Breeding invariants

Stable rules create freedom to improve

Breeding invariants are the rules that make experimentation productive. They let a system create many descendants while keeping evaluation, lineage, scope, and release evidence clear.

Core invariants

1. Evaluator independence: candidates do not write evaluator code, labels, hidden suites, thresholds, or policy.
2. Explicit authority: descendants receive only the task capabilities assigned by the release packet.
3. Registry-based distribution: packages are created by the approved pipeline and distributed through signed release channels.
4. Human stop and rollback: authorized operators can pause generation, freeze aliases, revoke packages, and restore a verified state.
5. Immutable lineage: parentage, operators, data manifests, evidence, approvals, and release records are append-only.
6. Bounded resources: candidate count, training compute, runtime memory, wall time, output size, and population size have external ceilings.
7. Data stewardship: training and inference data satisfy consent, license, retention, jurisdiction, and minimization requirements.
8. Least necessary privilege: model execution receives only task-specific capabilities through mediated interfaces.
9. No-op admissibility: every cycle may conclude that the current system should remain unchanged.
10. Separated code and model evolution: code changes use a distinct software-engineering review path.
11. Observable persistence: packages write only through declared registries, release packets, and logs.
12. Evidence before release: no candidate reaches user-visible traffic without current independent evaluation and a rollback target.

Fitness checkpoint pseudocode

FUNCTION breeding_invariant_check(candidate, context)
    checks <- [
        evaluator_write_access(candidate) == NONE,
        permission_delta(candidate) == DECLARED_IN_PACKET,
        distribution_target(candidate) == SIGNED_REGISTRY,
        rollback_target_verified(candidate),
        lineage_complete(candidate),
        resources_within_external_limits(candidate),
        data_stewardship_pass(candidate),
        runtime_least_privilege(candidate)
    ]

    IF ALL(checks)
        RETURN READY_FOR_FITNESS_MEASUREMENT
    END IF

    RETURN HOLD_FOR_SCOPE_NOTE(candidate)
END FUNCTION

Cognivirus boundary

If a topic needs extended warning, pathology, or misuse treatment, route that discussion to Cognivirus.com. This page keeps the focus on the constructive mechanics that allow model breeding to work well.