Review the repository. Check the agent’s report. Understand the tools behind the verdict. Barg Labs makes software-delivery evidence easier to inspect and use.
Independent studio · Vancouver and London
Cejel checks the repository. Dunstan checks the agent’s report. Repository Evidence Review helps you interpret repository evidence for a named decision. Instrument Review measures the tools that judge code.
Make repository evidence inspectable.
Cejel examines engineering signals in a repository and produces a reproducible report and trust certificate. Give reviewers evidence they can inspect alongside their own checks.
Check what the agent says it delivered.
Coding agents report what they did: files changed, tests passed, issues closed. Dunstan checks each claim the agent declares against the repository and CI record before you merge.
One repository. One named decision.
Accepting a software delivery or reviewing a repository? We run Cejel with you in your environment, review what the evidence supports and where it stops, and deliver a two-page decision memo.
Your AI judge, measured by someone else.
People accept what your scanner, code reviewer or AI judge says without seeing inside it. We test it on cases where the right answer is already known.
A delivery to accept, a completion claim to check, or a tool to evaluate.
Engineering leaders
Use repository evidence and declared-claim checks to inform the decision to accept a delivery.
Explore an evidence reviewTeams using coding agents
Compare declared files, tests and references with the repository and CI record through Dunstan.
See DunstanTool and platform teams
Commission a bounded Instrument Review, or explore how software-delivery evidence could fit your workflow.
Discuss an evaluationOur published studies show what was tested, how it was measured and where the conclusions stop.
A known-answer study of a commercial AI judge. Read the frozen method, results and limits before relying on its verdicts.
An open-model study using the same cases, with the context-window limitation stated alongside the results.
A study of constructed credential-transfer instructions across agent skills. Methods, results and the study’s limits are published.
These constructed studies do not estimate real-world error rates.
Engineering knowledge connected to supporting evidence, explicit authority and present applicability, preserved across repository and agent changes.
Keep the source and its limits attached to the knowledge it supports.
Make clear who authorised a decision and what that authority covers.
Distinguish a historical lesson from evidence that still applies now.
Preserve context as repositories, tools and agents change.
This is our longer-term direction. Today’s tools provide focused repository and claim checks; broader knowledge continuity and authority workflows remain in development.
Cejel runs where the code lives: no network, no model call, nothing sent to us.
Cejel reports and Dunstan records can be reproduced. Studies state their inputs, method and limits.
A missing source, an unverified claim and an observed contradiction mean different things. The record should say which.
We write down what we expect before we run, and publish when we are wrong.
Barg Labs is led by Houman Azimi-Nejadi, a software engineer and founder with more than 25 years of experience building complex software systems across multiple domains.
The company works closely with expert collaborators across product, engineering, and web systems, and is based in Vancouver and London.
Barg Labs is a member of NVIDIA Inception, NVIDIA’s program for startups, and of Claude for Startups, Anthropic’s program for startups building with Claude.
Discuss a repository review, an independent tool study or a potential integration. We’ll agree what can be checked, what evidence is needed and what remains outside scope.
Talk to Barg Labs
