Agent architectural compliance

Question

Can machine-readable architectural constraints outperform natural-language instructions?

Background

This experiment will compare compliance outcomes across two guidance modes: natural-language and machine-readable constraints.

Success means more architecture-consistent changes with less reviewer correction effort.

The objective is to find a governance model that scales with increasing agent capability.