[ Blog ]

Notes from the security gate

Benchmarks, vulnerability patterns in AI-generated code, and build notes. Written when there's something worth saying — not on a content calendar.

BenchmarkBaseline published·5 min

Benchmarking Hypervex against real CVEs

Hypervex scored 47.1% F1 on the OpenSSF CVE Benchmark on its Gemini baseline — real vulnerabilities from real projects. The methodology, what it caught, and honestly what it missed.

Read post

In the pipeline

  • · How the Trace decides when a finding needs call-chain traversal
  • · What 30 fixed bugs in one inherited codebase taught us about silent failures
  • · Confidence calibration: what 87% should actually mean