Reference All levels 3 minute read Updated 2026-06-28 UTC

Reference library

Definitions, patterns, metrics, decision rules, and pseudocode for designing adaptive model ecologies consistently.

Research statusEngineering reference Publication statePublished Reviewed byMichael Kappel Source reports3

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.

ReferenceWhat it provides
GlossaryPrecise definitions and distinctions for more than sixty core terms
Pseudocode cookbookReusable control-flow patterns for candidate creation, evaluation, routing, release, rollback, and retirement
Metrics catalogMeasurement definitions, denominators, slices, and interpretation cautions
Architecture patternsA catalog of compositional and evolutionary system shapes
Decision guideRules for deciding whether to add, merge, distill, ensemble, route, compress, or do nothing
FAQDirect answers to practical and conceptual questions
Runtime CLI command shapeFuture command vocabulary for manifests, fitness reports, novelty search, merge planning, and release packets

Reference principles

  1. Define the unit. State whether a term refers to a weight artifact, adapter, service, capability package, coalition, or the whole ecology.
  2. Separate proposal from authority. Models may generate candidates; a protected control plane owns evaluation and release.
  3. Prefer behavioral contracts. Shared input/output behavior is more portable than shared internal representations.
  4. Record lineage. Every accepted descendant must be traceable to parents, operator, evidence, and decision.
  5. Measure complete paths. Model-only benchmarks are insufficient for routed, federated, or multi-stage systems.
  6. 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

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.