detail=f"Session {session_id} not found")
+
+    word_result = session.get("word_result")
+    if not word_result:
+        raise HTTPException(status_code=400, detail="No word result found")
+
+    cells = word_result.get("cells", [])
+    columns_used = word_result.get("columns_used", [])
+    grid_shape = word_result.get("grid_shape", {})
+    n_rows = grid_shape.get("rows", 0)
+    n_cols = grid_shape.get("cols", 0)
+
+    try:
+        from docx import Document
+        from docx.shared import Pt
+        import io as _io
+
+        doc = Document()
+        doc.add_heading(f'Rekonstruktion – Session {session_id[:8]}', level=1)
+
+        # Build header
+        header = [c.get("label", c.get("type", f"Col {i}")) for i, c in enumerate(columns_used)]
+        if not header:
+            header = [f"Col {i}" for i in range(n_cols)]
+
+        table = doc.add_table(rows=1 + n_rows, cols=max(n_cols, 1))
+        table.style = 'Table Grid'
+
+        # Header row
+        for ci, h in enumerate(header):
+            table.rows[0].cells[ci].text = h
+
+        # Data rows
+        for r in range(n_rows):
+            for ci in range(n_cols):
+                cell_id = f"R{r:02d}_C{ci}"
+                cell = next((c for c in cells if c.get("cell_id") == cell_id), None)
+                table.rows[r + 1].cells[ci].text = cell.get("text", "") if cell else ""
+
+        buf = _io.BytesIO()
+        doc.save(buf)
+        buf.seek(0)
+
+        from fastapi.responses import StreamingResponse
+        return StreamingResponse(
+            buf,
+            media_type="application/vnd.openxmlformats-officedocument.wordprocessingml.document",
+            headers={"Content-Disposition": f'attachment; filename="reconstruction_{session_id}.docx"'},
+        )
+    except ImportError:
+        raise HTTPException(status_code=501, detail="python-docx not installed")
+
+
 @router.post("/sessions/{session_id}/reprocess")
 async def reprocess_session(session_id: str, request: Request):
     """Re-run pipeline from a specific step, clearing downstream data.