v0.3.2: PyMuPDF extractor, DBLP always-on for low-scoring refs

- Switch primary PDF text extractor to PyMuPDF (fitz); fixes multi-column
  layout garbling (Cinkusz DOI now found, Kostka/Tran titles correct)
- Fall back to pdfminer when PyMuPDF is not installed
- Add pymupdf as a core dependency
- Fix DBLP trigger: query domain sources whenever best candidate score is
  below pass threshold, not only when candidates list is empty; fixes
  Leviathan (ICML) and ReAct (ICLR) reliably finding via DBLP

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

Author: mkassaf

citesentry/__init__.py

@@ -1,3 +1,3 @@
 """citesentry — citation verification tool."""
 
-__version__ = "0.3.1"
+__version__ = "0.3.2"

citesentry/__pycache__/__init__.cpython-311.pyc

@@ -163,7 +163,10 @@ async def check_existence(
             except Exception as e:
                 evidence[f"{src.name}_error"] = str(e)
 
-    if not candidates and domain_sources:
+    # Query domain sources when no good candidate found yet — this handles papers
+    # (e.g. ICML/ICLR proceedings) that Semantic Scholar misses but DBLP covers well.
+    best_score_so_far = max((c[0] for c in candidates), default=0.0)
+    if (not candidates or best_score_so_far < _TITLE_PASS_THRESHOLD / 100.0) and domain_sources:
         for src in domain_sources:
             try:
                 if effective_doi:

citesentry/checks/__pycache__/existence.cpython-311.pyc

@@ -17,11 +17,20 @@
 
 
 def _extract_text(path: Path) -> str:
+    # PyMuPDF handles multi-column layouts and line order far better than pdfminer
+    try:
+        import fitz  # PyMuPDF
+        doc = fitz.open(str(path))
+        text = "\n".join(page.get_text() for page in doc)
+        doc.close()
+        return text
+    except ImportError:
+        pass
     try:
         from pdfminer.high_level import extract_text
         return extract_text(str(path))
     except ImportError as e:
-        raise ImportError("pdfminer.six is required: pip install pdfminer.six") from e
+        raise ImportError("Install pymupdf or pdfminer.six: pip install pymupdf") from e
 
 
 def _find_ref_section(text: str) -> str | None:

citesentry/checks/existence.py

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "citesentry"
-version = "0.3.1"
+version = "0.3.2"
 description = "Citation verification tool: existence, URL liveness, and content relevance checks"
 readme = "README.md"
 requires-python = ">=3.10"
@@ -20,6 +20,7 @@ dependencies = [
     "pdfminer.six>=20221105",
     "mcp[cli]>=1.0",
     "platformdirs>=4",
+    "pymupdf>=1.27.2.3",
 ]
 
 [project.optional-dependencies]