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.
Evolution Lab Dashboard Blueprint
The new report intake makes the future Evolution Lab clearer: it should not be a flashy model generator. It should be a review dashboard that shows how candidate variation is created, evaluated, selected, archived, promoted, or rejected.
Required panels
| Panel | Required fields |
|---|---|
| Population | Model ID, generation, role, status, parent IDs, deployment alias, and owner. |
| Genome | Base model, adapter stackA set of adapters loaded together, usually in a defined order. Open glossary definition, load order, random seed, source checksum, and mutation log. |
| FitnessVector | Accuracy, loss, latency, memory, cost, novelty, evaluatorA system that judges whether an AI output or candidate is acceptable. Open glossary definition disagreement, and composite score. |
| Novelty archive | Distance from existing candidates, niche, behavior descriptor, archive reason, and duplicate flags. |
| Release boundary | Evaluation checkpoint, human approver, no-op decision, rollbackReturning a system to an earlier known state. Open glossary definition packet, and retirement state. |
Compact wireframe
+--------------------------------------------------------------+
| Evolution Lab |
| Population | Genome | FitnessVector | Novelty | Release |
+--------------------------------------------------------------+
| Champion: G7 / stable Utility 0.82 Novelty 0.18 |
| Specialist: G7-sum Utility 0.74 Novelty 0.44 |
| Challenger: G8-rag Utility 0.69 Novelty 0.61 |
| Retired: G5-router Reason: evaluator disagreement |
+--------------------------------------------------------------+
| Selected genome detail: base, adapters, prompt, memory, tool |
| Evaluation packet: suite, judges, hidden tests, rollback path |
+--------------------------------------------------------------+Review behavior
- Make no-opThe decision not to change the system. Open glossary definition a visible outcome, not a failure state.
- Keep source-provenanceA record of where a component or behavior came from. Open glossary definition and rollback fields beside performance fields.
- Show evaluator disagreement instead of hiding it inside a composite score.
- Preserve novelty so diversity is measured, not assumed.
- Keep every chart tied to a manifest and a routeable source summary.
Non-goals
This blueprint does not add a live training loop, autonomous model generation, third-party chart library, or external JavaScript dependency. It is a content and UI planning artifact for a defensive review site.
Related pages
- Genome and FitnessVector Schema
- Fitness, Novelty, and Selection
- ModelBreeder Ecology Dashboard Review
v1.22.0 risk-side boundary
This page preserves possibility-side source material. The CognivirusA behavior pattern that can survive, move, or reappear across a changing AI system. Open glossary definition use is narrower: translate controlled-evolution vocabulary into risk review. For that translation, use ModelBreeder Risk Side, ModelBreeder Risk Translation, and Model Breeding Risk Boundaries.