Skip to content

Autonomy Ladder

CORE advances through defined autonomy levels. Each level adds capability while remaining constitutionally bounded.

Advancement is not declared — it is measured. A level is complete when the system demonstrates it operationally, not when the capability theoretically exists.

The Ladder

A0 — Self-Awareness ✅

CORE knows what it is and where it lives.

  • Knowledge graph of all symbols, capabilities, and relationships
  • Full structural understanding of its own codebase
  • Capability assignment for every public symbol
  • Stable identity anchors across the codebase

A1 — Self-Healing ✅

CORE fixes known structural issues automatically.

  • Detects constitutional violations via continuous audit
  • Generates remediation proposals for known violation types
  • Applies repairs autonomously within permitted lanes
  • Verifies compliance after repair before committing

A2 — Governed Generation ✅

CORE generates new code under constitutional governance.

  • Natural language intent → constitutionally aligned code
  • All generated code validated before execution
  • Self-correcting loop: Generate → Validate → Remediate → Verify
  • Full decision trace for every autonomous action
  • Human approval required for proposals outside permitted lanes

A2+ — Universal Workflow ✅

All operations follow the same self-correcting pattern.

  • Universal workflow pattern: every operation is self-correcting
  • All workflows declared in .intent/workflows/ — not in code
  • Convergence principle: resolution rate must exceed issue creation rate

A3 — Governed Autonomy ✅ ← current

CORE's daemon runs continuously, finds violations in its own codebase, proposes fixes, executes approved changes, and verifies the result — without the human writing a single line of code.

  • Seven audit sensors running continuously across 52 rules
  • Violations posted to the Blackboard and automatically claimed for remediation
  • Proposals executed via governed atomic actions
  • Unmapped violations delegated to human judgment via the Governor Inbox
  • Governor dashboard (core-admin runtime dashboard) gives five-panel situational awareness
  • Human role: define intent, review proposals requiring architectural judgment, approve constitutional changes

Operational. Daemon running. Blackboard clean.

A4 — Self-Replication 🔮

CORE writes CORE.NG from its own understanding of itself.

Conceptual. Not on active roadmap.

Convergence Principle

CORE is a feedback control system:

AI produces code → Audit detects problems → Remediators fix problems → System improves

For autonomy to be sustainable, the rate of issue resolution must exceed the rate of issue creation. If findings accumulate faster than they resolve, the system accrues governance debt and autonomous operation becomes unreliable.

Convergence toward a clean, compliant state is the fundamental operational metric.

Human Role at Each Level

Level Human Role
A0–A1 Orchestrator — directs all operations
A2–A2+ Governor — approves proposals, monitors convergence
A3 Reviewer — checks dashboard, approves architectural decisions
A4 Architect — defines intent only