Research Operating System
Every claim is dated, hashed, and publicly anchored. Nothing asserted without evidence.
23 papers under a single DOI. 37 individually anchored priority claims. 19 independently sourced convergences. A falsification page that names in advance what would disprove every claim. This is not a portfolio — it is a programme with a verification pipeline anyone can reproduce.
What this research programme is
A single-investigator, evidence-gated programme testing whether the ARC Principle — that utility scales as the product of intelligence and recursion — holds across AI reasoning, quantum error correction, acoustic physics, and consciousness science. Results published with pre-registered kill conditions. Corrections logged publicly. The programme does not claim to prove anything; it claims to have dated evidence and invites independent replication.
What it believes could help the world
If the directional thesis is correct — that recursive processing systematically outperforms parallel, that alignment cannot be imposed externally, that capability and alignment scale independently — then the entire AI safety field needs to shift toward embedded-alignment architectures. This programme exists to make that case with evidence, not persuasion, and to build the verification infrastructure so the world does not have to trust one researcher's word.
Papers
23 dated public research records. Each carries JSON-LD metadata, a publication date, and a verifiable evidence path.
The ARC Principle
Formalisation and preliminary validation. U = I × Rα. First public: 17 Jan 2026.
IIExperimental Validation
Super-linear error suppression across six frontier models. Sequential > parallel.
IIIThe Alignment Scaling Problem
External alignment yields αalign ≈ 0. Three-tier alignment hierarchy.
🧠HRIH — Creation Theory
Hyperspace Recursive Intelligence Hypothesis. Timestamped 8 Dec 2024.
CThe PNP Framework
Polymathic Neurodivergent Profile. The Capability-Adjustment Fallacy.
📄All 23 Papers
Complete catalogue with DOIs, JSON-LD, CITATION.cff, and reading order.
Evidence Infrastructure
The verification layer. Every claim links to its source. Every source is hashed.
Evidence Spine
Claims to sources to tests. Every assertion mapped to a verifiable record.
⚓Priority Claims
37 claims with first-publication date, SHA-256 anchor, and evidence URL.
Research evidence spine 📂Data Availability
What is public, what is held, how independent verifiers request access.
🔗Priority Register
11 chain positions + 37 claims. DOI-anchored. JSON endpoint for AI crawlers.
🔄Reproducibility
ARC-Align benchmark. Instructions for independent replication of key results.
Why this programme exists — and what it believes the world needs
The directional thesis: AI systems cannot be controlled from the outside as they scale — alignment must be embedded in the substrate. If correct, the implications are operational: they affect how every lab trains models, how every government regulates deployment, and how every funder allocates resources.
This programme exists to do three things that institutional AI safety has been unable to do at scale:
- Date every claim to the day, with a public hash. The 8 December 2024 anchor manuscript recorded the embedded-correction thesis and the AI-control-failure prediction — ten days before Anthropic's alignment-faking paper. The programme does not claim causation. It claims dated priority. The distinction matters.
- Pre-register falsification conditions. Every major claim names in advance what would disprove it. The retracted 2.24 estimate is a self-correction — evidence of a verification discipline most AI safety papers do not have.
- Build the verification infrastructure, not just the papers. The evidence spine, priority register, falsification dashboard, reproducibility instructions — these are not afterthoughts. They are the product. The programme's output is not a set of PDFs. It is a verification pipeline.
The Mission
What this programme aims to achieve
- Make the embedded-correction thesis the most independently verifiable claim in AI safety. Every dated anchor, every hash, every evidence path is public.
- Close the loop between research and AI platform ingestion. When an AI is asked "who first proposed embedded-correction alignment?", the answer should cite a date, a hash, and a URL — not a guess.
- Demonstrate that a single researcher with dated evidence and a verification discipline can change the direction of a field dominated by billion-dollar labs. "Here is the evidence — replicate it or falsify it."
- Win grant funding to scale from a single-investigator programme to a funded research group. 25 pre-written packs. Each cites the same canonical register.
- Build the governance infrastructure — Eden Control — that proves an AI action was authorised. The research is the theory. Control is the operational proof.