Runs 2 real golden-set questions through the actual connected model — hybrid retrieval vs. a naive keyword-overlap baseline.