# Data-Driven UX Inference
Feed an AI prototyping tool a real, structured dataset with minimal accompanying instructions, and use whatever UI it infers — filters, sort controls, grids, groupings — as a brainstorming signal rather than a spec to follow literally.
Given only a raw album dataset with almost no spec, a tool invented UX features nobody asked for: a 'follow this genre' action, an essential-albums grid, and tag-based filters. These weren't requirements — they were candidate ideas surfaced by the shape of the data itself.
As an early-stage brainstorming step before writing a full Context Engineering prompt: let the data suggest structure, then decide which inferred elements to keep, discard, or formalize into the functional-context section of the eventual full-stack prompt. Complements the Separate Data File Pattern (data.json Modularity), since the same data file used for inference can later back the finished prototype.
Adam Mosseri (Instagram) cites the platform's feed as a case study in stated preference being empirically wrong. Users consistently say they want a purely chronological feed; when Instagram tested it, chronological ranking rewarded high-frequency institutional/brand posters over individual accounts, and both usage and long-run satisfaction fell relative to a ranked feed. He generalizes the finding: nearly every contentious product debate — including privacy versus safety — involves genuine trade-offs rather than one clearly 'right' side, and stated preference alone, as opposed to measured behavior and outcomes, is an unreliable guide to which trade-off users actually prefer once they live with it.