Use when debugging or verifying numerical parity between pipeline implementations (e.g., research repo vs diffusers, standard vs modular). Also relevant when outputs look wrong — washed out, pixelated, or have visual artifacts — as these are usually parity bugs.