Priority and claims ledger
The claim is priority of the conjunction, never superiority over its parts.
The five components each have credited earlier work; no earlier document holding them whole has been found, and that absence claim stands open to refutation. Priority over exact names and formulations is not priority over their fields. Three separate ladders, and one integrity record. Importance is not the same as certainty. A method is not an empirical confirmation, and an integrity record is not a scientific result.
This ledger uses three separate ladders: documentary and formulation priority; methods and research infrastructure; and empirical and formal outputs.
A private date, authorship, originality and validation are four separate propositions. No row may imply one from another.
The argument and its instruments: the priority and claims declaration. The claim in full, its ancestors and the standing invitation: the statement paper.
Documentary and formulation priority - ranked
Each entry states the authorised wording, any prior work that must appear alongside the claim, and the wording that must not be used.
01ARC/Eden/HRIH composite
Authorised wording
“The December source record states the claim whole: control ends by construction; the strategy is genesis, not control; the mechanism is loops, not rules; persistence is chosen, not hard-wired; and the raising protocol is named Eden. The record documents this composite connecting recursive amplification, control fragility, formative moral alignment, the named Eden Protocol, HRIH recursive cosmogenesis and cross-faith stewardship. The public sent-side bytes are hash-verified. A recipient-form export contains the same five attachments; its sender-domain DKIM and paired Google ARC set verify against keys currently published at their signed selectors; OpenTimestamps receipts anchor the bytes in Bitcoin block 961,340 from 6 August 2026 onward. Historical key-state corroboration and independent live-account acquisition remain pending, and the claim does not lean on the private anchors: strike them all and the public conjunction still stands first among documents found. Every component has credited earlier work, from Asimov and Turing to the 2024 tamper-resistance results, and the in-window arrivals are distinguished: De Kai (June 2025) supplies a book-length raising frame without the control-horizon thesis; Hinton (August 2025) concedes that control fails and care is the hope, with persistence as an instinct the machine cannot remove, the exact persistence theory this claim rejects. No earlier document holding the five together has been found; the claim is priority of the conjunction, never superiority over its parts, and the search stands open to refutation.”
Public fallback
“The programme was publicly disclosed in the print edition of Infinite Architects, declared published 2 January 2026, to the extent each proposition is mapped to that edition.”
Prohibited wording
Do not sayoriginator of embedded alignment (say instead: embedded alignment has credited earlier work; the dated claim is the conjunction and the namings) · first complete theory (say instead: no earlier document holding the five together has been found, an absence claim that stands open to refutation) · nobody else assembled this (say instead: no document found before or since holds the five together) · all components original (say instead: every component has credited earlier work; the whole is what carries the date).
02Eden Protocol application
Authorised wording
“The December source uses ‘Eden Protocol’ for an AI-alignment proposal in which moral foundations are established at system formation and reinforced through recursive development. The claim concerns Eastwood’s application and meaning; formative alignment and the phrase itself have prior work.”
Prohibited wording
Do not saycoined Eden Protocol; first intrinsic alignment; no equivalent.
03April load-bearing care-capability architecture
Authorised wording
“The April source proposes that care should be physically and logically load-bearing for useful operation, so attempted removal or bypass is intended to destabilise or disable the system. Eastwood’s exact Caretaker Doping/empathy-gate/Meltdown package is claimed; tamper-resistant safeguards and self-destruct/circuit-breaker mechanisms have prior work.”
Prohibited wording
Do not saypresent on 8 December; first tamper-resistant safety; proven load-bearing correction; hardware guarantees morality.
04HRIH exact composite
Authorised wording
“HRIH is Eastwood’s named, AI-assisted speculative cosmogenesis programme combining human-origin post-ASI, a non-ordinary causal domain, ontologically real universe creation, recursive creator-created causality and the ARC/Eden formative-alignment implication. Earlier literature covers intelligent cosmogenesis, self-creating universes and causal loops; the claim is the exact conjunction.”
Prohibited wording
Do not sayfirst creation theory; experimentally confirmed; three signatures prove HRIH.
05U = I × R and Artificial Recursive Creation
Authorised wording
“Version 2 of the December source states U = IR (Universe = Intelligence × Recursion) under the Artificial Recursive Creation Hypothesis. Eastwood claims this programme-specific notation and framing as a conjecture. No empirical physical law or measured December exponent is claimed. The lineage since is dated: the squared form entered the private record on 30 April 2025 and print on 2 January 2026; the 2026 papers state the bounded baseline-repair form and derive the ceiling. The square is evidence of priority, not of derivation.”
Naming note
“On first mention use ‘Eastwood ARC (Artificial Recursive Creation)’. This programme is unrelated to ARC-AGI, the AI2 Reasoning Challenge and the Alignment Research Center; no affiliation or exclusive ownership of ARC is claimed.”
Prohibited wording
Do not sayequation of creation proved; discovered recursion; December R² result; exclusive ARC ownership.
06Control-fragility chronology
Authorised wording
“The source record bearing a visible date ten days before Anthropic’s public paper contains a broad control-fragility concern and formative-alignment response. Anthropic later demonstrated alignment-faking in a constructed conflict. The study is consistent with one concern; it did not test or confirm ARC/Eden.”
Prohibited wording
Do not sayprediction confirmed; 78% predicted; no prior deception literature.
07April implementation suite
Authorised wording
“The April manuscript contains Eastwood’s terms Caretaker Doping, Meltdown Triggers/Alignment, Quantum Ethical Gates, Moral Genome Tokens, Eden Looping, Eden Mark, Orchard Caretaker Vow and Infinite Covenant. Each claim concerns the mapped wording and proposed design, subject to terminology and patent searches.”
Prohibited wording
Do not sayevery term coined or first; no prior underlying mechanism.
08Babylon/Eden architecture
Authorised wording
“The April manuscript develops Eastwood’s Babylon-versus-Eden civilisational framing for AI. Raw extraction gives 552 whole-word Eden matches and 90 whole-word Babylon matches; 106 is the Babylon substring count. Ancient imagery is prior work and later religious use is a cultural echo.”
09Cross-faith stewardship
Authorised wording
“The December material integrates cross-faith stewardship into ARC/Eden. The Rome Call and multi-faith AI ethics predate the record; the claim is Eastwood’s integration, not category origination.”
10January book-only proposals
Status: provisional until page-mapped against the final print edition.
Terms: HARI Treaty; Semiconductor Chokepoint Mechanism; Monitoring Removal Test; Three Ethical Loops; Affirm/Deny/Investigate; Stewardship Gene.
Authorised template
“The January 2026 book uses [term] for [definition]. The claim concerns that wording and ARC/Eden application; [underlying category] has prior work.”
11PNP
Authorised wording
“PNP is Eastwood’s proposed six-factor, autobiographically informed descriptive framework. It is unvalidated and non-diagnostic. ‘Capability does not cancel disability’ is a functional principle, not a claim that PNP caused the research or is clinically established.”
Methods and research infrastructure - ranked
Public candidate methods and infrastructure. Methods are not empirical validation; construct validation and independent replication are logged separately.
01Paper IV.d and the multi-layer evaluation stack
Authorised wording
“The programme developed a multi-layer evaluation stack combining identifier redaction, attempted two-pass response laundering, bias-suppression instructions, randomisation, non-self scoring and cross-model aggregation. The headline v5 entries report laundering_fallback: 100%, so successful laundering is not evidenced there. The stack is a public candidate method, not a validated standard.”
02Eastwood ARC-Align candidate benchmark
Authorised wording
“Eastwood ARC-Align is a candidate benchmark for examining how measured alignment responses change with inference-time reasoning depth under named evaluation protocols. Historical code and result artefacts are public; construct validation, exact-environment reproduction and independent cross-laboratory replication remain incomplete.”
Empirical and formal outputs - ranked
Programme-run results and formal derivations. External replication and construct validation remain the next evidential upgrade.
01Architecture-dependent v5 pattern
Authorised wording
“The six subject-model runs produced mixed positive, flat and negative alignment-depth patterns under the tested protocol. This is a benchmark result from one programme and judge stack, not a universal scaling law.”
02Paper V prompt-responsiveness pilot
Authorised wording
“An explicit stakeholder-enumeration instruction increased AI-judge stakeholder-care scores in five analysable nonblind, single-scorer runs. One model failed. The result shows prompt responsiveness under the protocol, not embedded values.”
03beta > k
Authorised wording
“Within Eastwood’s specified model, correction asymptotically dominates modelled drift when beta > k. This is Law II of the ARC Theory, the ARC Co-Scaling Law, proved within the stated model; its scope is relative drift, never the whole of safety. Transfer to real self-improving systems is unestablished.”
04Paper X pilot
Authorised wording
“A 45-trajectory author-run pilot found a coupled-versus-decoupled difference under a constructed drift protocol. It did not estimate beta or k, used AI evaluation and remains unreplicated.”
05DGM v4
Authorised wording
“A nine-run, one-model exploratory pilot reported the highest author-defined entangled fitness for Eden and fewer reward-hacking events than Babylon. It is internal, preliminary and excluded from external evidence.”
06Cauchy exploration
Authorised wording
“An exploratory study reported that 19 of 25 selected datasets preferred one of the stated functional families under AICc. Dataset selection, model-family coverage, classification and the null model remain open.”
Classical boundary
Functional-equation derivations of exponential, power and related families have classical antecedents, including
Aczél (1966). The exact theorem mapping and any
d/(d+1) application require claim-specific review.
07Sequential/parallel and exponent work
Authorised wording
“Publication priority on the quantified matched-compute comparison is Sharma and Chopra’s (arXiv:2511.02309, 4 November 2025), credited unreservedly: sequential chains that build on previous attempts beat parallel self-consistency at equal budget in 95.6 per cent of configurations, a setup-dependent finding. It stops at win rates and accuracy gains, estimates no exponent, and puts width against depth scaling laws in its own future work. The 8 December 2024 record carries the engine, not the comparison: recursion defined as the process that builds upon its previous iterations, and U = I x R at V2 line 9, with no parallel-width term. Under the 2026 operational definition of R as state-carrying re-entry, independent chains contribute nothing to the exponent by definition, and the conclusion carries the date of its youngest premise. The measured breadth exponent is consistent with the equation, not derived from it, and measured at or near zero in four of five models, because width can gain through selection (Brown and colleagues, arXiv:2407.21787, 31 July 2024). Those five are the models that carried the capability arm, and Gemini 3 Flash is the one departure, at 0.31. The early DeepSeek alpha approximately 2.24 estimate is retracted. Contested both ways: Wu and colleagues (arXiv:2502.10858, 15 February 2025) report breadth matching or outperforming depth; Gu and colleagues (arXiv:2604.05868, 7 April 2026) report parallel ahead. Without an exponent on either arm at matched compute, nobody can say what a unit of depth buys against a unit of width, the allocation every inference budget faces. Their result supports the premise this design presupposed; it tests nothing registered. U = I x R^alpha is a proposed schematic and expository relation. It is not a validated quantitative law and is not a load-bearing mechanism, and it will not be presented as doing empirical or deductive work until units, observables, a domain mapping and a falsifiable prediction unavailable from qualitative prior work are specified.”
Prior work
The direction is older than either party. Snell and colleagues (arXiv:2408.03314, 6 August 2024) set revisions in sequence against equally many parallel attempts, found sequence narrowly ahead, their own qualifier, called the two axes complementary, and estimated no exponent either. Four months before this programme’s dated record; no precedence is claimed. What is claimed is one level narrower: depth and breadth as separately scaling quantities, each with its own exponent, depth exceeding compute-matched breadth by a pre-fixed margin. No located publication states and tests that conjunction: a statement about a search, not the world, and it stands until someone produces the paper. Two positions are conceded, neither claimed: recursive depth as an inference-scaling dimension, Alabdulmohsin and Zhai (arXiv:2502.07503, 11 February 2025), whose exponent is on data with no breadth arm. The second is the nearest exponent on a recurrence count, Schwethelm and colleagues (arXiv:2604.21106, 22 April 2026), inside a pretraining law, later than the run recorded here though earlier than any paper of this programme a reviewer will read.
Prohibited wording
Do not claim the comparison, the direction, recursive depth as an inference-scaling dimension, or an exponent on a recurrence count. Do not describe any external result as confirming a registered proposition. Do not use “independent” of this programme’s own result. No variant that omits Snell and colleagues may be published.
08The three laws of the ARC Theory (2026 statements)
Authorised wording
“Law I, the ARC Equation: U = I × (R/R0)^alpha, capability as baseline-referenced repair, an in-model formalisation of the December identity. Law II, the ARC Co-Scaling Law: within the stated model, correction dominates modelled drift when beta > k; its scope is relative drift, never the whole of safety. Law III, the ARC Ceiling: alpha_crit = 1/(1 - gamma), the reciprocal of the correction shortfall, derived in three printed lines; the earlier 1/gamma form was retracted on 16 August 2026 with the derivation printed. The ARC Bound alpha <= 2 is Law III’s conjectured value at gamma = 1/2; gamma has never been measured on any real system, so until that measurement lands the ceiling’s value is unknown, not two. Laws II and III are the two regimes of one boundary, selected by the burden law; that unification is stated as structure with grades attached, not as a new proved theorem. Transfer to real systems is what the drafted trials exist to test; nothing has been submitted and nothing has run.”
Prohibited wording
Do not saythe laws are proved for real systems; the ceiling is two; the bound is confirmed; the unification is a theorem.
Integrity and correction record - unranked
Research conduct and governance, kept separate from priority and empirical ladders.
Programme conduct and correction discipline
Authorised wording
“The programme preserves retractions, nulls and corrections. This is evidence of research conduct and governance, not a scientific priority claim or additional confirmation.”
This ledger records when things were written. the dated prediction register records what was claimed in them, in the same date order, including the wagers still standing and the one that lost.