fix(workflow): extract bulk admissions from decisions, raise pg_trgm threshold to 0.5

Meeting 2 had 5 companies in one agenda item with separate decisions.
Old code only extracted from title. Now also parses each decision for
"Przyjęto firmę X jako" pattern. Raised similarity threshold from 0.3
to 0.5 to avoid false positives (e.g. "Konkol Sp. z o.o." matching "INPI Sp. z o.o.").

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

services/admission_workflow.py

@@ -167,9 +167,8 @@ def extract_admitted_companies(proceedings: list) -> list:
         if isinstance(decisions, str):
             decisions = [decisions]
 
-        # Check if this is an admission proceeding
+        # Collect all admission decisions from this proceeding
-        is_admission = False
+        admission_decisions = []
-        decision_text = ''
         for d in decisions:
             d_lower = d.lower()
             if ('przyjęt' in d_lower and 'jednogłośnie' in d_lower
@@ -177,22 +176,15 @@ def extract_admitted_companies(proceedings: list) -> list:
                     and 'program' not in d_lower
                     and 'protokół' not in d_lower
                     and 'protokol' not in d_lower):
-                is_admission = True
+                admission_decisions.append(d)
-                decision_text = d
-                break
 
-        if not is_admission:
+        if not admission_decisions:
             continue
 
-        # Extract company name from title
+        # Try to extract company name from title first
-        # Pattern 1: "Prezentacja firmy X -- kandydat na czlonka Izby"
+        # Pattern 1: "Prezentacja firmy X — kandydat na członka Izby"
-        # Pattern 2: "Prezentacja firmy X - kandydat na czlonka Izby"
+        # Pattern 2: "Prezentacja: X – coach/mentoring (kandydatka na członka Izby)"
-        # Pattern 3: "Prezentacja: X -- coach/mentoring (kandydatka na czlonka Izby)"
+        title_name = None
-        # Pattern 4: "Prezentacja i glosowanie nad kandydatami..." (bulk - extract from decisions)
-
-        company_name = None
-
-        # Try title patterns
         for pattern in [
             r'[Pp]rezentacja\s+firmy\s+(.+?)\s*[—–\-]\s*kandydat',
             r'[Pp]rezentacja:\s+(.+?)\s*[—–\-]\s*',
@@ -200,25 +192,30 @@ def extract_admitted_companies(proceedings: list) -> list:
         ]:
             match = re.search(pattern, title)
             if match:
-                company_name = match.group(1).strip()
+                title_name = match.group(1).strip().rstrip('.')
                 break
 
-        # If no match from title, try to extract from decision text
+        if title_name:
-        # Pattern: "Przyjeto jednoglosnie firme X jako nowego czlonka Izby"
+            # Single company from title — use first admission decision
-        if not company_name:
-            match = re.search(r'[Pp]rzyjęt[oa]\s+jednogłośnie\s+firmę\s+(.+?)\s+jako', decision_text)
-            if match:
-                company_name = match.group(1).strip()
-
-        if company_name:
-            # Clean up: remove trailing dots, Sp. z o.o. standardization
-            company_name = company_name.rstrip('.')
             results.append({
                 'title': title,
-                'extracted_name': company_name,
+                'extracted_name': title_name,
-                'decision_text': decision_text,
+                'decision_text': admission_decisions[0],
                 'proceeding_index': i
             })
+        else:
+            # Bulk admission — extract company names from each decision
+            # Pattern: "Przyjęto jednogłośnie firmę X jako nowego członka Izby"
+            for d in admission_decisions:
+                match = re.search(r'[Pp]rzyjęt[oa]\s+jednogłośnie\s+firmę\s+(.+?)\s+jako', d)
+                if match:
+                    company_name = match.group(1).strip().rstrip('.')
+                    results.append({
+                        'title': title,
+                        'extracted_name': company_name,
+                        'decision_text': d,
+                        'proceeding_index': i
+                    })
 
     return results
 
@@ -253,7 +250,7 @@ def match_company_by_name(db, name: str) -> tuple:
     # 4. pg_trgm similarity (if extension available)
     try:
         result = db.execute(
-            text("SELECT id, name, similarity(name, :name) as sim FROM companies WHERE similarity(name, :name) > 0.3 ORDER BY sim DESC LIMIT 1"),
+            text("SELECT id, name, similarity(name, :name) as sim FROM companies WHERE similarity(name, :name) > 0.5 ORDER BY sim DESC LIMIT 1"),
             {'name': name}
         ).first()
         if result: