Take the author line from the first page, not PDF metadata

Without an authors line the model invented author names on a Space-generated poster.
Lines between the title block and the abstract are used; e-mails and stray combining marks removed.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

pdf_pipeline.py

@@ -1,6 +1,7 @@
 """Paper-only prompt construction. No poster GT or external generation API."""
 import json
 import re
+import unicodedata
 from pathlib import Path
 from PIL import Image
 import fitz
@@ -84,7 +85,20 @@ def prepare(pdf, job, instruction, canvas, accent, columns, custom_spec=''):
         meta_title=clean_text(doc.metadata.get('title',''))
         if meta_title and len(meta_title)>8 and (len(title)<8 or len(title.split())>3 and sum(len(w) for w in title.split())/len(title.split())<2.5):
             title=meta_title[:400]
-        authors=clean_text(doc.metadata.get('author',''))[:500]
+        # authors: the first-page lines between the title block and the abstract (PDF metadata is usually empty,
+        # and without an authors line the model may invent names)
+        title_lines=[l for l in lines if max(s['size'] for s in l)>=largest*.85 and l[0]['bbox'][1]<doc[0].rect.height*.45]
+        title_bottom=max((s['bbox'][3] for l in title_lines for s in l),default=0)
+        author_lines=[]
+        for l in sorted((l for l in lines if l[0]['bbox'][1]>=title_bottom-1 and l not in title_lines),key=lambda l:(round(l[0]['bbox'][1]),l[0]['bbox'][0])):
+            txt=clean_text(join_line(l))
+            if re.match(r'^\s*(abstract|a\s*b\s*s\s*t\s*r\s*a\s*c\s*t)\b',txt,re.I) or l[0]['bbox'][1]>doc[0].rect.height*.5: break
+            if re.search(r'published as a conference|under review|preprint|anonymous author|proceedings of|copyright|workshop',txt,re.I): continue
+            author_lines.append(txt)
+        authors=' '.join(author_lines)
+        authors=re.sub(r'\{[^}]*\}?|\S+@\S+','',authors)            # e-mail lists
+        authors=re.sub(r'\s+([\u0300-\u036f])',r'\1',authors)          # 'Balunovi ́c' -> 'Balunović'
+        authors=clean_text(unicodedata.normalize('NFC',authors))[:500] or clean_text(doc.metadata.get('author',''))[:500]
         asset_lines=[];assets=[]
         for a in chosen:
             cap=clean_text(a.get('caption',''))