Enhance OpenAlex conference scoring to recognize high-quality single-year instances (fixes #77)

The previous scoring algorithm applied a harsh 70% penalty to all conference
instances with 1-2 years of activity, regardless of their quality metrics.
This incorrectly penalized major conferences like CVPR 2022 (2,082 papers,
179,530 citations) and ICCV 2021 (1,618 papers, 124,236 citations), causing
them to be ranked below poor-quality matches.

The enhanced algorithm now:
- Calculates publication volume and citation impact before applying year-span logic
- Identifies high-quality conferences (>50k citations OR >1000 papers)
- Applies bonuses to high-quality single-year instances instead of penalties
- Only penalizes low-quality short-span conferences (with reduced penalty)

This ensures legitimate year-specific conference instances receive appropriate
scores based on their actual quality metrics.

[AI-assisted]

Author: florath

src/aletheia_probe/openalex.py

@@ -119,14 +119,6 @@ def _score_source_match(self, source: dict[str, Any], journal_name: str) -> floa
         elif any(word in display_name for word in search_name.split() if len(word) > 3):
             score += 0.2
 
-        # Avoid year-specific conference instances (conferences with only 1-2 years active)
-        if source_type == "conference" and first_year and last_year:
-            years_active = last_year - first_year + 1
-            if years_active <= 2:
-                score *= 0.3  # Heavily penalize single-year instances
-            elif years_active >= 10:
-                score += 0.1  # Bonus for long-running venues
-
         # Publication volume (30% of score)
         if works_count > 1000:
             score += 0.3
@@ -147,6 +139,26 @@ def _score_source_match(self, source: dict[str, Any], journal_name: str) -> floa
         elif cited_by_count <= 10:
             score *= 0.5  # Penalize low-impact sources
 
+        # Enhanced conference scoring: Consider quality metrics before penalizing short spans
+        if source_type == "conference" and first_year and last_year:
+            years_active = last_year - first_year + 1
+
+            # Determine if this is a high-quality conference instance
+            is_high_quality = (cited_by_count > 50000) or (works_count > 1000)
+            is_medium_quality = (cited_by_count > 10000) or (works_count > 500)
+
+            if years_active <= 2:
+                # High-quality single-year instances are legitimate (e.g., CVPR 2022)
+                if is_high_quality:
+                    score += 0.15  # Bonus for major conference instance
+                elif is_medium_quality:
+                    score += 0.05  # Small bonus for good conference instance
+                else:
+                    # Only penalize low-quality short-span conferences
+                    score *= 0.4  # Reduced penalty (was 0.3)
+            elif years_active >= 10:
+                score += 0.1  # Bonus for long-running venues
+
         # Recency (10% of score) - penalize inactive sources
         if last_year:
             current_year = datetime.now().year