How much genre identity does a chord progression alone carry, and where does it run out? A companion 5-method x 11-genre x 3-seed adaptation study over a frozen pop-pretrained Music Transformer answers bounded, and unevenly: every method improves macro-averaged held-out next-chord accuracy, but none is decisive after multiple-comparison correction, a wrong-genre adapter still beats the base in 81 of 110 cases, and a chord-only genre classifier stays weak (macro-F1 0.171, balanced accuracy 0.225 vs. 11-class chance 0.091). The boundary is a gradient, not a cliff. This demo is designed to make that gradient hearable. We fix a genre-neutral 16-bar frame (plain-triad bookends in C major) and let only the middle eight bars be generated, changing one variable at a time: the un-adapted base, a bossa nova adapter (largest pop-family gain, +3.55 pp), or a rock adapter (smallest gain, +1.14 pp). The expected reading (the author's informal impression, to be verified by listeners) is base ~ rock << bossa: the chords keep some color, never enough to pin the genre. The fixed frame doubles as the stimulus a future controlled listening study would need. We seek feedback on protocol, genre panel, and contrasts before that perceptual step.