Figures as Interfaces: Toward LLM-Native Artifacts for Scientific Discovery

This project identifies a critical friction point in scientific discovery: the loss of semantic data when plots are rendered to pixels. While the conceptual framework is sophisticated, it faces extreme 'frontier risk' from the rapid evolution of LLM interfaces. Anthropic's 'Artifacts' and OpenAI's 'Canvas' are already moving toward this interactive, code-backed visual paradigm. With 0 stars and 7 forks, this is currently a localized research artifact (likely from a specific lab or paper) rather than a community-driven project. Its defensibility is low because the moat is purely intellectual; any platform with a code-execution sandbox can implement 'figures-as-interfaces' by simply mandating Vega-Lite or Bokeh outputs. The project's value lies in its formalization of how agents should 'reason' about these objects, but it is likely to be absorbed as a feature of larger AI scientific assistants like those being built by FutureHouse or the frontier labs themselves.