A deterministic scanner with 85 rules for security, accessibility, and
quality problems in AI-generated code and writing. A labeled test
corpus tracks missed issues and false positives.
Scanner and test corpus
A web UI review team that is itself evaluated: labeled interface
fixtures test whether its findings are accurate and specific. I used it
on this site.
Review evaluation harness
Seven agents that design and review SwiftUI. Limerino’s interface work
goes through it, and its reviewer must reach 0.8 recall and precision
on planted defects.
SwiftUI review harness