A shared language for a multidisciplinary system
Model breeding crosses machine learning, software delivery, distributed systems, security, governance, and operations. A compact shared vocabulary prevents the same word from carrying incompatible meanings across teams.
| Reference | What it provides |
|---|---|
| Glossary | Precise definitions and distinctions for more than sixty core terms |
| Pseudocode cookbook | Reusable control-flow patterns for candidate creation, evaluation, routing, release, rollback, and retirement |
| Metrics catalog | Measurement definitions, denominators, slices, and interpretation cautions |
| Architecture patterns | A catalog of compositional and evolutionary system shapes |
| Decision guide | Rules for deciding whether to add, merge, distill, ensemble, route, compress, or do nothing |
| FAQ | Direct answers to practical and conceptual questions |
| Runtime CLI command shape | Future command vocabulary for manifests, fitness reports, novelty search, merge planning, and release packets |
Reference principles
- Define the unit. State whether a term refers to a weight artifact, adapter, service, capability package, coalition, or the whole ecology.
- Separate proposal from authority. Models may generate candidates; a protected control plane owns evaluation and release.
- Prefer behavioral contracts. Shared input/output behavior is more portable than shared internal representations.
- Record lineage. Every accepted descendant must be traceable to parents, operator, evidence, and decision.
- Measure complete paths. Model-only benchmarks are insufficient for routed, federated, or multi-stage systems.
- Keep no-op eligible. A system that must change will eventually change for the wrong reason.
Suggested use in projects
Copy the glossary terms and metric definitions that apply into the project’s architecture decision records. Copy pseudocode into design reviews before writing implementation code. Treat deviations as deliberate choices that need a written reason.
New reference material
The reference library now includes an Operator catalog that classifies model-breeding and code-breeding operators by compatibility requirements, evidence needs, and scope tier.
New positive implementation references
Use the Local AI Audience Map to connect privacy, regulation, and cognitive-liberty adoption pressure to concrete model-breeding audiences.
Use Genome and FitnessVector schemas to describe descendants and evidence, use Runtime CLI command shape to keep future tools consistent, and use the ModelBreeder/Cognivirus boundary to keep this site benefit-centered.
v2.7 practical reference additions
- Schema Reference defines Genome, FitnessVector, and ReleasePacket objects.
- Pattern Catalog maps champion/challenger, local-first inference, evidence packets, no-op, and reversible release.
- Model Ecology Glossary Tool provides a quick learning path for core terms.
- Merge Recipe Sketchpad and Adapter Stack Planner turn theory into project worksheets.
Local AI audience references
Use the local AI maps to connect model-breeding theory to actual adoption segments.
Local AI reference
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.