This talk lays out Scuemata, a CUE-based framework for expressing data schemas that aims directly at these issues of schema evolution and backwards compatibility. It does not seek to innovate in defining schema themselves (beyond what CUE natively offers vis-a-vis other schema languages). Rather, Scuemata defines a logical meta-structure around schemas that allows for:

• All versions of a schema to be declared in a single place, easing analysis for humans and machines
• Enforcement that migrations are written when a particular schema change is backwards-incompatible
• Standard ways of expressing the risks and insufficiency of automated migrations
• Encapsulating the problem of data validation and migration to a boring, standardized process that can be isolated to process boundaries of programs
• Standard interfaces and tools for consuming any scuemata-defined schema, allowing generic tooling to be written for tasks like language translation, API generation, and more
• Standardized composition of scuemata, allowing large schema to be broken down into discrete subcomponents managed by different groups, each evolving independently
• Seamless integration into larger programs and ecosystems. (Scuemata are currently used by Grafana to express schema are complex, pluggable, and compositional, and canonical in three languages.)

Scuemata is based on a premise borne of years of experience with how most developers actually deal with versioning and compatibility in their day-to-day programming. Schema authors generally want to either maintain backwards compatibility, or else define migration paths - but absent simple structures, easy verification, and supportive tooling, doing so is often too onerous and nonstandardized to justify. By providing formalisms that encapsulate the meta-problems of schema evolution, we aim to reduce friction to the point where that calculus can shift.