AI-generated code often passes linting and tests while still violating team engineering standards — a phenomenon called 'slop'. The solution proposed is an invariant registry: a catalog of team-defined rules that automatically apply to every code change before it merges. Using Aviator Verify, teams can populate this registry by mining PR history, parsing contribution docs, or writing rules manually. A selector LLM matches relevant invariants to each change, flags violations with evidence, and prevents slop from shipping without adding review headcount. Invariants differ from acceptance criteria in that they are written once and apply universally, covering things like mandatory auth middleware, structured logging, and data access patterns.
Table of contents
What “Slop” Actually MeansWhy Existing Guardrails Miss ItRegistry As a Catalog of InvariantsBuilding the Registry Without Writing It by HandHow the Registry Stops Slop AutomaticallyWriting an Invariant That Holds UpWhere the Registry FitsNext StepsFrequently Asked Questions (FAQ)23 Impressions