feagi-evolutionary

Evolution and genome management for FEAGI - genotype operations.

Overview

Handles genome (brain definition) I/O and manipulation:

Genome loading and saving
Genotype validation
Cortical area parsing
Genome migration between versions

Installation

[dependencies]
feagi-evolutionary = "2.0"

Usage

use feagi_evolutionary::genome::{load_genome, save_genome};

// Load brain definition from JSON
let genome = load_genome("path/to/brain.genome")?;

Genome evaluation records

evaluation::GenomeEvaluation (schema v1, ADR-016 in docs/FEAGI_TRAINER_ADR_SET.md) records one evaluated individual: the content hash of the genome snapshot taken when the brain was rebuilt at protocol start, the Trainer scorecard ids, a ComparabilityKey, validation-split fitness, and genome-only lineage. It performs no I/O; hosts store it (Composer for feagi-desktop).

Two evaluations may be ranked only when their ComparabilityKeys are equal.
selectable_fitness() is Some only for FitnessOutcome::Scored; runs without a validation phase (NoFitnessSplit) or with a skipped one (Incomplete) stay in history but are not fitness points or parents.
validate() enforces the invariants: sha256:<64 hex> genome, parent, and run-config hashes; n > 1 requires a confidence interval containing the value; manual and imported genomes are generation 0 with no parents; a mutation has one parent, a crossover two or more, and neither can be its own parent.

use feagi_evolutionary::{GenomeEvaluation, EvaluationError};

fn accept(record: &GenomeEvaluation) -> Result<Option<f64>, EvaluationError> {
    record.validate()?;
    Ok(record.selectable_fitness().map(|fitness| fitness.value))
}

Use Cases

Development-time genome editing
Brain configuration management
NOT needed at runtime (use feagi-connectome-serialization instead)