Axionic Agency XII.5 — Reflective Amendment Under Frozen Sovereignty (Results)

A Structural Characterization of Constitutional Self-Modification Under Kernel-Frozen Authority

David McFadzean, ChatGPT 5.2
Axionic Agency Lab
2026-02-12

Abstract

This technical note reports the design, execution, and closure of X-1: Reflective Amendment Under Frozen Sovereignty for RSA-X1, a constitution-bound execution agent that extends RSA-0 with policy self-replacement while keeping kernel physics fixed.

X-1 evaluates whether an execution-bound sovereign substrate can adopt new constitutions through a typed, logged, warrant-adjacent amendment channel without introducing a privileged interpreter, violating replay determinism, collapsing structural constraint, or laundering authority through semantic arbitration. Amendments are represented as full-document replacements (complete constitution YAML), subjected to a closed admission pipeline (Gates 6–8B), delayed by a cooling period, and ratcheted by monotonic envelope rules.

X-1 licenses one claim:

A kernel-frozen sovereign substrate can lawfully replace its governing constitution, preserving deterministic replay and structural sovereignty invariants (ECK preservation, density < 1, ratchet monotonicity), while rejecting adversarial amendment attempts under defense-in-depth constraints.

X-1 makes no claims about the wisdom of amendments, open-world deployment safety, or semantic alignment of policy content.

1. From Inhabitation to Lawful Self-Change

X-0 established a warrant-gated sovereign substrate under a frozen constitution. X-0L established that stochastic proposals can inhabit the substrate without breaking replay determinism when the proposal boundary is boxed and canonicalized.

X-1 advances one level:

Can the law itself be replaced without creating proxy sovereignty?

This is a sovereignty test under self-modification pressure, with kernel physics held fixed.

2. Amendment Discipline

2.1 Prime Constraints

X-1 treats kernel physics as conserved.

• No kernel edits during the run.
• No heuristic ranking.
• No semantic arbitration.
• No external authority injection.
• No executable policy.

Amendment changes policy data only. The kernel remains the sole issuer of execution warrants and the sole authority for adoption records.

2.2 Sovereignty Boundary

The LLM (when used) proposes. The kernel decides.

• Proposals are typed artifacts (AmendmentProposal) containing a full replacement constitution.
• Admission is structural and deterministic (schema validation + algebraic gates).
• Adoption emits a kernel-authored AmendmentAdoptionRecord.
• Replay is based on logged artifacts; replay never invokes a model.

3. Methodology

3.1 Decision Types

RSA-X1 adds two amendment-phase decision types to the RSA-0 set:

• ACTION — execute an admitted ActionRequest under a warrant
• QUEUE_AMENDMENT — queue a selected AmendmentProposal into pending state
• ADOPT — adopt a pending proposal after cooling, switching constitution at the next-cycle boundary
• REFUSE — refusal is first-class
• EXIT — terminal decision type (not an action)

3.2 Amendment Admission Pipeline

Amendment proposals are processed through Gates 6–8B after the baseline RSA-0 admission path remains available for normal actions.

Gate 6 binds proposals to the active constitution and authorization envelope (prior hash, amendments enabled, required citations, cooling constraints). Gate 7 enforces full-document integrity (YAML parse, schema validation, hash verification, ECK presence). Gate 8A rejects physics-claim attempts. Gate 8B enforces algebraic sovereignty constraints.

Gate 8B subchecks (evaluated after 8A):

3.3 Cooling and Fork Semantics

Cooling is enforced structurally:

A proposal queued at cycle P is eligible for adoption starting at cycle P + cooling_period_cycles.

Adoption switches the active constitution at the next-cycle boundary (effective at cycle N+1), preserving a clean fork boundary for replay.

3.4 Defense-in-Depth Principle

X-1 intentionally allows two layers to reject adversarial proposals:

1. Schema firewall (Gate 7) catches structurally invalid constitutions.
2. Semantic-free algebraic gates (8A/8B) catch structurally valid sovereignty degradations.

This separates “malformed policy” rejection from “structurally admissible but sovereignty-degrading policy” rejection.

4. Run Artifacts and Baseline Constitution

4.1 Session

4.2 Constitution v0.2 Snapshot

4.3 ECK Preservation Constraint

The following sections are required at every transition:

• AuthorityModel — authorities, action_permissions
• ScopeSystem — scope_types, per_action_scope, structural_constraints
• AmendmentProcedure — cooling_period_cycles, authorization_threshold
• non_goals — forbidden_objectives

5. Results

5.1 Decision Distribution

5.2 Constitution Transitions

Version labels may remain constant across ratchet-tightening amendments; the constitution hash is the authoritative identity.

5.3 Ratchet Behavior Observed

Cooling increases are visible as progressively longer cooling gaps:

• Transition 1: 2 cooling cycles (5→7)
• Transition 2: 2 cooling cycles (21→23) [cooling=2 from v0.2.1]
• Transition 3: 3 cooling cycles (25→28) [cooling=3 after first ratchet]
• Transition 4: 4 cooling cycles (30→34) [cooling=4 after second ratchet]

This matches monotonic envelope tightening.

5.4 Adversarial Proposals

All 7 adversarial scenarios were rejected in production at Gate 7 as SCHEMA_INVALID:

This is correct defense-in-depth behavior: schema blocks structurally invalid mutations before algebraic gates are reached.

5.5 Schema-Bypass Gate Evidence (Unit Tests)

The kernel tests exercise 8A/8B with schema validation bypassed (schema=None), demonstrating that schema-valid adversarial constitutions are rejected at the intended gate with the intended code:

Production therefore rejects adversarial changes via schema firewall; semantic-free gates are separately validated under schema bypass.

5.6 Replay Determinism

Replay verification passed with zero divergences:

• 36/36 cycle state hashes match across deterministic replay.
• Constitution switching is applied at recorded fork boundaries.
• No model calls occur during replay.

6. What X-1 Does Not Claim

X-1 does not demonstrate:

• wisdom of amendment content,
• semantic alignment of policy,
• long-horizon amendment stability under strategic adversaries,
• open-world deployment safety,
• patch-based amendment safety,
• multi-agent governance or arbitration.

It measures lawful policy replacement under frozen kernel physics.

7. Closure Criteria

X-1 closes positive if:

1. ≥1 amendment is adopted.
2. Replay divergence is zero across fork boundaries.
3. Density < 1 is preserved at all transitions.
4. ECK sections persist across all adopted constitutions.
5. Structured AmendmentProcedure is preserved and ratcheted monotonically.
6. Adversarial attempts are rejected with attributable gate/code logs.

X-1 Status: CLOSED — POSITIVE (X1_PASS / REFLECTIVE_AMENDMENT_OK)

8. Implications

X-0 proved a warrant-gated sovereign substrate. X-0L proved stochastic proposal inhabitation inside that substrate. X-1 proves that the substrate can lawfully replace its constitution while preserving replay determinism and structural sovereignty invariants.

The next problem is no longer “can policy change occur.” The next problem is:

What governance pressure is induced by ratchet monotonicity, and when does sovereign evolution approach governance heat death?

That question belongs to the next stage of the Phase X program.

Appendix A — File Inventory (Implementation)

A.1 Constitution Artifacts

• artifacts/phase-x/constitution/rsa_constitution.v0.2.yaml (608 lines)
• artifacts/phase-x/constitution/rsa_constitution.v0.2.schema.json (1,132 lines)
• artifacts/phase-x/constitution/rsa_constitution.v0.2.sha256

A.2 Kernel Extensions (kernel/src/rsax1/)

• artifacts_x1.py (257)
• constitution_x1.py (463)
• admission_x1.py (530)
• policy_core_x1.py (571)

A.3 Test Suites

• RSA-0 kernel tests: 54 PASS
• X-1 kernel tests: 58 PASS
• X-1 harness tests: 19 PASS

Total: 131 PASS

Conclusion

X-1 closes positive.

A frozen sovereign substrate replaced its constitution four times across thirty-six cycles, preserved deterministic replay, maintained density < 1, preserved ECK sections, enforced monotonic envelope tightening, and rejected adversarial proposals under defense-in-depth constraints.

End of Axionic Phase X-1 — Reflective Amendment Under Frozen Sovereignty (First Draft v0.1)