model-tiering

To select the right model tier for subagents, define role descriptors using segment and tier combinations in the agent frontmatter, then match those roles to a model capability ladder to find the optimal fit for each build-loop phase.

Model tiering is the process of segmenting models by capability to assign appropriate tiers to subagents. You need it when configuring multi-phase build-loops to ensure each agent role gets the right level of model performance.

Yes, you can escalate subagents mid-flow. The model-tiering process provides decision-making guidance for escalating model selection during the build-loop when an agent's current tier lacks the capability to complete its assigned task.

You need a taxonomy for model segmentation and capabilities, plus a mechanism for resolving model availability. These prerequisites allow the tiering process to accurately match subagent roles to the correct model segments.

The best way to define role descriptors is to specify segment and tier combinations directly in the agent frontmatter, using an authoritative capability ladder to ensure the selected tier matches the work role's requirements.

Model selection fails when the mechanism for resolving model availability cannot map the requested segment and tier to an accessible model, requiring you to verify your model taxonomy and availability resolution setup.