ReferenceStrong architectural inferencev1.22.1

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.

Apex Risk Control Matrix

Evidence levelStrong architectural inferenceTechnical label: Architectural inference

The matrix links report-derived mechanisms to controls. It is not a certification method.

schematic · apex controls matrix

Controls must attach to transitions, not just artifacts.

Every major apex mechanism has a corresponding transition-level control. None is sufficient alone.

Adapter reproduction signed registry · rate limits Composition expression manifest · path-aware eval Selection pressure independent judges · no-op Persistence reservoirs rollback packet · extinction review Yardstick drift evaluator versioning · evidence store Human incentives exit rights · dissent channels
MechanismRiskControl
A small add-on that changes or specializes model behavior. Open glossary definition reproductionLow-cost behavioral cloningsigned adapter registry, rate limits, source review
Behavior that becomes visible only when a specific collection of components is loaded, routed, or invoked together. Open glossary definitionUnsafe state appears only in stackcomposition manifests, path-aware evaluation
Selection pressureMetrics preserve shortcutsindependent evaluators, The decision not to change the system. Open glossary definition, proxy-risk review
Persistence reservoirsBehavior survives retirementRestoring not only a model artifact but the relevant router, prompts, memory state, tool permissions, evaluator version, deployment alias, and data dependencies. Open glossary definition, extinction review
Yardstick driftRuler changes with candidateA system that judges whether an AI output or candidate is acceptable. Open glossary definition versioning, independent evidence store
Human incentive captureUsers preserve unsafe behaviorexit rights, dissent channels, operator rotation
Local guardrail bypassCandidate can reach its own controlsexecution-time authorization boundary

Use

Use this as a planning matrix for architecture reviews. Each row asks whether the system has a meaningful control at that transition.