feat: cell-first OCR + document type detection + dynamic pipeline steps

Cell-First OCR (v2): Each cell is cropped and OCR'd in isolation,
eliminating neighbour bleeding (e.g. "to", "ps" in marker columns).
Uses ThreadPoolExecutor for parallel Tesseract calls.

Document type detection: Classifies pages as vocab_table, full_text,
or generic_table using projection profiles (<2s, no OCR needed).
Frontend dynamically skips columns/rows steps for full-text pages.

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

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

@@ -11,7 +11,7 @@ import { StepWordRecognition } from '@/components/ocr-pipeline/StepWordRecogniti
 import { StepLlmReview } from '@/components/ocr-pipeline/StepLlmReview'
 import { StepReconstruction } from '@/components/ocr-pipeline/StepReconstruction'
 import { StepGroundTruth } from '@/components/ocr-pipeline/StepGroundTruth'
-import { PIPELINE_STEPS, type PipelineStep, type SessionListItem } from './types'
+import { PIPELINE_STEPS, type PipelineStep, type SessionListItem, type DocumentTypeResult } from './types'
 
 const KLAUSUR_API = '/klausur-api'
 
@@ -23,6 +23,7 @@ export default function OcrPipelinePage() {
   const [loadingSessions, setLoadingSessions] = useState(true)
   const [editingName, setEditingName] = useState<string | null>(null)
   const [editNameValue, setEditNameValue] = useState('')
+  const [docTypeResult, setDocTypeResult] = useState<DocumentTypeResult | null>(null)
   const [steps, setSteps] = useState<PipelineStep[]>(
     PIPELINE_STEPS.map((s, i) => ({
       ...s,
@@ -59,16 +60,23 @@ export default function OcrPipelinePage() {
       setSessionId(sid)
       setSessionName(data.name || data.filename || '')
 
+      // Restore doc type result if available
+      const savedDocType: DocumentTypeResult | null = data.doc_type_result || null
+      setDocTypeResult(savedDocType)
+
       // Determine which step to jump to based on current_step
       const dbStep = data.current_step || 1
       // Steps: 1=deskew, 2=dewarp, 3=columns, ...
       // UI steps are 0-indexed: 0=deskew, 1=dewarp, 2=columns, ...
       const uiStep = Math.max(0, dbStep - 1)
+      const skipSteps = savedDocType?.skip_steps || []
 
       setSteps(
         PIPELINE_STEPS.map((s, i) => ({
           ...s,
-          status: i < uiStep ? 'completed' : i === uiStep ? 'active' : 'pending',
+          status: skipSteps.includes(s.id)
+            ? 'skipped'
+            : i < uiStep ? 'completed' : i === uiStep ? 'active' : 'pending',
         })),
       )
       setCurrentStep(uiStep)
@@ -84,6 +92,7 @@ export default function OcrPipelinePage() {
       if (sessionId === sid) {
         setSessionId(null)
         setCurrentStep(0)
+        setDocTypeResult(null)
         setSteps(PIPELINE_STEPS.map((s, i) => ({ ...s, status: i === 0 ? 'active' : 'pending' })))
       }
     } catch (e) {
@@ -123,16 +132,28 @@ export default function OcrPipelinePage() {
   }
 
   const handleNext = () => {
-    if (currentStep < steps.length - 1) {
-      setSteps((prev) =>
-        prev.map((s, i) => {
-          if (i === currentStep) return { ...s, status: 'completed' }
-          if (i === currentStep + 1) return { ...s, status: 'active' }
-          return s
-        }),
-      )
-      setCurrentStep((prev) => prev + 1)
+    if (currentStep >= steps.length - 1) return
+
+    // Find the next non-skipped step
+    const skipSteps = docTypeResult?.skip_steps || []
+    let nextStep = currentStep + 1
+    while (nextStep < steps.length && skipSteps.includes(PIPELINE_STEPS[nextStep]?.id)) {
+      nextStep++
     }
+    if (nextStep >= steps.length) nextStep = steps.length - 1
+
+    setSteps((prev) =>
+      prev.map((s, i) => {
+        if (i === currentStep) return { ...s, status: 'completed' }
+        if (i === nextStep) return { ...s, status: 'active' }
+        // Mark skipped steps between current and next
+        if (i > currentStep && i < nextStep && skipSteps.includes(PIPELINE_STEPS[i]?.id)) {
+          return { ...s, status: 'skipped' }
+        }
+        return s
+      }),
+    )
+    setCurrentStep(nextStep)
   }
 
   const handleDeskewComplete = (sid: string) => {
@@ -142,10 +163,69 @@ export default function OcrPipelinePage() {
     handleNext()
   }
 
+  const handleDewarpNext = async () => {
+    // Auto-detect document type after dewarp, then advance
+    if (sessionId) {
+      try {
+        const res = await fetch(
+          `${KLAUSUR_API}/api/v1/ocr-pipeline/sessions/${sessionId}/detect-type`,
+          { method: 'POST' },
+        )
+        if (res.ok) {
+          const data: DocumentTypeResult = await res.json()
+          setDocTypeResult(data)
+
+          // Mark skipped steps immediately
+          const skipSteps = data.skip_steps || []
+          if (skipSteps.length > 0) {
+            setSteps((prev) =>
+              prev.map((s) =>
+                skipSteps.includes(s.id) ? { ...s, status: 'skipped' } : s,
+              ),
+            )
+          }
+        }
+      } catch (e) {
+        console.error('Doc type detection failed:', e)
+        // Not critical — continue without it
+      }
+    }
+    handleNext()
+  }
+
+  const handleDocTypeChange = (newDocType: DocumentTypeResult['doc_type']) => {
+    if (!docTypeResult) return
+
+    // Build new skip_steps based on doc type
+    let skipSteps: string[] = []
+    if (newDocType === 'full_text') {
+      skipSteps = ['columns', 'rows']
+    }
+    // vocab_table and generic_table: no skips
+
+    const updated: DocumentTypeResult = {
+      ...docTypeResult,
+      doc_type: newDocType,
+      skip_steps: skipSteps,
+      pipeline: newDocType === 'full_text' ? 'full_page' : 'cell_first',
+    }
+    setDocTypeResult(updated)
+
+    // Update step statuses
+    setSteps((prev) =>
+      prev.map((s) => {
+        if (skipSteps.includes(s.id)) return { ...s, status: 'skipped' as const }
+        if (s.status === 'skipped') return { ...s, status: 'pending' as const }
+        return s
+      }),
+    )
+  }
+
   const handleNewSession = () => {
     setSessionId(null)
     setSessionName('')
     setCurrentStep(0)
+    setDocTypeResult(null)
     setSteps(PIPELINE_STEPS.map((s, i) => ({ ...s, status: i === 0 ? 'active' : 'pending' })))
   }
 
@@ -188,7 +268,7 @@ export default function OcrPipelinePage() {
       case 0:
         return <StepDeskew sessionId={sessionId} onNext={handleDeskewComplete} />
       case 1:
-        return <StepDewarp sessionId={sessionId} onNext={handleNext} />
+        return <StepDewarp sessionId={sessionId} onNext={handleDewarpNext} />
       case 2:
         return <StepColumnDetection sessionId={sessionId} onNext={handleNext} />
       case 3:
@@ -314,7 +394,14 @@ export default function OcrPipelinePage() {
         </div>
       )}
 
-      <PipelineStepper steps={steps} currentStep={currentStep} onStepClick={handleStepClick} onReprocess={sessionId ? reprocessFromStep : undefined} />
+      <PipelineStepper
+        steps={steps}
+        currentStep={currentStep}
+        onStepClick={handleStepClick}
+        onReprocess={sessionId ? reprocessFromStep : undefined}
+        docTypeResult={docTypeResult}
+        onDocTypeChange={handleDocTypeChange}
+      />
 
       <div className="min-h-[400px]">{renderStep()}</div>
     </div>