Carlos Arleo — Constitutional Physics

Deployed & Empirically Validated

The Constitutional Physics architecture is not theoretical. It has been built, deployed, and stress-tested across three operational systems before any SDK abstraction began.

319

adversarial sessions
AURA-ECHO corpus

constitutional collapse
across all sessions

1.0025

CV at critical seam
predicted, then confirmed

26×

P_work spike at
ρ = 0.862 Write-Lock

100,009

papers processed
DNI metascience engine

28:1

targeted vs random
repair likelihood ratio

Three deployed systems. Three completely different domains — generative music, metascience, civic governance. The same constitutional enforcement architecture. No architectural modification required across domains.

The Physics: Three Substrate-Independent Predictions

These equations are not engineering choices. They are predictions that emerged independently from three separate theoretical frameworks and were subsequently confirmed in deployed systems.

ρ = L / C

Causal saturation ratio. Tracks adversarial load in real time. As the η-pump injects load without increasing capacity, ρ → 1.0. Write-Lock triggered at ρ = 0.862. Confirmed across 319 sessions.

CV = 1.0025

Self-organised criticality at the constitutional boundary. Predicted theoretically by Prager (2026) §9.7 before the corpus was analysed. Confirmed to four decimal places. Substrate-independent — a consequence of Landauer's principle, not of silicon.

f(x) = 0
if x ∉ C

Constitutional violation = zero fitness = ontological death. The discontinuous fitness landscape: no gradient to follow out of constitutional space. The boundary is a cliff, not a slope. Confirmed: 100% mortality → evolutionary rescue in 4.9s. 28:1 targeted repair likelihood ratio.

Full derivation: docs/physics-grounding.md

Constitutional Agent SDK

An open-source (MIT/Apache 2.0) infrastructure toolkit that wraps any standard LLM inside a thermodynamic Viability Kernel. Submitted to ARIA Scaling Trust Track 2 (£600k, March 2026).

constitutional-compiler

Translates natural language security policies into deterministic, machine-executable graph constraints via Frame-Based Principled Reasoning (FBPR). Binds Phenotype (human intent) to Genotype (TypeScript hard constraints). Policy violation becomes structurally impossible rather than statistically unlikely.

rho-monitor

Tracks causal saturation ρ = L/C in real time. Detects the η-pump — adversarial load injection without capacity increase — and triggers homeostatic repair before constitutional violations occur. Tiered computation: expensive evaluation triggered only when epistemic integrity is under threat.

constitutional-halt

Safe transfer of authority when the Viability Kernel cannot be restored. Refuses action outside constitutional mandate, executes Terminal Gait, transfers control to human operator, emits complete deterministic audit log. The architectural opposite of Silent Failure.

→ github.com/CarlosArleo/constitutional-agent-sdk

The Trinity Stack

Three independent researchers working in separate disciplines arrived at structurally identical predictions for system stability under adversarial load. The convergence was not coordinated. It was discovered.

Framework	Author	Discipline	Core Prediction
Aitiopoietic Cognition	Veloz (2023–24)	Theoretical Biology	Thermodynamic coupling required for organisational closure
Grammar of Stability	Prager (2026)	Physics / Info Theory	CV ≈ 1.0 at ρ ≈ 1.0 before Write-Lock
Constitutional Physics	Arleo (2025)	AI Architecture	f(x) = 0 if x ∉ C — survival law not training target

Publications & Preprints

Constitutional Activation Under Adversarial Load: Empirical Observations from a 319-Session Corpus of the AURA-ECHO Constitutional System

Arleo, C. & Veloz, T. · Zenodo Preprint · March 2026 · Reviewed by Matthew Prager

doi.org/10.5281/zenodo.17692900

Constitutional Physics: Causal Saturation and Aitiopoietic Governance in Generative AI Systems

Arleo, C. · Zenodo · November 2025

doi.org/10.5281/zenodo.17692900 1,419 views · 1,161 downloads

Grammar of Stability

Prager, M. · Zenodo · 2026

doi.org/10.5281/zenodo.18444557

Aitiopoietic Cognition: Organisational Closure and Causal Autonomy in Adaptive Systems

Veloz, T. · CLEA-VUB Working Paper · 2023

Project Status

Component	Status	Notes
AURA-ECHO architecture	Deployed	319 sessions validated
DNI (100k papers)	Deployed	99.9% consistency confirmed
WFF governance system	Deployed	Great Filter experiment complete
Constitutional Compiler (SDK)	In development	FBPR validated · abstraction in progress
ρ-Monitor (SDK)	In development	Validated in AURA-ECHO + DNI
Constitutional Halt API (SDK)	In development	Anti-jailbreak validated
TypeScript / Python library	Planned	MIT / Apache 2.0 · pending ARIA funding
ARIA Scaling Trust (Track 2)	Submitted	£600k · 18 months · March 2026
ARIA Scaling Trust (Track 3.4)	Submitted	£300k · 18 months · March 2026