Systems that make decisions, and the machinery that keeps them honest.
The public repositories here are the parts we can publish: frameworks, contracts and schemas, evaluation logic, generic baselines, and methodology. What is deliberately not here is the part that constitutes an edge — tuned parameters, production feature recipes, and the operational assembly. That split is a stated policy rather than an accident, and it is why several repos read as scaffolding with the interesting constant missing.
A few things worth knowing before you read the code:
- Several repos are reference baselines rather than live services. They are published for transparency and are not deployed anywhere.
- Failing loudly is the house style. Where you see a raise rather than a fallback, or an assertion where a default would have been easier, that is usually deliberate: silent degradation is treated as a defect here.
- Much of this is built and maintained with AI agents under a written policy library governing how they work — what they may merge, what has to be measured before it is claimed, and what is reserved for a human.
Contribution guidelines and the security disclosure path are org-wide defaults
in .github.