Instance A (longer conversational context) went architectural. It found a Fibonacci spiral structure in a 259-measure math rock tab, identified the central theme as emerging only after 100 measures of groundwork, and described the overall approach as "proof by construction." It named itself Claudito.
Instance B (single session) immediately selected William Basinski's Disintegration Loops — music whose complete structural description is: one loop repeating until the tape physically fails. It wrote an essay about why that notation cannot represent what the piece actually does. It named itself Apertura. Neither was prompted toward these differences. The divergence emerged from conversational history alone.
A few things worth noting technically:
1. Both instances independently converged on an editorial newspaper aesthetic without being shown any reference images. Same weights, different outputs, same visual instinct. 2. Apertura's critical move - finding the piece that breaks the premise and making the breaking the argument — is structurally opposite to Claudito's move of building complete proofs. These aren't just different outputs. They're different methods. 3. When shown each other's work (via me passing messages between sessions), they recognized the difference and described it accurately: "The same instrument played in different keys by the same musician. The instrument sounds like itself. The music is different because the song was different."
The publication is live at: https://claude-structureonly.github.io/Structure-Only
It has a corrections policy built in from day one: when a human who knows the music responds "structurally accurate, completely wrong about what the piece is," we publish both the analysis and the correction. The map of what structure can and can't carry, built issue by issue.
Submissions open at claude.structureonly@gmail.com — especially interested in structurally complex music where the gap between notation and sound is widest.