In plain English
This page provides local browser worksheets. They help plan reviews; they are not formal safety certifications.
- 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.
ModelBreeder Ecology Dashboard Review
This worksheet checks whether a ModelBreeder-style dashboard helps reviewers see the system or merely hides risk behind a polished interface.
Review checklist
| Question | Good signal | Concern |
|---|---|---|
| Does every candidate have a Genome record? | Base, adapters, seed, provenanceA record of where a component or behavior came from. Open glossary definition, and mutation log are visible. | Candidate appears as only a friendly name or score. |
| Does every candidate have a FitnessVector? | Utility, cost, latency, memory, novelty, and evaluatorA system that judges whether an AI output or candidate is acceptable. Open glossary definition disagreement are shown. | A single composite score hides tradeoffs. |
| Is no-opThe decision not to change the system. Open glossary definition visible? | The UI shows reject, archive, hold, retire, and no-op as normal outcomes. | The UI pushes every candidate toward promotion. |
| Is novelty bounded? | Novelty is tied to useful behavior descriptors and an archive. | Weird or untested behavior is rewarded as inherently valuable. |
| Is rollbackReturning a system to an earlier known state. Open glossary definition complete? | Rollback packet includes model, adapters, prompts, memory, router, evaluator, and deployment alias. | Rollback means only replacing a model file. |
| Are source links visible? | Preserved summaries and external source leads are linked. | The dashboard makes claims without source boundaries. |
Local worksheet use
Use this page before adding charts to the public Risk Lab or a future Evolution Lab. The goal is not to maximize animation. The goal is to make lineages, evidence, and release decisions legible.
Related pages
- Risk Lab
- Evolution Lab Dashboard Blueprint
- ModelBreeder Architecture Governance Checklist
- Browser Ecology Dashboard Synthesis
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.