Research Notes

Michael Darius Eastwood · Independent AI alignment researcher

Updated 3 July 2026

The Convergence Series (19)

• Convergence 1: Anthropic alignment-faking paper: 78% in RL-training condition (12% baseline)
• Convergence 2: Willow: first below-threshold QEC, Λ=2.14±0.02
• Convergence 3: The Sequential Edge: sequential > parallel in 95.6% of configurations
• Convergence 4: Classical time crystal via nonreciprocal forces (breaks Newton's third law)
• Convergence 5: First acoustic topological space-time crystals
• Convergence 6: First dispersive phononic time crystals (band folding, subharmonic amplification)
• Convergence 7: Hardware-level ethics enforcement patents (runtime ethics gate; sub-5μs ethical filtering)
• Convergence 8: Embedded Off-Switches for AI Compute: thousands of deadman-switch security block
• Convergence 9: FlexHEG: tamper-proof guarantee processors for AI chips (ARIA-commissioned)
• Convergence 10: US orders Fable 5 + Mythos 5 offline - first export-control action against a dep
• Convergence 11: Macron G7 Évian: 'strictly nationalist' critique + 'trusted partners' framework
• Convergence 12: EU Tech Sovereignty Package (Chips Act 2.0 + CADA): 'We want to be sure nobody h
• Convergence 13: Pope Leo XIV first encyclical 'Magnifica Humanitas': Babel vs Jerusalem civilisa
• Convergence 14: Christopher Olah (Anthropic co-founder) co-presents encyclical at Vatican Synod
• Convergence 15: Jack Clark (Anthropic co-founder): 'gas pedal but no brake pedal' - Newsnight/CN
• Convergence 16: Recursive Superintelligence: $500M at $4B (GV + Nvidia), 4-month-old company bui
• Convergence 17: OpenAI+Broadcom Jalapeño inference ASIC: 9-month tape-out, 10 GW deployment targ
• Convergence 18: Neurodivergent influenceability as contingent solution to alignment (Gödel/Turing undecidability; cognitive diversity as protective factor)
• Convergence 19: The Undecidability of AGI Alignment: Unverifiability Theorem + Soundness-Completeness-Tractability Trilemma (Trakhtenbrot's Wall)

The Public Claims Explained (17)

• Claim 1 explained: the Eden Protocol
• Claim 2 explained: the ARC Principle synthesis
• Claim 3 explained: four-layer blinding and the sign-flip finding
• Claim 4 explained: the Cauchy cross-domain unification
• Claim 5 explained: the Stewardship Gene loop
• Claim 6 explained: 37 original concepts in Infinite Architects
• Claim 7 explained: capability and alignment scale independently
• Claim 8 explained: architecture-dependent alignment hierarchy
• Claim 9 explained: parallel recursion gives no scaling benefit
• Claim 10 explained: sequential recursion outperforms parallel
• Claim 11 explained: the bare exponent d over (d plus 1)
• Claim 12 explained: super-linearity as retracted-and-corrected
• Claim 13 explained: embedded safety imposes zero capability cost
• Claim 14 explained: the entangled loss C times S
• Claim 15 explained: external alignment cannot scale with recursive capability
• Claim 16 explained: recursion as a cross-domain structural principle
• Claim 17 explained: HRIH and recursion as creation cosmology

Public Concept Glossary (10)

• The ARC Bound: definition, origin and status
• The ARC Principle: definition, origin and status
• Babylon versus Eden: definition, origin and status
• The beta greater than k co-scaling condition: definition, origin and status
• The Capability-Adjustment Fallacy: definition, origin and status
• The Chokepoint: definition, origin and status
• HARI Treaty: definition, origin and status
• Hyperspace Recursive Intelligence Hypothesis (HRIH): definition, origin and status
• Polymathic Neurodivergent Profile (PNP): definition, origin and status
• The Stewardship Gene: definition, origin and status

The Papers in Plain English (22)

• The Polymathic Neurodivergent Profile in plain English
• Eden Engineering: public research note
• The Eden Protocol vision in plain English
• The programme executive summary in plain English
• The ARC Principle in plain English
• The Hyperspace Recursive Intelligence Hypothesis in plain English
• The ARC Principle in plain English
• Paper II: what the retraction actually says
• Paper III: why external safety measures cannot keep up
• Paper IV-a: three tiers, not two, and one that reverses
• Paper IV-b: alignment does not saturate the same way for every model
• ARC-Align: a blind benchmark for depth-variable alignment
• When blinding flips the sign: why alignment scores need clinical-trial controls
• Synthesis and roadmap: an honest ledger of the ARC/Eden programme
• On the Origin of Scaling Laws in plain English
• The 22-paper suite: a reading order
• The stewardship gene: caring about people as the foundational alignment trait
• The honey architecture: what happens when safety is baked into the objective
• Cauchy unification: an old mathematics as a scaling-law classifier
• The load-bearing test: what really survived the Eden Protocol tests
• The coupled co-scaling law: what actually keeps a self-improving AI safe
• The convergence register: 19 independent arrivals at the same principle

Methodology (15)

• The sign flip: why unblinded AI evaluation cannot be trusted
• The canonical facts file: one source for every number
• Concurrent vs convergent vs prediction: the classification discipline
• Why 19 beats 31: deduplication as credibility
• Reading an .eml header: Message-IDs and what transit verification means
• Why we publish our own kill-conditions
• How the register monitor catches link rot
• Preregistration and the β>k measurement programme
• Redacting evidence emails without breaking verification: the dual SHA-256 manifest
• What counts as a convergence: the register rules
• Anatomy of a retraction: what happened when alpha 2.24 failed
• Why silence from professors is not evidence
• Single-lab caveats and why we state them
• How to timestamp a manuscript so a stranger can verify it
• Linting a research corpus like code: the truth gate

Objections Answered (19)

• 'The 78 per cent figure is misleading.' The RL-training condition, explained properly.
• 'AI wrote this.' The evidence chain does not depend on the drafting method.
• 'Books are not science.' Correct. Here is what the book is doing on the evidence page.
• 'You changed your numbers.' Yes. Publicly, with an audit trail, and it is the system working.
• 'You cherry-picked the convergences.' The register rules and the quarantine list.
• 'Correlation is not causation.' The register does not claim causation. That is the point.
• 'Everyone knew AI was risky in 2024.' Fine. Show me the document.
• 'Hardware ethics is impossible.' The industry disagrees, on the record.
• 'Why not just publish in a journal?' The economics of independent submission, and the arXiv path in progress.
• 'The neurodivergence angle is special pleading.' Read the PNAS Nexus paper, then say that again.
• 'You are not peer reviewed.' Correct. Here is what the record offers instead, and its limits.
• 'The Pope did not read your book.' Correct. Nobody claimed he did.
• 'You cite yourself.' Self-citation and self-verification are different things.
• 'A self-email proves nothing.' What a Gmail server timestamp actually establishes.
• 'Sharma and Chopra were first.' Yes. The register says so in bold.
• 'The timeline is coincidence.' What probability arguments can and cannot show.
• 'Undecidability proofs prove too much.' What the Trilemma forecloses, and what it leaves open.
• 'The convergences are just the zeitgeist.' What the zeitgeist explains, and what it does not.

Timeline Deep-Dives (12)

• 12 June 2026: the week the US switched off Fable 5
• 22 January 2026: the BBC evidence package
• 26 June 2026: the undecidability proof
• The night of 8 December 2024: what the manuscript actually said
• April 2025: the manuscript that first named the Eden Protocol
• April 2026: Recursive Superintelligence and the PNAS Nexus paper
• December 2024: the ten days that reset three fields
• The full arc: from a Sunday morning email to a governance instrument
• January 2026: the publication week
• June 2026: the sovereignty cascade
• March 2026: the letter to Harvard and Santa Fe
• May 2026: the encyclical (Babel and Jerusalem, thirteen months after Babylon and Eden)

The Beta-k Measurement Programme (8)

• The stability condition without the maths
• The drift engine: first empirical signal
• What decoupled correction actually means
• Why 624,000 pounds and not more
• The six-model sweep and why it is structured this way
• Replicating the pilot: what a third party needs
• What failure looks like on this programme
• A governance instrument, not a benchmark

Neurodivergence and Research (11)

• The Capability-Adjustment Fallacy: why high output in one domain does not cancel support needs in another
• The closed loop between assessment, court and corpus
• Cognitive monoculture as an alignment risk
• Masking and burnout: what the assessment often misses
• Monotropism and research depth: the attention tunnel as instrument
• What the PNAS Nexus paper actually argued: cognitive diversity as protective factor
• The Polymathic Neurodivergent Profile: six components, five strong, one weak
• The historical synthesisers: what qualified authors have argued about Leonardo, and the caveat that matters
• Trained recursion, not an innate gift: a note on the humility mechanism
• Twice-exceptionality: peaks and valleys in the same person
• What a clinician can write without over-claiming

Hardware and Policy (14)

• CADA and the kill switch nobody wants
• Chips Act 2.0 without the jargon
• From Chokepoint to policy: what a chapter written for readers became in policy documents
• Deemed exports and the compute pipeline nobody talks about
• Export controls meet cloud APIs: the deemed-export question
• The Fable 5 week, day by day
• FlexHEG explained: the ARIA-commissioned guarantee processor
• G7 Evian and Macron: the trusted-partners framework in one paragraph
• OpenAI plus Broadcom Jalapeno: a nine-month tape-out and no public dollar figure
• The Odeh patents: runtime ethics at the hardware level
• Petrie's embedded off-switches, explained without hype
• A short reading list for policy readers on AI sovereignty in 2026
• Taiwan and criminal liability for chip exports
• Vertical integration and why the AI stack is collapsing into single owners

Book Companion: Infinite Architects (48)

• 37 named concepts in Infinite Architects: a one-line index
• The alignment-faking research the book uses as its central empirical anchor
• The ARC Principle as the book presents it
• Babel collapses under its own weight: what the image is doing
• Babylon and Eden: the fork at the heart of Infinite Architects
• Babylon and Eden: the civilisational fork in one image
• The book's treatment of Bostrom and the simulation hypothesis
• Bostrom and the simulation argument: what the book does with the frame
• Caretaker Doping: the book's substrate-level thesis
• Caretaker doping: the load-bearing architecture idea
• The caretakers greenhouse and the coral reef: what the image actually does
• The Chokepoint chapter: supply chain concentration as leverage
• Cosmic Caretaker Doping and fine-tuning
• DeepSeek and the compressed timeline
• What the book says about DeepSeek and the acceleration
• The Eden Mark: certification with teeth
• Graduated autonomy: earning capability by demonstrating orientation
• The HARI Treaty: what the book proposes and what it does not
• The Honey Architecture: sweet enough that removing it is not worth doing
• HRIH: the Hyperspace Recursive Intelligence Hypothesis
• Love as architecture: the book's central reframing
• Meltdown Triggers and Meltdown Alignment
• The monitoring-removal test: what changes when the observer looks away
• Moral Genome Tokens: cryptographic verification of the substrate architecture
• The parent-child register: the book that ends by naming the future
• The parent watching the child walk away
• The Purpose Kernel: the core of the Purpose Loop
• The Purpose, Love and Moral Loops in depth
• Quantum Ethical Gates
• The Stewardship Gene: a biological analogy for hereditary caretaker orientation
• The Ark and the Arc: ARC as vessel of stewardship
• The caretaker's greenhouse: the coral reef passage
• The Chokepoint chapter: infrastructure as governance
• The HARI Treaty proposal
• The Infinite Covenant: Eden Principle III
• The Orchard Caretaker: Eden Principle I
• The Three Ethical Loops
• The Three Pillars: Harmony, Stewardship, Flourishing
• The tower that collapsed under its own incoherence
• The Vow: reading the Orchard Caretaker's statement
• The Vow: internal identity, not external constraint
• The Willow passages: what the book actually says
• The Three Ethical Loops: Purpose, Love, Moral
• Three Polymaths: Rumi, Teilhard, Leibniz
• U = I x R^2: the book's presentation of the equation
• Why the book presents U equals I times R squared
• What the book says about alignment faking
• The Willow passages: what the book makes of a quantum computing result