Apex review playbook

Review sequence

1. Use Composition Explosion Calculator to estimate the state space.
2. Use Assurance DecayProposed Cognivirus terminology for the loss of confidence in an evaluation result as system components, routes, permissions, models, prompts, memory, tools, or environments change. Open glossary definition Worksheet to identify what changed since evaluation.
3. Use Behavioral Persistence Explorer to inspect memory, descendants, synthetic data, routers, and evaluatorA system that judges whether an AI output or candidate is acceptable. Open glossary definition reservoirs.
4. Use Ecological RollbackRestoring not only a model artifact but the relevant router, prompts, memory state, tool permissions, evaluator version, deployment alias, and data dependencies. Open glossary definition Planner to inventory rollbackReturning a system to an earlier known state. Open glossary definition dependencies.
5. Use Evaluator Independence Checklist to check whether selection pressure is trustworthy.
6. Use Responsibility DiffusionThe inability to identify one accountable component, developer, operator, or decision point after a distributed system produces harm. Open glossary definition Mapper to identify missing accountable owners.
7. Use Multi-LoRA Apex Threat EnvelopeA proposed Cognivirus review category for systems where adapter reproduction, dynamic composition, persistent memory, adaptive routing, evaluator selection, and incomplete rollback reinforce one another. Open glossary definition to decide whether ordinary component review is inadequate.

Apex review triggers

Run the playbook when any of these are true:

• adapters are generated, merged, or selected automatically;
• a router changes which safety policy serves an input;
• memory persists across model replacement;
• evaluator versions changed since the last approval;
• synthetic data from prior outputs enters training or evaluation;
• rollback does not include aliases, prompts, memory, and permissions;
• human approval is mostly based on model-generated summaries.

Output

The output should be a decision record with one of four outcomes:

Reminder

The worksheets do not calculate real-world probability. They identify missing evidence, stale assuranceConfidence, backed by evidence, that a system meets safety or governance requirements. Open glossary definition, and unexamined transition paths.