fix: validator false-positive on EXTRACT(YEAR FROM col) + remove hardcoded SQL hints

Agent/agentic_workflow.py

@@ -1648,9 +1648,12 @@ def enhanced_validate_sql(query: str) -> Dict[str, Any]:
         issues.append("Missing SELECT or FROM clause")
 
     # Extract referenced tables (rough pattern)
+    # IMPORTANT: Strip EXTRACT(...FROM...) expressions first to prevent false table detection.
+    # e.g. EXTRACT(YEAR FROM hire_date) would otherwise match 'hire_date' as a table name.
     referenced_tables = set()
     try:
-        table_tokens = re.findall(r'(?:from|join)\s+([a-zA-Z_][a-zA-Z0-9_]*)(?:\s+as)?\s*([a-zA-Z_][a-zA-Z0-9_]*)?', query, flags=re.IGNORECASE)
+        query_for_table_scan = re.sub(r'\bEXTRACT\s*\([^)]+\)', '', query, flags=re.IGNORECASE)
+        table_tokens = re.findall(r'(?:from|join)\s+([a-zA-Z_][a-zA-Z0-9_]*)(?:\s+as)?\s*([a-zA-Z_][a-zA-Z0-9_]*)?', query_for_table_scan, flags=re.IGNORECASE)
         for base, _alias in table_tokens:
             referenced_tables.add(base.lower())
     except Exception:
@@ -3390,24 +3393,12 @@ def query_craft(question: str) -> str:
                 print('[MICRO-HINT] Using micro-generated gender pay gap SQL as LLM hint only.')
                 question = f"{question}\n\n# HINT: The following SQL may help, but do not copy blindly.\n{v}"
         # 7c) Trend over time / over the years (temporal aggregation)
-        # Qwen 1.5b often fails to group by year and returns raw rows ordered by date instead
+        # NOTE: No hardcoded SQL hints here — the semantic clarifier at line ~3511 already rewrites
+        # the question with explicit EXTRACT(YEAR FROM ...) and GROUP BY instructions, which is
+        # sufficient for the LLM. Hardcoded hints were removed to avoid schema coupling.
+        # The validator fix (strip EXTRACT from table-scan) ensures valid EXTRACT queries pass.
         if re.search(r"trend over the years|changed over time|over time|over the years|by year|per year", question.lower()):
-            def gen_trend_hint():
-                q_lower = question.lower()
-                if "hiring" in q_lower or "hire" in q_lower:
-                    if 'employee' not in available_tables:
-                        return None
-                    return "SELECT CAST(EXTRACT(YEAR FROM hire_date) AS INTEGER) AS hire_year, COUNT(*) AS employees_hired FROM employee GROUP BY hire_year ORDER BY hire_year;"
-                elif "salary" in q_lower or "salaries" in q_lower:
-                    if 'salary' not in available_tables:
-                        return None
-                    # NOTE: No to_date filter — we want ALL historical records for trend analysis
-                    return "SELECT CAST(EXTRACT(YEAR FROM from_date) AS INTEGER) AS year, AVG(salary) AS avg_salary FROM salary GROUP BY year ORDER BY year;"
-                return None
-            v = gen_trend_hint()
-            if v:
-                print('[MICRO-HINT] Injecting temporal trend hint (EXTRACT YEAR + GROUP BY)')
-                question = f"{question}\n\n# HINT: The following SQL may help, but do not copy blindly. You MUST use EXTRACT(YEAR FROM date_column) and GROUP BY the year.\n{v}"
+            print('[MICRO-HINT] Temporal trend query detected — semantic clarifier will inject EXTRACT(YEAR) instruction.')
 
         # 8) maximum and minimum salary overall or for a specific department (user question may use max/min or maximum/minimum)
         ql = question.lower()