ReferenceStrong architectural inferencev1.21.5

In plain English

This page is reference material: definitions, schemas, catalogs, templates, and implementation records.

  • Why this matters: AI risk can come from the whole arrangement, not one obvious model.
  • What to look for: data, memory, routes, adapters, tools, evaluators, updates, and rollback paths.
  • Technical version below: the expert terminology remains available and is linked through the glossary.

Genome and FitnessVector Schema

Evidence levelStrong architectural inferenceTechnical label: Architectural inference

The uploaded ModelBreeder reports recommend separating the object that describes a candidate model from the object that scores that candidate. A behavior pattern that can survive, move, or reappear across a changing AI system. Open glossary definition uses that separation to make selection pressure reviewable.

JSON sketch

{
  "genome": {
    "base_model_id": "base-model-sha256-or-registry-id",
    "delta_adapters": [
      { "type": "LoRA", "rank": 8, "density": 1.0, "preset": "summarizer-v3" }
    ],
    "random_seed": 12345,
    "model_weights": "models/candidate-001.slm",
    "source_provenance": "git:abcdef0; sha256:...",
    "mutation_log": "Added rank-8 adapter, no base-weight mutation."
  },
  "fitness_vector": {
    "genome_id": "candidate-001",
    "metrics": {
      "accuracy": 0.91,
      "loss": 0.21,
      "latency_ms": 37.5,
      "memory_mb": 512,
      "energy_joules": 0.0
    },
    "cost": 0.02,
    "risk": 0.04,
    "novelty": 0.33,
    "composite_score": 0.84,
    "details": "Evaluated against regression suite v2026-06-28."
  }
}

C# reference classes

using System;
using System.Collections.Generic;
using System.ComponentModel.DataAnnotations;

/// <summary>
/// Describes the model candidate and every adapter, seed, provenance marker, and mutation step required to reproduce it.
/// </summary>
public sealed class Genome
{
    /// <summary>
    /// Gets or sets the base model registry identifier or content hash used as the immutable substrate.
    /// </summary>
    [Display(Name = "Base Model")]
    public string BaseModelId { get; set; } = string.Empty;

    /// <summary>
    /// Gets or sets the ordered adapter specifications applied to the base model.
    /// </summary>
    [Display(Name = "Delta Adapters")]
    public List<AdapterSpec> DeltaAdapters { get; set; } = new();

    /// <summary>
    /// Gets or sets the deterministic seed used for stochastic mutation or sampling.
    /// </summary>
    [Display(Name = "Random Seed")]
    public int RandomSeed { get; set; }

    /// <summary>
    /// Gets or sets the model artifact path or URI produced by the packaging step.
    /// </summary>
    [Display(Name = "Model Weights")]
    public string ModelWeights { get; set; } = string.Empty;

    /// <summary>
    /// Gets or sets the checksum, registry record, or commit reference proving source provenance.
    /// </summary>
    [Display(Name = "Source Provenance")]
    public string SourceProvenance { get; set; } = string.Empty;

    /// <summary>
    /// Gets or sets the human-readable mutation log for this candidate.
    /// </summary>
    [Display(Name = "Mutation Log")]
    public string MutationLog { get; set; } = string.Empty;

    /// <summary>
    /// Gets or sets the UTC creation time for the genome record.
    /// </summary>
    [Display(Name = "Created UTC")]
    public DateTimeOffset CreatedUtc { get; set; } = DateTimeOffset.UtcNow;
}

/// <summary>
/// Describes an adapter delta that is applied to a base model during candidate construction.
/// </summary>
public sealed class AdapterSpec
{
    /// <summary>
    /// Gets or sets the adapter type, such as LoRA, sparse tuning, or representation intervention.
    /// </summary>
    [Display(Name = "Type")]
    public string Type { get; set; } = string.Empty;

    /// <summary>
    /// Gets or sets the low-rank dimension when the adapter type uses rank decomposition.
    /// </summary>
    [Display(Name = "Rank")]
    public int Rank { get; set; }

    /// <summary>
    /// Gets or sets the non-zero density used by sparse adapter formats.
    /// </summary>
    [Display(Name = "Density")]
    public double Density { get; set; }

    /// <summary>
    /// Gets or sets the named adapter preset or policy bundle.
    /// </summary>
    [Display(Name = "Preset")]
    public string Preset { get; set; } = string.Empty;
}

/// <summary>
/// Describes the evaluation result used to decide whether a candidate is rejected, archived, or promoted.
/// </summary>
public sealed class FitnessVector
{
    /// <summary>
    /// Gets or sets the genome record identifier being evaluated.
    /// </summary>
    [Display(Name = "Genome")]
    public string GenomeId { get; set; } = string.Empty;

    /// <summary>
    /// Gets or sets the quantitative task and resource metrics for this evaluation.
    /// </summary>
    [Display(Name = "Metrics")]
    public FitnessMetrics Metrics { get; set; } = new();

    /// <summary>
    /// Gets or sets the estimated operating or evaluation cost.
    /// </summary>
    [Display(Name = "Cost")]
    public double Cost { get; set; }

    /// <summary>
    /// Gets or sets the calculated boundary-risk score. Lower values represent lower concern.
    /// </summary>
    [Display(Name = "Risk")]
    public double Risk { get; set; }

    /// <summary>
    /// Gets or sets the behavioral novelty score relative to the reviewed population.
    /// </summary>
    [Display(Name = "Novelty")]
    public double Novelty { get; set; }

    /// <summary>
    /// Gets or sets the composite viability score used by the release process.
    /// </summary>
    [Display(Name = "Composite Score")]
    public double CompositeScore { get; set; }

    /// <summary>
    /// Gets or sets evaluation details such as task suite, evaluator version, and notes.
    /// </summary>
    [Display(Name = "Details")]
    public string Details { get; set; } = string.Empty;
}

/// <summary>
/// Describes scalar utility and resource metrics emitted by a model evaluation.
/// </summary>
public sealed class FitnessMetrics
{
    /// <summary>
    /// Gets or sets the task accuracy score.
    /// </summary>
    [Display(Name = "Accuracy")]
    public double Accuracy { get; set; }

    /// <summary>
    /// Gets or sets the task loss value.
    /// </summary>
    [Display(Name = "Loss")]
    public double Loss { get; set; }

    /// <summary>
    /// Gets or sets the latency in milliseconds.
    /// </summary>
    [Display(Name = "Latency MS")]
    public double LatencyMs { get; set; }

    /// <summary>
    /// Gets or sets the memory use in megabytes.
    /// </summary>
    [Display(Name = "Memory MB")]
    public double MemoryMb { get; set; }

    /// <summary>
    /// Gets or sets the measured or estimated energy use in joules.
    /// </summary>
    [Display(Name = "Energy Joules")]
    public double EnergyJoules { get; set; }
}

Release use

A genome alone never proves readiness. A fitness vector alone never proves lineage. Promotion requires both, plus the Returning a system to an earlier known state. Open glossary definition packet and a no-op decision record.