Blog

From the team

ProductAugust 2026

Your Benchmark Is Grading the Wrong Thing

Golden sets measure agreement with the findings reviewers recorded, not an exhaustive inventory of a codebase's defects. Introducing Deep Audit (beta): evidence-gated review where every Bug ships with a citation.

TechnicalMay 2026

There Is No Best Code Reviewer

Research shows specialized multi-agent AI code review outperforms single-pass approaches by 40%+. No single model dominates across all bug types, languages, or codebases.

PerspectiveMay 2026

AI Writes Code Fast. Who Checks If It's Good?

AI shifted the bottleneck from writing code to knowing whether it's good. The data on AI code quality is sobering — and it points to a missing layer.

BenchmarkMay 2026

Martian Benchmark Testing

Measured tier results on the Martian-50 code-review benchmark, with methodology notes and official-submission status.