Evolution Dashboard Tool

Purpose

This tool is a local teaching surface for model breeding. It does not train real models. It lets a builder adjust population size, candidate count, novelty pressure, local-device budget, and generation count, then see how a dashboard would summarize the ecology.

How to read it

A high champion score means the current best candidate is strong. A high novelty spread means the population is still exploring multiple useful strategies. A high local-fit score means small specialists are likely to repay their footprint. A strong promotion count means the ecology found candidates that deserve a release packet.

The best dashboard is not a scoreboard for one winner. It is a workbench for preserving a portfolio: champions for dependable work, specialists for narrow high-value tasks, challengers for future improvement, and archives for reusable experiments.

Pseudocode

PROCEDURE dashboard_experiment(settings)
    population <- CREATE_SEEDED_POPULATION(settings.population_size)
    history <- []
    FOR generation IN 1..settings.generations
        children <- CREATE_CANDIDATES(population, settings.candidate_count)
        scored <- COMPUTE_FITNESS_VECTOR(children, settings.novelty_pressure, settings.local_budget)
        promoted <- SELECT_BY_UTILITY_COST_AND_NOVELTY(scored)
        population <- KEEP_CHAMPIONS_SPECIALISTS_CHALLENGERS(population, promoted)
        history.ADD(SUMMARIZE(generation, population, promoted))
    END FOR
    RETURN DASHBOARD(history, population)
END PROCEDURE

Next implementation target

The educational version should evolve into a real file-backed lab where every candidate has a manifest, artifact digest, source recipe, evaluation report, and release packet. The first runtime bridge can remain browser-local and deterministic while the heavy model work is simulated.