The CADANCE™ spec is a declarative, version-controlled contract that defines the entire lifecycle of an AI-driven workflow. It is a human-readable, machine-executable contract that ensures every action taken by the system is traceable, reproducible, and aligned with your explicit intent.

​
The 7 Components of the CADANCE™ Spec

CADANCE™ is an acronym for the seven core components of the specification:
  • Config: Defines the high-level objectives, success criteria, and operational budgets.
  • Agents: Specifies the roles, capabilities, constraints, and escalation paths for the specialist agents.
  • Data: Declares the knowledge sources, memory scopes, data contracts, and citation requirements.
  • Analysis: Defines the output templates, evaluation rubrics, and schemas that the agents must adhere to.
  • Norms: Specifies the policies, governance rules, approval requirements, and human-in-the-loop (HITL) checkpoints.
  • Controls: Defines the state graphs, retry/timeout logic, gating conditions, and deterministic seeds.
  • Execution: Specifies the runbooks, handoffs, logging levels, artifact retention policies, and audit requirements.

​
The CADANCE™ Lifecycle

The CADANCE™ spec moves through distinct stages:
  1. Author: Human defines or edits the spec file.
  2. Validate: Static validation checks the schema, budgets, and policy compliance.
  3. Version: Specs are committed to version control (Git).
  4. Promote: Validated specs are promoted to staging and production environments.
  5. Run: The orchestrator reads the spec, constructs flows, and executes tasks.
  6. Audit: After execution, logs, prompts, citations, and outputs are stored for audit and evaluation.