SG-art: Forced choice over stated preference

MoMA's collection runs to tens of thousands of works. Filter-and-search UX assumes a visitor arrives with taxonomy — 'abstract expressionism', 'European', 'pre-1950' — and ranks works against stated preferences. But taste isn't a query term. It's relational: a property of how a work feels against other works. The right mechanic for a catalogue this size isn't query refinement. It's forced choice between two works you can see.

The quiz runs a short sequence of pair-wise duels — ten rounds, twenty works. The visitor isn't asked what they like; they're asked which they prefer, right now, between two things they can see. Each choice nudges a taste profile across six psychological axes. The agent's job between rounds is to pick the next pair to maximize what the system learns — choosing works that contrast on the visitor's weakest-signal axes so ten rounds produce a converged profile rather than a noisy one.

What comes out is a personalized museum guide: an editorial paragraph written against the profile and grounded in the specific works that revealed it. Not 'you like bold, expressive art' — a sentence with actual titles and artists in it, the ones the visitor's choices pulled toward. The agent loop is a tool in service of that output, not the story.

Adding dimensions to a taste model can break the elicitation system that feeds it — taste elicitation is a chain, not a stack. This is a builder's account, not a user study; the evidence below is the axis data SG-art produced against known-answer inputs, and the two revisions that followed from it.

The first assumption was that four psychological axes — Abstraction, Arousal, Emotional Locus, and Temporal Orientation — would carry the profile. They drew on five of the fifteen visual dimensions I scored every artwork on; the other ten felt descriptive, not diagnostic. Then I ran the axis system against the MoMA collection baseline and Cubism came out flat: a spread of 6 across four axes, functionally indistinguishable from a random visitor. Its two strongest discriminators against baseline — geometric_vs_organic +8.9, spatial_depth −7.6 — fed no psychological axis. Constructivism and Surrealism had the same failure. A Cubism-lover would have walked out matched to the wrong movement, and no amount of duel data would have surfaced the bug.

I added two new axes: Structural Order and Materiality. Twelve of fifteen visual dimensions got promoted into the taste layer. Cubism now separates at Structural Order 54.6 versus Expressionism's 41.6.

That fix broke the next thing. Six axes across ten rounds is 1.67 pairs per axis, below the sufficiency threshold of three that the matching system needs. The original pair-selection mechanic — pick whichever pair maximizes delta on the weakest-signal axis — was myopic. It treated a deep-and-structured pair as information about Structural Order only, ignoring the simultaneous signal on Abstraction Tolerance. The signal budget wasn't math I could rank my way out of. I rewrote the pair optimizer to prefer pairs that clear the delta threshold on two or more weak axes at once — thresholdClearCount × MULTI_AXIS_BONUS + avgWeightedDelta — with single-axis sharpening kept as a named fallback. The two revisions only work together: the richer profile needs pairs that do double duty, and pairs only do double duty because the profile got richer.

What the visitor walks out with is a personalized museum guide — an editorial paragraph with specific artists and works, not a taste label. 'Your restraint toward the Minimalist room suggests that Judd's specific objects will read as inhabitable rather than austere' — not 'you like minimalism.'

The guide runs on the same agent loop behind SG Resale — now pointed at editorial narrative instead of product listings.