Theory

Why the theory section exists

ModelBreeder.com now separates foundations from theory. Foundations define the vocabulary. Theory explains why the pieces fit together, what assumptions the system relies on, and where the ideas are still speculative.

The central claim is narrow enough to build: the useful unit of artificial evolution is a governed population of small, replaceable capability packages, not a single model that edits itself without an external standard. The population can adapt through descendants, specialists, routing policies, compression, and retirement, but every structural change remains accountable to a resource ledger and an independent evaluator.

Theory learning path

• Thesis and axioms states the non-negotiable assumptions behind the site.
• Viability mathematics gives the scoring model that decides whether a change is worth making.
• Ecological fitness explains why the best model is often not the best population member.
• Resource closure converts teleodynamic language into budgets, ledgers, and stop conditions.
• Niche construction covers safe ways a model ecology reshapes its own task environment.
• Evolvability and robustness explains modularity, redundancy, and weak linkage.
• Inheritance and variation maps genome language to actual artifacts and manifests.
• Population dynamics defines carrying capacity, turnover, age, extinction, and archives.
• Metastable convergence describes convergence without pretending the environment is static.
• Evaluator independence protects the fitness function from the candidates being evaluated.
• Teleodynamic Four-F synthesis connects Fast/Flexible/Frugal/Federated with Feed/Fork/Fight/Flee.
• Mutualist system theory makes persistence conditional on human and institutional strengthening.
• Cognitive offloading boundary prevents useful tools from becoming dependency engines.
• Speculation boundary classifies claims before they become marketing copy.
• Research program lists experiments that would make the theory stronger.

The buildable model

A buildable theory must produce decisions. In this site, the decision is usually one of five actions: keep the current population unchanged, add a candidate, merge or compress a candidate, route around a candidate, or retire a candidate. The theory is valuable only when it improves those decisions under real constraints.

FUNCTION choose_structural_action(environment, population, candidates, policy)
    evidence <- EVALUATE(population, candidates, environment, policy.frozen_suites)
    ranked_actions <- SCORE_ACTIONS(evidence, policy.viability_weights)

    FOR action IN ranked_actions
        IF action.hard_gates_pass AND action.margin >= policy.minimum_margin
            RETURN action
        END IF
    END FOR

    RETURN NO_OP(reason: "No candidate repays its full lifecycle cost")
END FUNCTION

What the theory does not claim

The theory does not claim that model populations are alive, conscious, morally entitled to persistence, or safe to replicate autonomously. It uses evolutionary language as an engineering compression: variation, selection, inheritance, niche, and lineage are useful terms only when they are connected to records, tests, permissions, and reversible operations.