Build a precise vocabulary before building a system
Model-breeding discussions mix mature engineering methods with new metaphors. This section separates them. The central ideas—ensembles, distillation, neural architecture search, model merging, federated learning, lineage, canary releases, and multi-objective evaluation—already exist in established forms. Model breeding, the digital Four Fs, and a teleodynamic control plane are useful ways to organize those methods, but they should be presented as a synthesis rather than a settled standard.
Foundation guides
- What model breeding means defines the umbrella term and its boundaries.
- Research maturity map separates established, emerging, conceptual, and speculative material.
- Terminology guide normalizes overloaded phrases such as code beading, code breeding, and the two Four-F systems.
- Small-model ecologies explains specialists, niches, coalitions, redundancy, and population-level capability.
- The operational 4Fs covers Fast, Flexible, Frugal, and Federated design.
- The evolutionary 4Fs covers Feed, Fork, Fight, and Flee.
- Teleodynamic control translates resource-aware self-organization into a practical two-timescale controller.
- Code breeding versus model breeding establishes the most important safety boundary in the architecture.
- Interchangeability explains why contract compatibility matters more than weight compatibility.
- Lineage and inheritance defines the minimum provenance record for every descendant.
- Use biological metaphors carefully prevents agency, consciousness, and survival language from outrunning the engineering evidence.
- No-op and metastability makes stable non-growth an explicit success state.
The minimum conceptual model
A model ecology contains a population of versioned capability packages. A router selects one model or a bounded coalition for each task. Independent evaluators measure output quality and system cost. A slower controller may create, merge, compress, or retire descendants when the expected gain exceeds the full lifecycle cost. Governance rules define what the controller may never change.
That model is sufficient to reason about adaptive AI without assuming autonomous self-preservation, unlimited self-modification, or unrestricted replication.
Continue into theory
After the vocabulary is clear, continue to Theory. The theory section adds axioms, viability mathematics, ecological fitness, resource closure, evaluator independence, and mutualist governance. Foundations define the words; theory explains how those words make system-level decisions.
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.