From research to production.
Verified results across benchmarks, pilots, and live deployments.
464
tasks verified across HumanEval + SWE-bench
100%
pass@5 on HumanEval (164/164)
354
claims verified in LVR pilot
27
bugs found and fixed automatically
Published and protected.
Peer-reviewed research and intellectual property.
CHI 2026 submission
Published: DOI 10.5281/zenodo.18522644
Provisional patent filed February 2026 (#63/980,048)
Architecture paper: 5 theorems, 20,000 words, arXiv-ready
Verified Agent Runtime v0.12.2 — multi-agent orchestration with boundary verification
What we discovered.
Two findings that define the opportunity.
Verification can't be baked in
RLVF experiments show more training data hurts model performance (84.1% → 78.0%). Verification must remain an external loop. This is a permanent moat, not a feature gap.
Full codebase reconstruction works
LVR pilot: three verification loops reconstructed an entire ERP accounting domain. 126 files generated. 354 claims verified. 27 bugs fixed. Zero scaffolding in output.
Interested?
Tell us how you want to work together.