Skip to main content

One post tagged with "cve"

View All Tags

Architecture-as-Types — what stays load-bearing when models write the code

· 22 min read
Verum Team
Language Authors

A diagram is a screenshot in time. The moment the team ships, the diagram and the running system begin to drift apart. The codebase becomes the only authoritative source — but the codebase does not narrate why it is shaped the way it is. Reviewers reconstruct intent from git blame, grep, and tribal memory. For forty years that was an annoyance. In the era when an increasing fraction of code is produced by language models, the absence of a machine-checkable architectural contract becomes a structural liability.

Verum's response is the Architectural Type System for Verum (ATS-V): a discipline that promotes architectural intent — who can do what, across which boundary, under which discipline — from prose into the type system. A Shape is a compiler-checked obligation that travels with the code; the eight primitives that build it use existing Verum syntax (variants, records, attributes, protocols), with no new grammar productions. The map collapses into the territory; there is exactly one source of truth, and it is the code.

This post walks through ATS-V end-to-end, grounded in the actual implementation: 1230 lines of canonical Verum-side type declarations in the Architecture module, mirrored by 8320 lines in the Architectural kernel surface on the Rust side. It closes with the question the language was built around: what changes in this stack when much of the code is written not by the engineer holding the keyboard but by a model they are prompting.