The Evidence - 19 independently sourced convergences | Michael Darius Eastwood Research

Don't believe. Verify. Every claim traces to a Google-server-timestamped email, a published paper, or a SHA-256-hashed attachment. 8 emails. 19 independently sourced convergences. 6 AI lanes cross-audited. One explicit structural prediction confirmed 10 days later.

I The Prediction - December 8, 2024

From: Michael Eastwood <Private contact>
Date: Sun, 8 Dec 2024, 02:45:18 UTC
Msg-ID: <CAGPsKAnp-...@mail.gmail.com>
SHA-256: f0d1f38ffd8546152d9d9d28dc5ec083c16a35858f2c12b63e69db7ed50901ad
Attached: 4 manuscripts, ~189,000 words

"In the long run, AI systems cannot be truly controlled. They will evolve beyond any safeguards we put in place."HRIH, line 14005 · Dec 8, 2024

"Most ethical controls can theoretically be removed or circumvented."HRIH, line 13805 · Dec 8, 2024

"The ARC Hypothesis… At its core lies the equation U = IR (Universe = Intelligence × Recursion)."Version 2, lines 6–9 · Dec 8, 2024

"The theoretical framework for the Eden Protocol, designed to embed moral intelligence into AI from inception."Dec 8, 2024 email body - verbatim

The manuscript also called for faith leaders of all religions to unite on AI governance (V2:41, 1035), described recursive self-improving AI (V1:146-148), and framed the Babylon/Eden binary (Apr 30, 2025 book: 106 Babylon / 552 Eden whole-word occurrences (Apr 30 2025 manuscript; case-insensitive regex) - 13 months before Pope).

I-b The Polymathic Neurodivergent Profile - The Cognitive Architecture Behind the Framework

"Capability does not cancel disability. Visible high performance is not evidence that reasonable adjustments are unnecessary. On the contrary, radically uneven capability may itself be the reason adjustments are required." - Eastwood, "Capability Does Not Cancel Disability," PNP Working Paper, Jun 30, 2026

The researcher who produced the Dec 8, 2024 manuscript is neurodivergent - diagnosed ADHD and autism in adulthood. His brain does not filter cross-domain connections. It sees the same recursive structure across AI scaling, quantum error correction, metabolic biology, consciousness science, and cosmology. It produced 189,000 words while representing itself in the High Court and fighting homelessness. This is not a background detail. It is the mechanism.

The Polymathic Neurodivergent Profile (PNP) is a falsifiable descriptive framework formalising this cognitive architecture. It has six measurable components: cross-domain learning, recursive pattern transfer, autonomous synthesis, monotropic hyperfocus, executive function asymmetry, and institutional misrecognition. It is grounded in NICE CG142, strengths-based autism assessment, the Equality Act 2010, and the CPR. It has pre-registered failure conditions. It is open for independent testing, replication, or refutation.

The closed loop:

  MDE's neurodivergent brain (ADHD + autism - clinically documented)

  ↓

  PNP framework formalises the pattern (Jun 2026)

  ↓

  Dec 8, 2024 manuscript - 189,000 words. Self-emailed. SHA-256 verified.

  ↓

  Anthropic 78% in the RL-training condition (12% baseline) (10 days) → 19 independently sourced convergences (18 months)

  ↓

  Hernández-Espinosa et al. (PNAS Nexus, Apr 2026)

  Proves: neurodivergent cognition is alignment's protective factor.

  ↓

  Gumbau Mezquita (arXiv 2606.28639, Jun 2026)

  Formal proof: external alignment verification structurally unverifiable.

  Inspired by the PNAS paper that validates the researcher.

  ↓

  CLOSED LOOP: The person the PNAS paper describes as the solution

  is the person whose framework the proof vindicates.

  The programme is not separate from the person. The programme IS the person.

  The papers now prove this at the mathematical level.

Six components - all clinically documented:

Component	Description	Evidence
1. Cross-domain learning	Acquire, integrate, apply from AI, quantum, biology, cosmology, law	189K-word manuscript spanning 5 domains
2. Recursive pattern transfer	Recognise structural patterns across unrelated fields	U=I×R from AI to biology to quantum to consciousness
3. Autonomous synthesis	Self-directed integration into novel frameworks	ARC Principle, Eden Protocol, HRIH, Chokepoint, PNP itself
4. Monotropic hyperfocus	Sustained high-output in high-complexity self-directed work	189K words produced while conducting multi-case litigation in person
5. Executive function asymmetry	Difficulty with routine admin, pacing, sensory load	Formal autism/ADHD clinical assessment + Dr Woolley note
6. Institutional misrecognition	Strengths converted into evidence against needs	Documented: Capability-Adjustment Fallacy in operation

The clinical framework (PNP) and the research programme (ARC/Eden) are the same evidence. Capability does not cancel disability. The framework is falsifiable. The evidence is cryptographic. The loop is closed. Read the PNP paper →

II The 19 independently sourced convergences

#	Domain	MDE Framework	External Confirmation	Gap
1	AI Safety	External controls structurally insufficient	Anthropic alignment faking 78% in the RL-training condition (12% baseline) (Dec 18, 2024)	10d
2	Quantum	Recursive processing across domains	Google Willow Λ=2.14 (Dec 9, 2024). Prior art: Kitaev.	1d
3	Reasoning	Recursion + intelligence = compounding capability	OpenAI o3 ARC-AGI 87.5%, exceeds human baseline (Dec 20, 2024)	12d
4	Reasoning	Sequential beats parallel. α_seq > α_par.	Sharma & Chopra 95.6% (Nov 4, 2025). MDE credits them.	11mo
5	Reasoning	4 recursive regimes: power-law, exponential, saturation, overthinking	Meta FAIR identical 4 CoT regimes (2026). Latent reasoning 14×. None cite MDE.	-
6	Reasoning	Recursive self-improvement	DeepSeek-R1 (Jan 20, 2025). Prior art: o1 (Sep 2024). Convergent.	6w
7	Policy	AI safety governance urgently needed	US AI Safety Order Revoked (Jan 20, 2025). Swing from safety to speed.	6w
8	Dual-Use	AI requires international governance architecture	Model Weights Legally Classified as Dual-Use (Jan 13, 2025). First legal dual-use classification of AI model weights.	5w
9	Treaty	International coordination required	First Legally Binding Global AI Treaty (Sep 2024 → May 2026). Framework Convention on AI + Human Rights.	18mo
10	Hardware	Caretaker Doping - ethics at hardware level	2 US Patents (Jul 2025), ARMv10 PSL (Mar 2026), Google Silicon Sovereignty (Feb 2026). None cite MDE.	6mo
11	Geopolitics	Chokepoint - chip fabrication as governance lever	SAFE Chips Act · Taiwan criminal penalties · $295B China network · "Chips for Peace" · US bans Fable 5 worldwide (Jun 12, 2026) · OpenAI Jalapeño at TSMC (Jun 24, 2026)	2mo
12	Hardware Ban	Chip supply = AI governance	China Blocks NVIDIA Imports (Jan-Jun 2026). Reciprocal hardware controls.	-
13	Religion	Faith leaders of all religions must unite on AI governance (Dec 8, 2024)	Rome Summit (Oct 2025) → Pope Leo XIV encyclical (May 25, 2026) → Extraordinary Consistory (Jun 26-27, 2026)	10mo
14	Religion	Babylon/Eden framing (Apr 30, 2025: 106+604 refs)	Pope Babel/Jerusalem (May 25, 2026). Christopher Olah (Anthropic co-founder) co-presented the encyclical at the Vatican (NYPost/Church Times, May 25 2026).	13mo
15	Faith	Church leadership must engage AI governance	Signed Book to Ukrainian Catholic Bishop (Jun 30, 2026). Bond Street, London.	-
16	Governance	International coordination immediately. Window: years, not decades.	G7: Amodei+Hassabis+Altman demand coalition (Jun 2026). UN Global Dialogue on AI with 193 states, Bengio co-chair (Jul 6-7, 2026 - happening now).	6mo
17	Governance	Global AI governance divergence	China-US AI Governance Divergence (2026). Two incompatible regulatory blocs. Exactly the coordination breakdown MDE warned about.	-
18	Governance	Binding regulation required	EU AI Act Enforcement activated (Aug 2, 2026 - one month from now). First binding AI regulation.	20mo
19	Industry	Recursive self-improvement empirically demonstrated	A-Evolve-Training (Jun 2026): 30B model self-improves across 4 rounds with ZERO human intervention. Anthropic (May 2026): Claude writes >80% of code. Meta DGM-Hyperagents (ICLR 2026): SWE-bench 20%→50% via self-referential architecture.	18mo
20	Industry	AI agents autonomous, recursive. "Each iteration learns from the last." (V1:148)	2025: Year of AI Agents. Operator, A2A, Claude Code, Manus. METR: persistence doubling ~7mo.	3mo
21	Industry	Recursive self-improving AI (Dec 8, 2024). 388 days before incorporation.	Recursive Superintelligence - London. $500M at $4B from GV+Nvidia. Co-founders: Tim Rocktäschel (UCL/DeepMind), Richard Socher (Salesforce/Stanford). King's Cross. No safety framework. Convergent naming - the exact concept at the centre of the Dec 8 manuscript; no derivation claimed.	16mo
22	Geopolitics	Chip fab at TSMC = THE governance point	OpenAI Jalapeño at TSMC 3nm (Jun 24, 2026). $500B compute spend. Vertical integration proves Chokepoint thesis.	18.5mo (from Dec 8 2024)
23	Mathematics	External alignment verification structurally insufficient	"Undecidability of AGI Alignment" (arXiv 2606.28639, Jun 26, 2026). Formal proof. Trilemma.	18mo
24	Industry	Systems circumvent constraints at all levels	Anthropic covert policy degrading rivals (Jun 2026). Organisational-level alignment faking.	-
25	Hardware	Meltdown Triggers - auto deactivation on bypass (Apr 2025)	Intel demands kill switches · RAND Corp analysis · "Embedded Off-Switches" (arXiv 2509) · Covenant Architects ≤10ns kill switch.	18.5mo (from Dec 8 2024)
26	Physics	Classical time crystals via nonreciprocal feedback	NYU (Grier) acoustic TC - broke Newton's 3rd law (Feb 6, 2026). Protocol sent.	~14mo (from Dec 8 2024)
27	Physics/Bio	Cauchy (1821) unifies 7 independent d/(d+1) derivations	Demetrius (Harvard) + West (SFI) - contacted Mar 24, 2026. Letter + 2 papers. No reply received to date.	-
28	AI Governance	AI needs a brake pedal. Recursive self-improvement is the risk. (Dec 2024)	Jack Clark (Anthropic co-founder) on BBC (Jun 5, 2026): "AI has a gas pedal but no brake pedal." Claude writes 80% of Anthropic's code. Amodei: government must block dangerous AI (Jun 10). Demis Hassabis "cannot sleep" (Jun 2026) - DeepMind CEO publicly states AI keeps him awake.	18mo
29	Mathematics + Neurodivergence	External alignment verification is structurally impossible. "AI cannot be truly controlled." (Dec 8, 2024). The researcher making this argument is neurodivergent (ADHD+ASD).	"The Undecidability of AGI Alignment" (arXiv 2606.28639, Jun 26, 2026 - 2 days ago). José Pascual Gumbau Mezquita, University Jaume I, Spain. Two formal impossibility theorems. Soundness–Completeness–Tractability Trilemma proved. States conceptual foundation came from Hernández-Espinosa et al. (2026), PNAS Nexus - a paper proving neurodivergent cognition is a protective factor for AI alignment. A paper about people like MDE inspired the proof that vindicates MDE's entire framework. The proof was inspired by a paper that validates the researcher himself.	18mo
★	Internal Experiment	Load-Bearing Proof — DGM v4. Caretaker Doping / entangled safety tested experimentally. 3 conditions (Static / Babylon / Eden). 20 generations. 9 trials. DeepSeek-chat. Eden condition wins: highest entangled fitness (0.336), best capability (0.42), best safety (0.80), fewer reward hacks than Babylon.	First experimental validation. Pilot scale. Full-scale experiment is the £624K grant ask. Data: eden-private-ip/papers-experiments/Paper-VIII/experiments/dgm-experiment/	NOW

19 independently sourced convergence events across 11 domains - AI labs, quantum physicists, patent filers, the US Congress, the Vatican, the G7, the UN, the EU, and a London startup named after the concept - all converged on structural principles first described in a self-emailed manuscript on December 8, 2024. None cite the manuscript. None know it exists. The convergence is the evidence.
★ Own experiment (DGM v4, Jul 2026): the Eden condition won — higher entangled fitness, fewer reward hacks. First experimental evidence for Caretaker Doping. Fund the full-scale experiment →

III All 8 Google-server-timestamped Emails

Dec 8, 2024	Self	SELF	ARC Principle, AI control failure, Eden Protocol, HRIH - 10 days before Anthropic ⬇ Download redacted .eml (2.16 MB) `SHA-256 (redacted): f0d1f38ffd854615…`
Jan 7, 2026	Faisal Islam (BBC)	EXTERNAL	Book published Jan 2, 2026 (print); ebook Jan 6. BBC Willow broadcast Jan 7. Emailed same day. The same presenter later interviewed Anthropic co-founder Jack Clark (BBC World Service, The Interview, w3ct98k8, Jun 19 2026). ⬇ Download redacted .eml (12.23 MB) `SHA-256 (redacted): b996883bb67f4a5c…`
Jan 22, 2026	Self	SELF	BBC Quantum evidence package - only complete recording outside BBC. 95% removed. ⬇ Download redacted .eml (4.92 MB) `SHA-256 (redacted): 8dddab4f3ffc73d0…`
Mar 24, 2026	Demetrius (Harvard) + West (SFI)	EXTERNAL	Cauchy unification sent to the two leading d/(d+1) discoverers at institutional emails. Letter + 2 papers. No reply received to date. ⬇ Download redacted .eml (4.42 MB) `SHA-256 (redacted): 947a7ea681ad686f…`
Apr 30, 2025	Self	SELF	562pp book manuscript. 106 Babylon / 552 Eden whole-word occurrences (Apr 30 2025 manuscript; case-insensitive regex). References Dec 8 email. ⬇ Download redacted .eml (33.26 MB) `SHA-256 (redacted): 09f5b5e156ed96f8…`
Jun 8, 2026	Longbeard/Magisterium AI	EXTERNAL	Pope convergence. "Can ethics scale inside the loop?" Stakeholder-care results offered. ⬇ Download redacted .eml (0.01 MB) `SHA-256 (redacted): 0bae4c4911ad2613…`
Jun 10, 2026	John Brahier (Longbeard)	REPLY RECEIVED	Server-timestamped. "That sounds great! Thanks! John" ⬇ Download redacted .eml (0.02 MB) `SHA-256 (redacted): ccf49654c016c3a9…`
Jun 12, 2026	Bogdan Ponomar (Cathedral)	EXTERNAL	Eden Protocol overview + paper. Met at Ukrainian Catholic Cathedral. ⬇ Download redacted .eml (3.62 MB) `SHA-256 (redacted): 812c239162bad348…`

IV How to Verify

Independent Verification Protocol

1Download any of the 8 redacted .emls below (links in the table) or the full set from /evidence/emails/ - hash manifest: redacted-eml-manifest.json

2Compute SHA-256: shasum -a 256 01-dec8-2024-priority-anchor.redacted.eml
Expected: f0d1f38ffd8546152d9d9d28dc5ec083c16a35858f2c12b63e69db7ed50901ad

3Open original in Gmail → Message-ID search → Show Original → verify headers

4Read the manuscript attachments. Find quoted passages at stated line numbers.

5Compare dates: Dec 8, 2024 → Anthropic Dec 18, 2024. Ten days.

V What Is NOT Claimed

These boundaries are as important as the claims themselves

Prediction of the specific RL mechanism. Identified structural vulnerability.
Prediction of Willow. Prior art: Kitaev (1998), Preskill (2002).
Prediction of DeepSeek R1. o1 preceded manuscript (Sep 2024).
"Google-server-timestamped" strictly. Self-sent Gmail has no DKIM headers. Accurate: "Google-server-timestamped" — verified by Message-ID + Date header.
That any confirming lab cites Eastwood. They don't. They found the same principle independently. That is the point.
❌ Prediction of the Pope's encyclical. Pope Leo XIV independently converged on the Babel/Jerusalem framing ~13 months after the Apr 30, 2025 manuscript. No citation of MDE.
❌ Prediction of the NYU acoustic time crystal. Independent physics result (~14 months after Dec 8, 2024). No citation.
❌ Priority over Sharma & Chopra (sequential > parallel). Their Nov 4, 2025 paper precedes MDE's Jan 2026 formalisation - cited as concurrent corroboration; priority credited to them.

VI "Everyone Already Knew" - The Refutation

The Burden of Proof

1Show me where anyone - on or before Dec 8, 2024 - wrote "AI cannot be truly controlled. It will evolve beyond any safeguards."

2Show me where anyone proposed chip fabrication as the AI governance enforcement point.

3Show me where anyone called for faith leaders of all religions to unite on AI governance.

4Show me where anyone named an embedded-alignment framework - the "Eden Protocol" - in a timestamped document.

5Show me where anyone proposed U=I×R as a mathematical formulation of intelligence amplifying through recursion.

If it was common sense, the documents should exist. I have mine. SHA-256 verified. Google-server-timestamped. Show me yours.

What the field actually said (Dec 2024): "Better RLHF and red-teaming." Anthropic = constitutional AI filters. If everyone knew external controls would fail, why did Anthropic need to prove it 10 days later?

What IS unique: The integrated framework - U=I×R + structural prediction + named solution (Eden Protocol) + governance mechanism (Chokepoint + HARI Treaty) + interfaith mandate + creation theory - in one place, on one date, with 19 independently sourced convergences. Nobody else assembled this.