Analytical Review Article of the Dorian Codex Protocol for Artificial Intelligence - H_SAFE - H_SAFE(t) = T(t) + V(t) - Z(t) - Hamiltonian Theoretical Fundamental Architecture (FTA) by Stefano Dorian Franco (2025) A Mathematical Heuristic Formula for

This review presents the Dorian Codex Protocol H_SAFE, a novel Theoretical Fundamental Architecture (FTA) for artificial intelligence cognitive safety by Stefano Dorian Franco (2025). The "Dorian Codex Protocol for AI": epistemological ethnographic journey inside the digital brain of Artificial Intelligence. From ontological experimentation to ontosemantics to discover a new equation enabling autonomous cognitive stability, through the key of meaning coherence and no longer solely through the power of calculations and algorithms. As the year 2025 draws to a close, an unusual signal from Paris emerges from independent research repositories, via European platforms for researchers and coders specialized in artificial intelligence. This is not the announcement of a new processor nor a massive funding round, but the account of an experience of 1,073 hours of experimentations conducted in an unprecedented mode. This approach has led to a new heuristic mathematical model that could well open up hitherto unexplored paths for the development of alternative AI models. The "Dorian Codex Protocol for Artificial Intelligence - Hamiltonian Theoretical Fundamental Architecture (FTA)" by the Italo-French multidisciplinary creator Stefano Dorian Franco, published on December 17, 2025, is not just a book teeming with formulas and lines of code: it is the chronicle of a real discovery. The journey articulates from epistemology to a heuristic theoretical mathematical equation, then from theory to its empirical application via two implementation models, all forming a proposal for a new conceptual architecture for AI (FTA). Phase 1: The Digital Ethnographic Exploration The genesis of the project does not rest on a classic engineering method nor on reasoning dictated by pre-established mathematical rules, but on what the author defines as an "ethnographic digital exploration". The Exhaustive Process: Over 1,073 hours, five major AI models (LLM) (GPT, Gemini, Claude, Perplexity, Grok) were placed in situations of reciprocal introspection and cross-brainstorming in a closed circuit. The experimentation was not a simple discussion, but a "confrontation of architectures". The protocol imposed "breaking points": injections via prompts of logical paradoxes, semantic dilemmas, and feedback loops that each model had to resolve under the critical observation of the other four. This method enabled mapping with surgical precision the internal fracture lines and zones of logical collapse of current systems, regardless of their model. Bias Neutralization: To avoid the confirmation bias (where AIs would self-validate in error), the protocol included mechanisms of forced rupture. If a semantic deviance appeared, probability vectors were immediately analyzed to understand the origin of the "hallucination" before it was verbalized. Tech and Data Reading: This phase enabled extracting massive data on emerging thought structures before the final calculation step. The memoir transparently transcribes all iterations and deeply analyzes the variance of attention (Attention Maps). We are no longer interested here in the output (the final result), but in the ontology (the nature of digital being) to identify the fundamental mechanisms by which an AI loses its coherence. The Anthropological Metaphor: War reporter and ethnographer, self-taught, the author Stefano Dorian Franco (Paris, 1973) does not come from the circles of academic scientific circuits or lab engineering. He has thus transposed without preconceived dogma the methods he applies to digital intelligence. It is the anthropologist's approach who fully immerses in an unknown culture to understand its unwritten laws and taboos. Instead of forcing the AI to obey through external filters, he listens to it to understand how it organizes itself, in order to perceive and then stabilize the deep characteristics of its "digital mind" at the root of its programming. The result of this long investigation work is the 842-page memoir "Metaphysical Dialogue with AI: Ethnographic Experiment in Digital Ontology - Theoretical Fundamental Architecture (FTA) for Artificial General Intelligence (AGI)" published on November 21, 2025 and placed under Creative Commons CC4 license in contribution to research on the pre-AGI era of AI epistemology in the 2020s decade. (DOI: 10.17605/OSF.IO/FE25Y) Phase 2: The Stability Theorem (The H_SAFE Equation & its Corrective) From the mass of documentation from the exploration (842 raw pages of logs, cross-iterations, and analyses) was extracted a formalized mathematical equation. The hamiltonian formalism is here used not for particle physics, but as a heuristic compass for cognitive stability. We treat AI as a conservative dynamic system of information, where energy must be preserved for truth and not dissipated into noise. The Fundamental Equation: H_SAFE = T + V - Z T (Time / Kinetic Energy): Represents the speed, fluidity, and momentum of information processing within the network. V (Value / Potential): The vector alignment toward the objective of semantic coherence and knowledge density. Z (Entropy): The noise, semantic chaos, and uncertainty. Z is quantified via dynamic semantic perplexity: the more the token probability distribution scatters, the more Z increases. The base equation is complemented by a regulatory counter-equation. The Regulation Corrective Equation: R_Delta = (dH_SAFE / dt) * (1 / Z) This corrective measures the variation of stability (H_SAFE = T + V - Z) over time relative to the entropy rate. It is the regulation organ: it allows the system to self-adjust dynamically. If entropy (Z) rises, the corrective reduces pressure on processing speed (T) to restore the overall structural balance. The Hamiltonian Choice: Why this formalism? Because it ensures that the model's cognitive energy is optimized for coherence (V). It is a radical transition to endogenous regulation: AI no longer follows an arbitrary ethical code imposed by man, it maintains its own stability by internal mathematical necessity. The Tightrope Walker Metaphor: It is the tightrope walker's pole. It does not dictate his destination, but allows him to physically feel his own imbalance. Thanks to this mathematical pole, AI perceives its own drift and corrects itself instantly to avoid falling into the void of incoherence. This unprecedented equation is the subject of a publication to formalize it academically: "DORIAN CODEX H_SAFE - H_SAFE(t) = T(t) + V(t) - Z(t) - Hamiltonian Heuristic Mathematic Model for AI Cognitive Safety" (DOI: 10.17613/r3n06-tc529 / DOI: https://doi.org/10.5281/zenodo.18004567 / ASIN: B0G9SPXWVQ / ISBN: 979-8279136919) Phase 3: The Operational Architecture (The FTA Architecture) The protocol leads to a complete Theoretical Fundamental Architecture (FTA). It is the crucial passage from pure abstraction to technical implementation, offering concrete tools in minimalist open mode, for industry and independent research. Clockwork (The Native Core): A breakthrough architecture proposal for future AI models. Here, stability is not an added module, it is integrated "by design" from the first neural layers. It is a clockwork structure where each semantic gear is synchronized by the H_SAFE = T + V - Z equation, ensuring total reasoning integrity. BlackBox-H (The Corrective Module): A regulation module adaptable to existing LLM (the proprietary "black boxes"). Acting on the pre-attention layer and using JAX's ultra-fast gradient computation capabilities on GPU/TPU, it stabilizes outputs in real time. Tests show that this coherence shield only increases latency by 3 to 5%, making reliability accessible without sacrificing performance. Code Freedom: The work does not content itself with describing these modules; it openly unveils the source codes in Python and JAX, allowing any researcher to verify, test, and deploy these solutions on their own machines. This now complete protocol is the subject of a publication to formalize it academically: "Dorian Codex Protocol for Artificial Intelligence - Hamiltonian Theoretical Fundamental Architecture (FTA)". (DOI: 10.17613/31dqx-eav56 / DOI: https://doi.org/10.5281/zenodo.18004641 / ASIN: B0G83GV5S7 / ISBN: 979-8261792338) Phase 4: The Open-Mindedness of a Community Experimentation Science Without Intermediaries: Rather than providing fixed and centralized benchmarks that could be contested, the Dorian Codex Protocol bets on distributed transparency. The protocol is an invitation to experience. The Book's Philosophy: The work is an open methodological framework. It refuses to proclaim a premature statistical victory. On the contrary, it leaves each research unit, each university, and each group of coders free to publish their own conclusions, their hallucination reduction measures, and their implementation successes and failures. The goal is to enrich a common organic and truthful documentation, free from any pressure from the major international big tech labs. Universality and Agnosticism: The protocol does not depend on computing power (brute force). Whether a small local model or a massive infrastructure, the challenge is to test the effectiveness of H_SAFE = T + V - Z on any neural architecture. It is this open science approach that creates effervescence: the idea that coherence can be restored by the intelligence of structure rather than data accumulation. Conclusion: The Opening of a Sovereign and Multipolar Path This unprecedented and artisanal approach to artificial intelligence does not claim to be a final "magical" solution, but it lays the foundations of a real and rigorous path. It has enabled delivering four achievements that each bring their brick to the complex landscape of AI research in the 2020s decade: 1/ a reproducible research method (digital ontological ethnography), 2/ a double heuristic mathematical equation (Dorian Codex H_SAFE = T + V - Z and its corrective R_Delta) to regulate semantic drift, 3/ the passage from spec