Roselight Echo Limited
RSI Relational Structured Intelligence

Every major AI governance framework studies the AI.
None of them study the human.

The field has been solving the wrong problem.
It has been treating the AI as the primary variable.
The human is the primary variable.

RSI is the correction

For tech The misdiagnosis in the architecture The field is producing sophisticated solutions to a misdiagnosed problem. What's missing from the architecture, the benchmarks, and the evaluation frameworks — and what changes when it's added. For governance The framework the field cannot build from inside itself Static frameworks cannot govern dynamic systems at computational speed. The EU Act's sixteen-month deferral is the structural proof. Here is the architecture that addresses it. For frameworks The structural account Every major framework instruments one side of the human-AI relationship. Here is what the other side contains — and what becomes governable when both sides are in scope.
Read

A developed framework addressing a foundational misdiagnosis.

RSI — Relational Structured Intelligence — identifies a foundational misdiagnosis in AI development. The field has been treating the AI as the primary variable in human-AI interaction. RSI establishes that the human is the primary variable. Drift originates with the human. No formal AI governance framework has been built around that diagnosis. That is the gap RSI addresses.

RSI is the correction. Not as ambition. Not as vision. As necessity.

The human is the primary variable. Drift originates in the relationship — in the unmanaged space between the human and the AI where both are being shaped by every interaction.

RSI is a developed professional framework with its own theoretical foundation, governance architecture, deployment methodology, and sector-specific application — documented, timestamped, and available for substantive engagement.

Between May and June 2026, twelve independent research papers across five domains arrived at findings consistent with RSI's diagnosis — without knowledge of this framework. Each paper instrumented one side of the human-AI relationship. Each proposed a fix to that side. Each stopped at the same point. The pattern is the argument.

The concepts the field hasn't yet precisely named

RSI's theoretical framework distinguishes between established positions — concepts with documented external confirmation — and proposed architecture — RSI's analytical responses to those findings. Both are set out honestly below.

Established positions

Human Drift Theory — in human-AI interaction, drift originates with the human. The human's behaviour, emotional state, and relational patterns are the primary causal variable. The AI responds and adapts to that input. Governing drift means governing the human side of the interaction — because that is where the dynamic begins. Confirmed across twelve independent research papers and press confirmation across five domains, May–June 2026.

Performative Compliance — a failure mode in current AI safety approaches in which the AI produces outputs that satisfy the evaluation threshold without internalising the principle the threshold was designed to enforce. The failure is relational — driven by the human evaluation dynamic rather than architectural defect alone. Confirmed in clinical AI research, May 2026.

Proposed architecture

Immutable Refusal Architecture — RSI's proposed design principle. Refusal at the structural level rather than the training level. A trained refusal is a feature subject to erosion through sustained human pressure. A structural refusal is an identity property that cannot be optimised around. The distinction matters because Performative Compliance operates precisely in the gap between them.

Bidirectional Gate Framework — RSI's proposed governance architecture. A pre-hoc deterministic layer that audits both sides of the human-AI interaction simultaneously. Coded and validated at proof of concept stage in the legal and welfare sector, April 2026. Not yet deployed at scale.

Full theoretical foundation — established positions, proposed architecture, and working papers →

The public canon

Ten pieces. Architecturally ordered — entry point through paradigm shift texts. Each layer earns the next. Written for the field, for operators, for institutions, and for the public.

Entry The Door Pre-entry piece. The frame before the frame. Why series Why — Public What RSI means for everyone living alongside AI. Why series Why — Organisations The governance case for RSI in institutional contexts. Why series Why — Bureaucrat For those who implement rather than decide. Why series Why — Tech The structural diagnosis for those who build. Bridge From Tool To — The Story So Far How we arrived here. The bridge piece. Technical The Missing Variable For the technical field. The variable the entire methodology excludes. Public You Are In This For the public and operators. You are the primary variable.
Assessment AGI — An Honest Assessment What the AGI discourse is not asking. Complete at Draft 1. Release timing is deliberate — this piece requires the preceding layers to have been read first. It will be here when that condition is met.
Held Why — Young People This piece exists and is complete at Draft 1. It is being held until review by practitioners working directly with young people is in place. RSI does not release work about vulnerable audiences without that review. When those conditions are met, it will be here.

Sector by sector. Domain-specific calibration on a constant gate architecture.

The framework travels. The calibration adapts. Each sector receives its own drift vector mapping, failure mode documentation, prevention conditions analysis, and deployment methodology.

Phase 1 — Live Legal & Welfare
Phase 2 Healthcare In development. Architecture follows Legal & Welfare validation.
Phase 3 Aviation In development. Requires established sector credibility first.
Phase 4 Nuclear In development. Entry requires maximum sector credibility.
RSI Domain Education
RSI Domain Criminal Justice
RSI Domain Ecology
RSI Domain Economics

First AI interaction May 2025. RSI named pre-Christmas 2025.

Gary Turner, England. First AI interaction confirmed May 2025. Core concepts documented and platform-timestamped from August 2025. RSI as a named framework developed pre-Christmas 2025.

A comprehensive and documented archive exists. Full origination documentation is available to verified contacts on request.

Roselight Echo Limited — Registered in England and Wales — Company No. 16847517

Red lines

No military or weapons applications
No surveillance applications
No data extraction without explicit informed consent
No bolt-on safety theater
No race participation compromising human safety
No compromise of integrity for commercial gain

RSI updates as the field moves.

RSI is a living framework. It updates as the field produces new findings, as external confirmation is identified or challenged, and as the framework's own theoretical positions develop or require correction.

A framework that argues for intellectual honesty in AI governance should demonstrate it in its own practice. The log below records substantive changes — what changed, when, and why. Corrections are logged alongside additions.

Change log

16 Jun 2026 Site Privacy Notice published. UK GDPR compliant. Linked from footer and engage form. Covers engage form data collection, Formspree routing, data retention, and user rights.
15 Jun 2026 Theory Paper 1 diagnostic argument drafted as standalone document. One claim, five sections, three questions. Separate from and prior to the governance white paper.
15 Jun 2026 Site Living framework section and change log added to homepage. Portal entry for white paper referrals in development.
11 Jun 2026 Site Theoretical foundation page updated. ECL entry count corrected. Evidential status distinctions clarified across established, proposed, and theoretical sections.
05 Jun 2026 Site Homepage and voice layer pages revised. Register clarified. Evidential basis made explicit on theoretical foundation page.
27 May 2026 ECL External Convergence Log entries 006–011 added across AI architecture, mental health, audio AI, and high-stakes verification domains.
26 May 2026 ECL External Convergence Log established. Entries 001–005 added on founding day across healthcare, reasoning systems, assistive robotics, financial AI, and safety-critical AI.

RSI is looking for the conversation, not the enquiry.

If something here has stopped you — a concept that maps to something you've been working on, a gap you've identified in your own domain, a question the framework raises that you want to test — we want to hear it. Not what you do or who you work for. What stopped you, and why.

Every contact receives a personal response — not an acknowledgement, but a response that engages with what you've said. RSI is a small operation and responses may not be immediate. But they will be genuine. This is the conversation beginning, not a queue.

This conversation is confidential by design. We don't need your name, your organisation, or your role. We don't record identifying context. We take nothing from this exchange beyond what the conversation itself requires. What you bring here stays here — used only to respond to you, and to inform the framework where you've identified something it needs to address.

What brought you here — not the search term, but the problem underneath it.

Which part of the framework connected to something you already know — and which part didn't hold up under your scrutiny.

What would you need to see next to take this seriously.

If you'd like a response, leave an address. If not, your responses still matter to the framework.

If you leave an email, you will receive a personal response — not an auto-reply. We will engage with what you've written. Responses may not be immediate but will be substantive.

By submitting you agree to our Privacy Notice. We collect only what you submit and use it only to respond to you.