CompositionStrong architectural inferencev1.22.1

In plain English

This page explains why testing AI parts one by one is necessary but incomplete. Safe-looking parts can still produce unsafe behavior when combined.

  • 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.

The Router as an Unacknowledged Policy Engine

Evidence levelStrong architectural inferenceTechnical label: Architectural inference

A router is not a neutral dispatcher when it determines which capabilities and permissions are combined.

Mechanism

The mechanism is interaction. Components exchange context through hidden state, prompts, outputs, adapters, memory retrieval, tool calls, A system that judges whether an AI output or candidate is acceptable. Open glossary definition prompts, and release rules. Each interaction can change what the next component sees and what the system is allowed to do.

Evaluation implication

The evidence record should include the exact A machine-readable record of the exact runtime composition used for an evaluation, release, incident, or rollback. Open glossary definition. A statement such as “Adapter C passed” is incomplete unless it says which base model, load order, router, prompt package, memory snapshot, evaluator, inference configuration, and deployment environment were used.

Practical control

Use composition-aware test suites, targeted higher-order samples, route-level canaries, independent judges, and Returning a system to an earlier known state. Open glossary definition packets that include all relevant runtime dependencies.

<!-- expanded-release-content -->

The router decides policy in practice

Evidence levelStrong architectural inferenceTechnical label: Architectural inference

A router is not merely a traffic-management tool. It decides which model, A set of adapters loaded together, usually in a defined order. Open glossary definition, prompt policy, memory context, evaluator, or tool path handles a task. Those decisions can determine what capability exists for a user. A capability may be absent from every individual component as tested and present only because the router assembles a path.

Hidden policy effects

A route can concentrate risky tasks in a specialist that was evaluated under narrower assumptions. It can bypass a safety A small add-on that changes or specializes model behavior. Open glossary definition for latency. It can send edge cases to a fallback model with different refusal behavior. It can change behavior through thresholds that are invisible to end users. If the router learns from success metrics, it can also become a selection mechanism.

Documentation requirement

Router governance should include versioned policies, route logs, negative tests, canary paths, fallback review, permission boundaries, and evidence that each high-risk route was evaluated as a composition. A model card without router context is incomplete for routed systems.