Thanks for the valuable feedback! The 0.7/0.3 blend is part of my learning implementation, so I’ll test query variations and compare raw, normalized, and rank-fusion approaches next.