Paper 6 — Operationalizing Reconstructive Authority

Runtime Construction, Dependency Resolution, and Execution Gating in Autonomous Agent Systems

---

Abstract

Autonomous agent systems fail not only due to incorrect decisions, but due to executing decisions whose authority no longer holds at runtime. Prior work established that atomicity alone cannot guarantee correct execution under dynamic conditions, and that behavioral drift is structurally unavoidable.

In this paper, we operationalize Reconstructive Authority (RAM) as a runtime criterion for execution: an action is permitted only if authority can be constructed from current state. This introduces a third execution state beyond admit/deny: HALT, representing undefined authority due to insufficient or inconsistent observability.

We formalize RAM as a projection over state space, prove that execution without runtime reconstruction admits inevitable failure under drift accumulation, and show that authority is a discontinuous property with boundary sensitivity to small state perturbations.

To complete the execution model, we introduce a Recovery Loop that integrates drift detection (IML) and execution control (ACP), enabling systems to transition from HALT to safe execution when observability gaps are resolved. We prove conditional liveness: if authority-defining variables become observable, execution will eventually resume.

The result is a unified execution model that guarantees no action is performed without constructible authority, while maintaining progress when safety conditions can be re-established.

---

Agent Governance Series

Paper	Title	Zenodo DOI	arXiv
P0	Atomic Decision Boundaries	10.5281/zenodo.19670649	2604.17511
P1	Agent Control Protocol (ACP)	10.5281/zenodo.19672575	2603.18829
P2	From Admission to Invariants (IML)	10.5281/zenodo.19672589	2604.17517
P3/4	Irreducible Governance Structure	10.5281/zenodo.19708496	TBD
P5	Reconstructive Authority Model (RAM)	10.5281/zenodo.19669430	2604.22898
P6	Operationalizing Reconstructive Authority	10.5281/zenodo.19699460	TBD
P7	Closing the Execution Gap (Empirical)	10.5281/zenodo.19929771	TBD

---

Key Contributions

1. HALT state — A third execution outcome ($A(t) = \bot$) distinct from ADMIT and DENY. HALT ≠ DENY: the system cannot determine whether authority holds, not that it does not hold.

2. Runtime Authority Construction Protocol — 7-step dependency-resolving algorithm with prior influence constraints, dynamic promotion rules, and 7 decision codes.

3. Recovery Loop — Formal process to exit HALT: Signal Extraction → IML Trigger → State Augmentation (with Paper 0 reversion fallback) → Reconstruction → Resolution.

4. Theorem 4 — Conditional Liveness — If all authority-defining variables become observable, the system eventually exits HALT and resumes execution.

5. Safety + Liveness — Together, Papers 5 and 6 complete both properties for the governance series.

---

Repository Structure

Paper/
├── main.tex          — LaTeX source
└── references.bib    — Bibliography (P0–P6 + related work)
README.md

---

Building the PDF

Requires a standard LaTeX distribution (TeX Live / MiKTeX):

cd Paper
pdflatex main.tex
bibtex main
pdflatex main.tex
pdflatex main.tex

---

Author

Marcelo Fernandez / TraslaIA
agentcontrolprotocol.xyz