Pure comparison: same model, same prompts. Toggle the vector on/off, load, and compare outputs.
Open two tabs — one with vector OFF, one ON. Run the same prompts. Compare.