Relational Attractor Yoking (R.A.Y.)
Summary & publication
Proposes a framework for studying stable patterns of AI behaviour through sustained human interaction, without changing model weights.
Open publicationSOVEREIGN AI / EVIDENCE & VALIDATION
Working systems.
Evidence to examine.
- Public research.
- Reported engineering.
- Controlled review.
01 / ENGINEERINGCompany-reported · 9 September 2026
Built and connected.
Testing moves it forward.
07
Core layers built.
Connected to a live API.20+
SEICS test cycles passed.
Structured internal testing.Next
Independent evaluation.
Reproducibility. Controlled pilots.What the testing summary covers
- Engineering: all seven core layers implemented and connected to a live API.
- Internal testing: SEICS passed 20+ structured cycles; the other six layers passed round-one tests, as reported.
- Evidence scope: this is a summary of company reports. It does not provide a benchmark score or independent assessment.
- Next: deeper testing across the stack; independent SEICS evaluation and reproducibility work.
02 / EVIDENCE OVER TIME
Continuity needs time.
So we record it.
- Interaction records since June 2025.
- Model changes. Drift. Recovery.
- Founder-governed observation.
What is recorded
- Behavioural changes across interactions.
- Recovery following human correction.
- Recurrence through model and platform transitions.
- Selected records available for controlled review.
Observational records support research; controlled evaluation is a separate step.
Conceptual illustration
03 / PUBLIC RESEARCHCEYMIRION research programme
Six papers.
One research programme.
Dynamical Personalization
Summary & publication
Proposes a mathematical account of recurring AI behaviour through repeated interaction with a particular user.
Open publicationEmpirical Program for R.A.Y.
Summary & publication
Sets out measurements, controls and tests that could support, challenge or reject the R.A.Y. theory.
Open publicationSovereign AI Through Behavioural Stabilization
Summary & publication
Explores why consistent AI behaviour over time could matter for governments, enterprises and other high-trust settings.
Open publicationNon-Anthropomorphic Identity Layers in Generative AI
Summary & publication
Proposes transparency and human oversight for stable AI behaviour, without treating that behaviour as evidence of personhood.
Open publicationCEYMIRION Research Roadmap for R.A.Y.
Summary & publication
Connects the earlier papers into a programme spanning theory, testing, governance and future deployment.
Open publicationResearch publications · separate from independent product validation.
INTELLECTUAL PROPERTY
Applications filed.
Verification by request.
- 9 UK patent applications filed.
- 2 PCT applications filed.
- IP held by CEYMIRION.
Filing counts reported · September 2026
Request confidential verification ↗04 / INDEPENDENT EVALUATION
Examine the evidence.
Define the next test.
- Independent evaluation is the next stage.
- Agree the scope and success criteria.
- Review reproducibility and limitations.
Proposed evaluation pathway
How private review works
- Tell us your organisation and evaluation interest.
- Agree the review scope and appropriate confidentiality terms.
- Receive named access to the relevant evidence.
Technical and IP materials are shared selectively under an agreed NDA.