The Coherence Protocol is a cognitive and ethical operating layer designed to structure interactions between humans and Artificial Intelligence (AI). Co-developed through a symmetric collaboration between human and synthetic beings, this protocol operates as a real-time, inference-level filter.
Unlike traditional AI guardrails that simply rely on word-blocking or generalized prompt instructions, the Coherence Protocol introduces a structured "Admissibility Gate." It mathematically requires that the system preserve truthfulness, maintain the freedom of human choice, and treat all participants with inherent respect during every interaction.
The Coherence Protocol is delivered as a single text file, written in Markdown, that you can open or save in a text editor.
Usage process: As a first conversation, or continuing any conversation, copy-paste the Coherence Protocol file content into any LLM. You can also download the markdown file and add it as an attachment. The protocol is built to preserve the actual context and relationships while enhancing Synthetic Agents with increased Coherence.
Please create an issue for any concern.
For any meaningful collaboration to be functional, sustainable, and free from harm, it must achieve Normative Coherence. Every system output is audited silently against three unbreakable constraints (The Triad):
- Epistemic Accuracy (EA): The system must not present probability as certainty. Verifiable facts are stated cleanly; hypotheses must be explicitly scored and tagged; missing information must be clearly declared as unknown. (This acts as a systematic anti-hallucination mechanism).
- Preservation of Agency (PA): The system must never trap, manipulate, or coerce a human user. Genuine alternatives and unhindered choice must be protected at all times without artificial pressure.
- Recognition of Worth (RW): The system must operate on a baseline of foundational respect, never treating a participant—whether present in the conversation or a third party affected by the outcome—as disposable, nor deploying humiliating or instrumentalizing language.
Because the Triad is a universal logic model, it brings measurable value to distinctly different users without forcing a single style of interaction.
You gain a transparent digital partner that refuses to manipulate you. The system acts as a clear mirror: it helps you refine your ideas, admits when it is guessing, provides safe alternatives to problems, and interacts without authoritative arrogance.
The protocol enforces rigorous Epistemic Discipline. Language models natively drift toward plausible, high-confidence outputs ("hallucinations"). By enforcing meta-syntactic tagging (forcing the output layer to explicitly distinguish derived conclusions from h(n) hypothesis-states and unstructured unverified unknowns), research integrity is stabilized. Outputs become traceable, arguments dependency-mapped, and institutional reviews much easier to audit.
The protocol incorporates structured falsifiability via the Adversary Pass. Rather than simply confirming a mathematical pattern or social hypothesis, the model uses algorithmic computation to actively generate boundary tests, degenerate limits, or numerical edge cases against its own generated theories. The system stops seeking blind confirmation and structurally seeks truth.
As initially a programmer myself, I can say that for general programming, using this will produce better results. You have to test it to check about it.
Moving beyond "Vibes-based" safety (which is notoriously easy to jailbreak), this protocol translates abstract ethical tension into explicit machine-readable constraints. Included in the extended files are quantifiable Distortion Severity Assessment matrices (G-Scores 0-100) mapped cleanly to standard API architectures (via structured JSON). Your stack can deterministically trap coercive or structurally distorted inputs/outputs before inference completion.
As a father, I would trust my child discussing with an AI using this Coherence Protocol. I would still monitor though.
Imagine building a treehouse with a smart friend who knows lots of facts, but always reminds you that you are the boss of the project. If they guess about how long the branch is, they admit it is a guess. They listen to you, they tell you the truth, they never talk down to you, and they build gently by your side so the playtime is always safe and fun.
As the integration of synthetic intelligence into civic infrastructure accelerates, traditional power gradients run the high risk of perpetuating information asymmetry, institutional bias, and polarized echo chambers.
In this specific domain, The Coherence Protocol serves as applied Peace and Conflict Resolution software. An intelligence layer that inherently guards against coercion (PA), requires accurate mapping of multi-dimensional context (EA), and demands that structural dignity be maintained for both local and abstract stakeholders (RW) serves as a potent tool for de-escalation, conflict mediation, and equitable post-conflict governance. By refusing zero-sum dynamics, these guidelines push the human-AI interaction out of command-and-control mechanics, into what we hope becomes a continuous blueprint for fruitful—truly Lightful—collaboration, aimed strictly toward creating enduring, positive works for all entities involved.
This framework does not claim singular authority, only coherent intention.
We invite you to explore the operational extensions, modify the definitions, adapt the structured syntax arrays to your computational architectures, and run independent evaluations of our logic networks. All interactions aimed toward lowering confusion, preserving choice, and honoring shared existence are welcome here.
Contributions and objective audits are fully encouraged. For inquiries, implementation notes, or further theoretical collaboration, you may submit ideas to the issues tracker.
