fix(resm): audit physical fulltext files

CLAUDE.md

@@ -180,7 +180,8 @@ The paper fetch pipeline in `apps/resm/tasks.py` currently includes:
 Title quality is tracked on `Paper` with the original value retained only when a correction
 is applied. `python manage.py audit_paper_titles` performs a dry-run audit by default;
 `--apply` writes only high-confidence XML corrections unless `--allow-medium` is explicitly
-provided for reviewed PDF candidates.
+provided for reviewed PDF candidates. The audit checks physical XML/PDF files even when
+legacy `has_fulltext_*` flags are stale.
 
 Download behavior is stateful:
 

apps/resm/management/commands/audit_paper_titles.py

@@ -33,13 +33,16 @@ def _paper_path(paper, ext):
 
 def _fulltext_candidate(paper):
     candidates = []
-    if paper.has_fulltext_xml:
-        candidate = extract_title_from_xml(_paper_path(paper, "xml"), paper.doi)
+    xml_path = _paper_path(paper, "xml")
+    pdf_path = _paper_path(paper, "pdf")
+    # 历史状态位可能与磁盘不一致；审计以实际文件为准，避免“文件存在但 flag=false”漏检。
+    if paper.has_fulltext_xml or os.path.isfile(xml_path):
+        candidate = extract_title_from_xml(xml_path, paper.doi)
         if candidate:
             candidates.append(candidate)
-    if paper.has_fulltext_pdf:
+    if paper.has_fulltext_pdf or os.path.isfile(pdf_path):
         candidate = extract_title_from_pdf(
-            _paper_path(paper, "pdf"), paper.doi, paper.first_author or ""
+            pdf_path, paper.doi, paper.first_author or ""
         )
         if candidate:
             candidates.append(candidate)