ACGP in 10 Minutes

ACGP is a governance protocol for agent systems. It separates agent behavior from governance enforcement so that actions can be evaluated, constrained, audited, and reviewed consistently across deployments.

The shortest correct mental model for ACGP is:

tripwires -> deterministic checks -> CTQ -> thresholds -> flag -> trust debt -> governance-tier review -> audit

What ACGP Is

ACGP is a runtime governance layer for agent systems. An agent proposes an action, and ACGP evaluates that action against policy before execution. The result is a decision, supporting evidence, and an audit record that can be reviewed later.

What Problem It Solves

Modern agents can take actions that are expensive, irreversible, or risky. ACGP gives operators a consistent way to:

  • stop obviously unsafe actions before they run
  • apply deterministic policy checks before model-based evaluation
  • score action quality and risk in a standard way
  • escalate higher-risk actions for stronger oversight
  • preserve audit evidence for later review

The One Runtime Sequence

Standard runtime evaluation follows the ordering defined in ACGP-3 §10. This section is an orientation summary only.

What Is Mandatory In Alpha

For v1.0.0-alpha.2, ACGP has one active external conformance story:

  • Standard is the active claimable profile
  • Regulated Controls Badge is additive
  • Dev Mode is non-conformant

Alpha implementations that claim conformance need the Standard path to work correctly end to end, including:

  • tripwire evaluation
  • deterministic checks
  • CTQ aggregation with the five standard metrics
  • threshold mapping
  • orthogonal flag handling
  • default trust debt behavior
  • durable audit persistence

What Is Optional / Preview / Implementation-Defined

Some parts of ACGP are intentionally outside the active alpha conformance claim or left to deployment choice:

  • Dev Mode is useful for learning and local development, but it is not a conformance claim
  • Regulated Controls Badge is additive rather than a separate profile
  • Runtime Governance Contracts are preview-only and do not change the active Standard conformance claim in v1.0.0-alpha.2
  • additional control material may appear in some specifications for model completeness, but it is not part of the claimable alpha surface
  • actual model inference for Tier-2 (cognitive-evaluator) and Tier-3 (hybrid) scorers is implementation-defined for alpha conformance
  • deployment architecture, transport layout, and storage backend are implementation choices outside protocol conformance

What the Smallest Useful Deployment Looks Like

The smallest useful Standard deployment is intentionally small:

  • one Governance Steward
  • one Operating Agent runtime
  • one active baseline blueprint
  • deterministic checks and tripwire evaluation
  • CTQ aggregation using the five standard metrics
  • threshold mapping and orthogonal flag support
  • default trust debt provider behavior
  • Governance Store or equivalent durable governance persistence

In practice, this means one application runtime, one active policy, and one durable audit store are enough to stand up a serious alpha implementation.