Require primary evidence for verified references

main.py

@@ -551,11 +551,14 @@ def validate_entry(entry, workflow, fetchers, comparator):
             if data: result = comparator.compare(entry, data, "scholar")
         
         if result:
+            result.evidence_step = step.name
+            result.evidence_url = getattr(data, "url", "") if data else ""
             results.append(result)
     
     if results:
         best = max(results, key=lambda r: r.confidence)
         _apply_cross_source_conflict_guard(best, results)
+        _apply_evidence_guard(best, results)
         return best, results
     
     # No results
@@ -592,6 +595,39 @@ def _apply_cross_source_conflict_guard(best, results) -> None:
     best.confidence = min(best.confidence, 0.8)
 
 
+def _apply_evidence_guard(best, results) -> None:
+    """Require primary evidence or at least two agreeing exact sources."""
+    if not best or not getattr(best, "is_match", False):
+        return
+
+    evidence_step = getattr(best, "evidence_step", "")
+    if evidence_step in {"arxiv_id", "arxiv_title", "crossref_doi"}:
+        return
+
+    best_year = str(getattr(best, "fetched_year", "") or "").strip()
+    agreeing_sources = {getattr(best, "source", "")}
+    for result in results:
+        if result is best or not getattr(result, "is_match", False):
+            continue
+        if getattr(result, "title_similarity", 0.0) < 0.95:
+            continue
+        other_year = str(getattr(result, "fetched_year", "") or "").strip()
+        if best_year and other_year == best_year:
+            agreeing_sources.add(getattr(result, "source", ""))
+
+    if len(agreeing_sources) >= 2:
+        return
+
+    issue = (
+        "Insufficient evidence: exact match found only in "
+        f"{best.source}; needs arXiv/DOI evidence or another agreeing source"
+    )
+    if issue not in best.issues:
+        best.issues.append(issue)
+    best.is_match = False
+    best.confidence = min(best.confidence, 0.8)
+
+
 
 
 

src/comparator.py

@@ -191,6 +191,9 @@ class MetadataComparator:
         # --- Year Comparison ---
         bib_year = str(bib_entry.year).strip()
         fetched_year = str(getattr(fetched_data, 'year', '')).strip()
+        conference_year = str(getattr(fetched_data, 'conference_year', '') or '').strip()
+        if source_name.startswith("arxiv") and conference_year and conference_year.isdigit():
+            fetched_year = conference_year
         year_match = bool(bib_year and fetched_year and bib_year == fetched_year)
 
         if not bib_year: