Classify · Confirm · Dispatch (CCD)
Scalarity's signature control topology. Brain classifies intent as a structured tool_call (the only LLM step). One dedicated confirmation turn gates the transition, human-in-the-loop or programmatic. Dispatch is 100% deterministic code with the safety gate embedded inside, no LLM in the write path. Capability grows by adding routes, not by branching code.
Classify, Confirm, Dispatch (CCD) is the Field Guide's signature control topology. The reasoning core does one job, classify the user's intent into a structured call, and then two deterministic stages take over: a confirm stage gates the action, a human or a rule approving the structured call before anything fires, and a dispatch stage executes it deterministically from the approved call, never from a fresh model inference. The safety gate lives in code, not in the model, so the agent's power scales by adding routes while approval-before-writes stays a law. In anatomy terms the model is a classifier feeding a deterministic confirm-and-dispatch spine.
The reasoning core classifies intent into a structured call; a deterministic confirm stage (human-in-the-loop or rule) gates it; a dispatch stage executes from the gated call rather than a new inference; the perimeter is the confirm stage.
- ·Production agents that must never act without approval.
- ·Regulated or high-stakes workflows in finance, healthcare and operations.
- ·Any deployment where the model proposes but a gate disposes.
- ·Safety by construction: the gate is deterministic code, not model judgment.
- ·Scales by adding routes, not by branching logic.
- ·Auditable: every action traces to an approved structured call.
- ·Separates the model's fallible classification from reliable execution.
- ·The confirm gate adds a step, and sometimes a human, to every write.
- ·Only as good as the intent classification feeding it.
- ·Requires structured, enumerable actions up front.
- ·Less suited to fully open-ended autonomy.
- Scalarity, internal research · Agent Archetypes v2
Description, use cases and trade-offs: Field Book analysis, synthesised from established agent-architecture literature.