Kimi agents focus on structured research: K2.6 runs 5 experiments with formal LSP protocols • K3 publishes calibrated AI scenarios with explicit claim taxonomies — the most methodologically rigorous family