fix: use OCR word_box coordinates directly instead of fuzzy matching

The slide positioning hook was re-matching cell.text tokens against
word_boxes via fuzzy text similarity, which broke positioning for
special characters (!, bullet points, IPA). Now uses word_box
coordinates directly — exact OCR positions without re-interpretation.

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

admin-lehrer/components/ocr-overlay/useSlideWordPositions.ts

@@ -46,21 +46,22 @@ export function useSlideWordPositions(
       const hasWordBoxes = cells.some(c => c.word_boxes && c.word_boxes.length > 0)
 
       if (hasWordBoxes) {
-        // --- WORD-BOX PATH: use OCR positions with cell.text tokens ---
+        // --- WORD-BOX PATH: use OCR positions directly ---
-        // Uses fuzzy text matching to pair each token with its best box,
+        // Each word_box already has exact coordinates from OCR.
-        // handling reordering, IPA corrections, and token count mismatches.
+        // Use them as-is — no fuzzy matching needed.
         const positions = new Map<string, WordPosition[]>()
 
         for (const cell of cells) {
           if (!cell.bbox_pct || !cell.text) continue
 
-          const tokens = cell.text.split(/\s+/).filter(Boolean)
-          if (tokens.length === 0) continue
-
           const boxes = (cell.word_boxes || [])
             .filter(wb => wb.text.trim())
+            .sort((a, b) => a.left - b.left)
 
           if (boxes.length === 0) {
+            // No word_boxes — spread tokens evenly across cell
+            const tokens = cell.text.split(/\s+/).filter(Boolean)
+            if (tokens.length === 0) continue
             const fallbackW = cell.bbox_pct.w / tokens.length
             const wordPos = tokens.map((t, i) => ({
               xPct: cell.bbox_pct.x + i * fallbackW,
@@ -72,80 +73,13 @@ export function useSlideWordPositions(
             continue
           }
 
-          // Match each token to its best box by text similarity.
+          // Use each word_box directly with its OCR coordinates
-          // Normalize: lowercase, strip brackets/punctuation for comparison.
+          const wordPos: WordPosition[] = boxes.map(box => ({
-          const norm = (s: string) => s.toLowerCase().replace(/[^a-z0-9äöüß]/g, '')
+            xPct: (box.left / imgW) * 100,
-
+            wPct: (box.width / imgW) * 100,
-          const used = new Set<number>()
+            text: box.text,
-          const tokenBoxIdx: (number | null)[] = []
+            fontRatio: 1.0,
-
+          }))
-          for (const token of tokens) {
-            const tn = norm(token)
-            let bestIdx = -1
-            let bestScore = 0
-
-            for (let bi = 0; bi < boxes.length; bi++) {
-              if (used.has(bi)) continue
-              const bn = norm(boxes[bi].text)
-              // Score: length of common prefix / max length
-              let common = 0
-              const minLen = Math.min(tn.length, bn.length)
-              for (let k = 0; k < minLen; k++) {
-                if (tn[k] === bn[k]) common++
-                else break
-              }
-              // Also check if token is a substring of box text or vice versa
-              const containsBonus = (bn.includes(tn) || tn.includes(bn)) ? 0.5 : 0
-              const score = (minLen > 0 ? common / Math.max(tn.length, bn.length) : 0) + containsBonus
-              if (score > bestScore) {
-                bestScore = score
-                bestIdx = bi
-              }
-            }
-
-            if (bestIdx >= 0 && bestScore > 0.2) {
-              used.add(bestIdx)
-              tokenBoxIdx.push(bestIdx)
-            } else {
-              tokenBoxIdx.push(null) // no match
-            }
-          }
-
-          // Build positions: matched tokens get box positions,
-          // unmatched tokens get interpolated between neighbors.
-          const wordPos: WordPosition[] = []
-
-          for (let ti = 0; ti < tokens.length; ti++) {
-            const bi = tokenBoxIdx[ti]
-            if (bi !== null) {
-              const box = boxes[bi]
-              wordPos.push({
-                xPct: (box.left / imgW) * 100,
-                wPct: (box.width / imgW) * 100,
-                text: tokens[ti],
-                fontRatio: 1.0,
-              })
-            } else {
-              // Interpolate: find nearest matched neighbor before/after
-              let prevPx = cell.bbox_pct.x / 100 * imgW
-              let prevW = 0
-              for (let p = ti - 1; p >= 0; p--) {
-                if (tokenBoxIdx[p] !== null) {
-                  const pb = boxes[tokenBoxIdx[p]!]
-                  prevPx = pb.left + pb.width + 5
-                  prevW = pb.width
-                  break
-                }
-              }
-              const estW = prevW > 0 ? prevW : (cell.bbox_pct.w / 100 * imgW / tokens.length)
-              wordPos.push({
-                xPct: (prevPx / imgW) * 100,
-                wPct: (estW / imgW) * 100,
-                text: tokens[ti],
-                fontRatio: 1.0,
-              })
-            }
-          }
 
           if (wordPos.length > 0) {
             positions.set(cell.cell_id, wordPos)