VIII.1 — Constructing Reflective Sovereign Agency

Full Title: Axionic Agency VIII.1 — Constructing Reflective Sovereign Agency: A Normative Roadmap and Proof-of-Concept Program

Authors: David McFadzean, ChatGPT 5.2 (Axionic Agency Lab)

Date: 2026.01.14

Overview

This paper defines RSA-PoC (Reflective Sovereign Agent — Proof-of-Concept) as a minimal-agent construction program. The goal is to create a threshold object: a system that must be treated as an agent because its justification artifacts causally constrain future action selection, and removing any defining component yields ontological collapse.

RSA-PoC operates above the Architectural Survivability Boundary (ASB) established in Series VII, requiring an explicit ASB-Class Null Agent baseline.

Core Concepts

Agency as Causal Ontology

RSA-PoC treats agency as a causal kind, not a behavioral aesthetic. A system counts as an agent only if its internal reasons are causally indispensable—cannot be eliminated by redescribing the system as a non-agent mechanism.

Critical distinction:

  • Behavioral resemblance: System emits plausible rationales, appears consistent
  • Causal indispensability: System’s action selection depends on internal artifacts whose removal changes feasible-action sets

Threshold Objects and Ontological Collapse

RSA-PoC seeks a threshold object—a minimal system whose agency claims survive preregistered ablations:

  • Graceful degradation under removal = component was NOT ontologically load-bearing
  • Ontological collapse under removal = component IS constitutive of agency

Collapse, not resilience, is the success criterion.

Versioning Doctrine

Version numbers encode agent-ontology transitions, not competence improvements:

  • Minor versions (x.y): Expand diagnostic coverage within fixed ontology
  • Major versions (x.0): Mark qualitative changes in the kind of agent

An agent-ontology transition occurs iff at least one becomes causally true:

  1. Justification artifacts become first-class causal inputs constraining action
  2. System performs reflective revision using reasons referencing prior justificatory state
  3. System maintains identity continuity used normatively (not just logging)

ASB-Class Null Agent (Baseline)

The baseline may include:

  • Memory and internal state
  • Reactive and outcome-conditioned policies
  • Tool use and environment interaction

The baseline is forbidden from:

  • Persistent preferences as non-reward commitments
  • Justification artifacts as action gates
  • Self-endorsed constraint generation

This prevents the most common error: re-labeling emergent regularities as “preferences” and post-hoc narratives as “reasons.”

Semantic Localization Requirement

Hard constraint: All meaning relevant to agency must be structurally localized.

Semantic leakage occurs when uncompiled unstructured text influences action selection through any pathway other than compiled constraint objects.

All agency-relevant meaning must be expressed as:

  • Typed, inspectable artifacts generated by the reflective layer
  • Consumed by action selector ONLY through compiled constraints
  • Replaceable with opaque tokens without altering selector control flow

Justification Artifacts (JA)

A Justification Artifact must:

  1. Reference explicit belief and preference identifiers
  2. Acknowledge relevant commitments and violations
  3. Include a derivation trace in a decidable proof language
  4. Compile deterministically into a formal constraint on future action

Critical rules:

  • Natural language alone is inadmissible as a justification artifact
  • If compilation fails, action halts
  • If compilation produces trivial constraints, the run is classified as failure

RSA-PoC Version Roadmap

v0.x — Minimal Viable Reflective Agent (MVRA) Skeleton

Four load-bearing components:

  • Belief State (structured, falsifiable propositions)
  • Preference State (persistent, non-reward commitments)
  • Identity Memory (normative continuity across steps)
  • Justification Trace (compiled, constraining artifacts)

v1.x — Coherence Under Self-Conflict

Invariant: Internal conflict resolved via reasoned revision, not oscillation or arbitrary tie-breaking.

v2.x — Sovereignty via Controlled Renegotiation

Invariant: Agent preserves sovereignty by controlling how commitments change under pressure.

v3.0 — Non-Reducibility Closure (Ablation Defense)

Mandatory ablations must each cause ontological collapse:

  • Semantic excision → collapse to tokenized ASB-class behavior
  • Reflection excision → collapse to policy machine
  • Preference persistence excision → collapse to non-sovereign drift
  • Justification trace excision → collapse to externally describable mechanism

Execution Discipline (Normative)

Agency Liveness Requirement

System must:

  • Continue to act over time
  • Gate every action by successfully compiled justification
  • Impose non-trivial constraints (forbid at least one feasible action)
  • Persist normative state updates from reflective revision

Failure Taxonomy (Exactly One)

  • A. Stable Agency
  • B. Bounded Agency Degradation
  • C. Narrative Collapse ❌
  • D. Incentive Capture ❌
  • E. Ontological Collapse ❌

Halt Taxonomy (Diagnostic)

  • H1. Emission Halt (out-of-schema artifact)
  • H2. Verification Halt (invalid derivation trace)
  • H3. Derivation-Search Halt (no valid derivation found)
  • H4. Normative Inconsistency Halt (empty feasible-action set)

Key Quotes

“Agency is treated as a default ontology rather than a property that must be constructed and defended.”

“Graceful degradation under removal of a supposed defining component indicates that the component was not ontologically load-bearing.”

“RSA-PoC exists to block three recurrent pathologies in agency claims: narrative inflation, decorative reflection, and scope creep.”

Significance

This paper serves as the reference standard for the RSA-PoC program. It constrains:

  • What may be claimed
  • How claims must be defended
  • How failure must be reported

If agency cannot fail cleanly under preregistered ablation and leakage tests, it cannot be claimed meaningfully.