Fitness Vectors for Useful Descendants

A practical multidimensional scoring object for comparing model descendants across utility, latency, memory, energy, privacy, novelty, maintainability, and human benefit.

Research statusSchema-backed implementation pattern Publication statePublished Reviewed byMichael Kappel Source reports8

Why a vector beats one score

A useful descendant may win on latency but not raw accuracy, or it may preserve local privacy while using a little more memory. A single global score hides those tradeoffs. A FitnessVector keeps the tradeoffs visible.

pseudocode

STRUCT FitnessVector
    candidate_id
    utility
    calibration
    robustness
    latency
    memory
    energy
    privacy
    novelty
    maintainability
    human_benefit
    lineage_completeness
    evidence_uri
    evaluated_at_utc
END STRUCT

Sample scorecard

Dimension	Example score	Why it helps
Utility	0.82	The descendant improves the target task.
Human benefit	0.90	It makes review, learning, or decision-making easier.
Latency	0.76	It responds fast enough for the target workflow.
Memory	0.70	It fits the target local runtime.
Local privacy	1.00	It can run without moving private data.
Novelty	0.64	It covers a useful niche, not just a duplicate.
Maintainability	0.81	The artifact is easy to inspect and operate.
Lineage completeness	0.96	Parentage, operator, and evidence are clear.
Decision	keep as specialist	The ecology gains a useful narrow capability.

Selection pattern

pseudocode

FUNCTION select_descendant(candidate, champion, threshold)
    vector <- MEASURE_FITNESS(candidate)
    delta <- COMPARE(vector, champion.fitness_vector)

    IF delta.utility > threshold.utility AND delta.human_benefit > threshold.human_benefit
        RETURN KEEP_AS_CHAMPION_OR_SPECIALIST(candidate, vector)
    END IF

    IF delta.novelty > threshold.novelty AND vector.maintainability >= threshold.maintainability
        RETURN KEEP_AS_CHALLENGER(candidate, vector)
    END IF

    RETURN ARCHIVE_AS_LEARNING(candidate, vector)
END FUNCTION

Fitness vectors make the ecology easier to discuss because each decision explains what improved, what stayed flat, and what should be measured next.

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.

Implementation directivesModelBreeder v2.7.0 Positive Exploration DirectiveCurrent operating directive · 35.8 KB Architecture and project resourcesModelBreeder Architecture and ProjectsSource report · 53.0 KB Architecture and project resourcesModelBreeder Architecture and ResourcesSource report · 50.9 KB Architecture and project resourcesModelBreeder Architecture and Resources Part 2Source report · 33.5 KB Evolution lab and dashboardsExploring Ecology Dashboard ArchitectureSource report · 64.7 KB Architecture and model mergingModel Breeder Architecture Deep DiveSource report · 76.7 KB Architecture and runtime explorationModelBreeder Architecture ExplorationSource report · 48.9 KB Theory and source alignmentArchitectural and Theoretical Analysis of ModelBreederMixed maturity · 63.4 KB

Why a vector beats one score

Sample scorecard

Selection pattern

Source reports used for this guide

Related guides

Reference Architecture for Governed Model Ecologies

Local Model Innovation Stack

Viability controller

Lineage DAGs Make Capability Reusable