Mission
ModelBreeder.com is an implementation-oriented learning system for designing adaptive AI from governed populations of specialized models and deterministic components. It translates a broad report corpus into definitions, architectures, experiments, operational controls, safety boundaries, blueprints, decision tools, and pseudocode.
Project commitments
- preserve every supplied source report in the distribution;
- distinguish established methods from conceptual and speculative material;
- favor measurable, reversible engineering over anthropomorphic claims;
- keep model generation separate from evaluation and release authority;
- make no-op, rollback, retirement, human override, and user exit first-class;
- remain deployable as plain PHP without a database or third-party runtime service.
Documentation
Use Contact and About Michael Kappel for the public creator contact route. Use Site evidence and discovery for public discovery files, route inventory, canonical answers, support boundaries, metadata, and release evidence. Use the editorial method to understand how reports were synthesized. Use the research-status policy to interpret maturity labels. Use the site-architecture page to deploy or extend the codebase. Use the source-preservation page to verify that the archive remains intact. Use AI project memory and handoff to understand the repository-local .uai records, active file intake, docs-as-source memory routing, and validation workflow.
Scope
The site teaches architecture and governance. It does not ship a training framework, model weights, autonomous agent, or production control plane. Pseudocode is deliberately implementation-neutral and should be adapted to the target organization’s security, legal, data, and operational requirements.
Source reports used for this guide
These reports are preserved verbatim in the site archive. The guide above is an editorial synthesis and may narrow, qualify, or reorganize claims from the source material.