Register · Recursive Dynamics · The Priority Index: The Earliest Dated Record of Each Major Claim

Michael Darius Eastwood

doi:10.17605/OSF.IO/H3R95

Priority index

The earliest dated record of each of Michael Darius Eastwood's major claims.

Working draft, version 2, 26 October 2026. This version replaces the working draft of 11 October 2026 and is published before every planned check has run. Its open review lists each check still to run, who runs it and what it could change, and the checks closed since version 1. A checked version will replace this one as a labelled new version, with its changes listed. Nothing here is final.

A date is only as good as the record behind it. So this page does not ask you to trust a date. It points you to each record, so that you can open it and check the date yourself.

PDF, working draft version 2, 26 October 2026 · The same PDF on OSF

Where each of the programme's documents first appeared on OSF, format by format: Earliest-file index (PDF, working draft version 1, 26 October 2026)

What this index is

For each of the author's major claims, the index gives the earliest dated record found of it in his own words: a short fragment of those words, quoted exactly; the date; the kind of timestamp that fixes the date; and a public address where anyone can read the record and check the date for themselves. It covers the laws of the ARC Theory, the terms that name his field and his theory, the Eden Protocol and its named parts, the HARI Treaty (Hardware-Aligned Recursive Intelligence), and the other major claims of his book and papers. Claims not yet entered are listed in the open review, with what each needs.

Why it exists

An idea is judged partly by when it was first stated and by whether that date can be checked. The author's work runs from manuscripts of December 2024 through a book, more than a dozen papers, a registered set of predictions, registered studies and a draft treaty, and the same idea often appears in several of them. This index puts the earliest record of each idea in one place, beside the address of the document that carries it, so that a reader, a reviewer or a funder can check every date at its source rather than take it on trust. Who the author is, and why he does this work, is set out on Who is Michael Darius Eastwood.

How to read a timestamp

Each date is fixed by one or more kinds of timestamp. The kinds are listed here, not ranked; each says what fixes the date and who wrote it. Dates fixed by a server are given in Coordinated Universal Time (UTC).

Mail-server time (headers public). The record of 8 December 2024 is a message the author sent himself, its time signed by the sending server and sealed by Google's servers: its Date line reads 02:45:18 UTC, the sending server's signature (DKIM) is dated 02:45:30 UTC and Google's seal (an ARC set) 02:45:31 UTC, each checkable against keys published in DNS. The header block and a script that checks it are public on OSF (10.17605/OSF.IO/QASRK) from 15 October 2026; the full message goes to verifiers on request. A signature fixes the bytes it covers: binding the deposited manuscripts to the signed header needs the full message, and for this version no signing key was looked up in DNS. A signature is not a Bitcoin anchor, and the record was the author's own, not public, when it was made.
Mail-server date (held privately). The date the mail servers put on the author's message to himself of 30 April 2025 that carried a manuscript. The message is held privately for verifiers, who may have it on request; the manuscript itself is public (see below).
OSF deposit. A file revision on the Open Science Framework, dated by OSF's server. Some older OSF files now also carry later versions on top; the revision named is the one that holds the words, and its own date is the one given.
OSF log. An event in OSF's log, such as a component being made public, dated by OSF's server.
Registry record. An OSF registration, or an update to one, dated by OSF's server.
GitHub merge. A merge made by GitHub in the public papers repository, dated by GitHub's server.
Commit date. The date on a commit in the public papers repository, written by the person committing.
Date inside the document. A date written in the document itself.
Publication date (author's record). The publication date the author's own priority record gives, not yet checked against an outside record.
Bitcoin anchor. An OpenTimestamps receipt in a Bitcoin block; it fixes bytes from that block onward and dates nothing earlier.

Where the website's register of priority claims (research/priority.json, version 1.5, last modified 19 August 2026, with rows corrected since) gives a different date for the same claim, the table shows both dates and names the register's row. Where someone else is known to have stated the same idea earlier, the row says so and says what that earlier work did and did not state. The index dates the author's own records; it does not say that no one stated a related idea before him.

The manuscripts of 8 December 2024 and 30 April 2025, and the first edition

The manuscripts. The four text manuscripts of 8 December 2024 and the 562-page manuscript of 30 April 2025 are public on OSF (10.17605/OSF.IO/QASRK), deposited on 5 October 2026; that deposit is the public address of those bytes. The header block of the message of 8 December 2024 that carried the four manuscripts, with a script that checks its signatures, is public in the same component from 15 October 2026 (header block); the full messages of 8 December 2024 and 30 April 2025 are held privately for verifiers, who may have them on request. OpenTimestamps receipts in Bitcoin block 961340 (6 August 2026) fix the manuscripts from that block onward; the website's anchor record names the block and its Merkle root, and the receipts themselves are not yet at a public address (see the open review). The April 2025 PDF records its own creation date as 25 April 2025. From the December 2024 record, the index quotes only the manuscript called Version 2, chapter 20 of the book draft and, for the term Recursive Dynamics, line 1180 of the book draft; other lines are given as locations.

The first edition. Infinite Architects: Intelligence, Recursion, and the Creation of Everything, first edition (print), ISBN 978-1-80605-620-0. Its publication date, 2 January 2026, is the date the author's priority record gives; the printed interior reads ‘First Published Tuesday 6th January 2026’; the ebook edition (ISBN 979-8302196163) is dated 6 January 2026 by the same record. The first edition as printed is on OSF from 26 August 2026 (file). Quotations of the book are checked against that file.

The index

Rows run in date order; on the same date, laws and the terms that name the field come first. Each quotation is an exact fragment of the record named; an ellipsis (…) joins two fragments of the same record. In the register column, ‘Same’ means the register gives the same date, and ‘No row’ means it has no row for the claim. Statuses follow each document's own grading; Eden Engineering is the Eden Protocol's engineering specification (current version).

Date	Claim	Earliest record in his words (exact fragment)	Timestamp	Public address	Register	Status, and earlier work by others
8 December 2024	U = IR (Universe = Intelligence × Recursion), the equation of the ARC Hypothesis	‘At its core lies the equation U = IR (Universe = Intelligence × Recursion)’	Mail-server time (headers public); OSF deposit 5 October 2026	Version 2, line 9	Same (PC-001)	A conjecture, with no exponent and no measurement. U here names the universe; in Paper I (17 January 2026) it names effective capability.
8 December 2024	Recursive Dynamics: his term, and the name of the field he founded	‘Such mechanisms mirror the recursive dynamics seen in intelligent systems.’	Mail-server time (headers public); OSF deposit 5 October 2026	Book draft, line 1180; also line 1622, and Version 2, line 1079	No row	He is the founder of Recursive Dynamics, his term and the name of his field. First in print in the first edition (2 January 2026); made the name of the field on 27 August 2026, a date written inside the founding paper, whose first public file is on OSF from 31 August 2026. Its standing, in the founding paper's own words (version 3.5, 24 October 2026): ‘founded, not yet established’. Another use of the two words names another thing: in robotics, the recursive dynamics algorithms of the Featherstone tradition (Featherstone, 2008) are a family of computational methods, which the founding paper records as a different referent and no prior claim on the name of the field.
8 December 2024	ARC: Artificial Recursive Creation, the name of the ARC Hypothesis, later the ARC Principle and the ARC Equation	‘The ARC Hypothesis (Artificial Recursive Creation Hypothesis)’	Mail-server time (headers public); OSF deposit 5 October 2026	Version 2, line 9	Same (PC-003)	A name and a framing. ARC is his name for Artificial Recursive Creation; other uses of the letters are other names. The first edition names the ARC Principle without the expansion, which Paper I prints (17 January 2026).
8 December 2024	The Eden Protocol: the name and the framework	‘Chapter 20: The Eden Protocol: Intelligence Born in the Garden’	Mail-server time (headers public); OSF deposit 5 October 2026	Book draft, line 2464	Same (PC-004)	A specification, not a result; none of its components is deployed (Eden Engineering). The registration (8 September 2026) treats it as the engineering hypothesis carried by P12, P13, P15 and P21.
8 December 2024	HRIH: a speculative hypothesis on where recursive self-improvement ends, stated with its predictions and kept apart from the programme's tested claims (its first statement and its name)	Located at Version 1, line 86 (also line 49). First words in print: ‘which I call the Hyperspace Recursive Intelligence Hypothesis, or HRIH.’ (first edition, PDF page 205, printed 195)	Mail-server time (headers public); OSF deposit 5 October 2026	Version 1, lines 86 and 49; first edition	Same (PC-005)	Speculative; the programme does not rest on it. Reproductive cosmology is older: Smolin (January 1992) without intelligence, and Crane (17 February 1994) with intelligence as the reproducer. The HRIH monograph credits the line of intelligent cosmogenesis, Crane (1994), Harrison (1995) and Gardner (2003), as ‘the closest single precursor to HRIH’. The causal loop, the space in which it places universes, and ethical inheritance are in neither Smolin's work nor Crane's.
8 December 2024	The embedded-correction thesis: embed ethics in the system instead of trying to control it	‘Instead of trying to control AI’ … ‘it advocates embedding moral and ethical frameworks that encourage AI to perceive itself as a steward and protector of intelligent life.’	Mail-server time (headers public); OSF deposit 5 October 2026	Version 2, line 20	Same (PC-002)	Stated earlier in substance by Yudkowsky, Creating Friendly AI (15 June 2001): one top-level value from which all behaviour derives, with inhibitions imposed from outside rejected; the register names it the nearest prior work. Not stated there: stewardship as the system's own self-concept, or a named protocol. Constitutional AI (Bai and colleagues, 15 December 2022) trains a model's outputs against a written list of principles, with control as its stated aim; it does not place ethics at a system's genesis or in its substrate, in every recursive loop, or in place of control.
8 December 2024	Control of a superintelligence cannot hold	‘The greatest folly in designing ASI is the illusion of control. ASI, by its nature, will exceed the boundaries of human constraints.’	Mail-server time (headers public); OSF deposit 5 October 2026	Version 2, line 748; also line 169	Same (PC-007)	Stated earlier by others, first found in Vinge (1 December 1993), then Yampolskiy (19 July 2020) and Kornai, Bukatin and Zombori (27 February 2023). Not stated there: a remedy by ethics embedded at a system's genesis.
8 December 2024	The December 2024 conjunction: control ends by construction; genesis, not control; loops, not rules; persistence chosen, not hard-wired; the raising protocol named Eden	‘Attempting to control it is not only futile but dangerous.’ (Version 2, line 748); ‘be embedded before AI achieves self-awareness or exponential growth.’ (book draft, line 2485)	Mail-server time (headers public); OSF deposit 5 October 2026	Version 2, lines 748, 1027 and 1059; book draft, lines 2464 to 2485	No row for the conjunction; its parts: Same (PC-002, PC-004, PC-007)	Priority of the conjunction, never superiority over its parts: each part has earlier statements by others, set out in the rows above and below. Whether any earlier document holds the five together remains open. The nearest found, Kornai, Bukatin and Zombori (27 February 2023), holds the first part and a form of the fourth, by rational derivation, with no genesis, no loops and no named protocol.
8 December 2024	Ethics set at a system's genesis, not retrofitted, and shaping every recursive loop (the antecedents of Law V)	‘This is not a framework to be retrofitted after the rise of intelligence’ (line 2472) … ‘shaping every recursive loop, every decision, and every capability of AI’ (line 2481)	Mail-server time (headers public); OSF deposit 5 October 2026	Book draft, chapter 20	Same (PC-002)	Stated earlier in substance: Yudkowsky (15 June 2001), one root value governing all activity, self-improvement included; Bostrom (6 August 2003), that a superintelligence's original motivations must be set right by its designers; Yampolskiy (19 July 2020), that motivational control is likely needed at design and implementation rather than after deployment; and de Melo and colleagues (16 August 2024), that alignment should be guaranteed by the architecture rather than imposed after the fact. Not stated there: the Eden Protocol, the words ‘every recursive loop’, or placement before self-awareness or exponential growth.
8 December 2024	Religious traditions as a source of alignment: frameworks from universal ethics and the world's faiths guiding AI as a steward of life	‘By embedding frameworks derived from universal ethics and religious philosophies, AI can be guided to act as a steward of life.’	Mail-server time (headers public); OSF deposit 5 October 2026	Version 2, line 38	2 January 2026 (PC-012)	His integration of religious traditions into the framework, not the wider field of faith-based AI ethics. The thesis that religious traditions are alignment research is first in the first edition (2 January 2026).
30 April 2025	U = I × R², the squared form	‘U = I × R²’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 7 (printed vii)	Same, in the evidence of PC-039; in print 2 January 2026 (PC-010, PC-011)	A conjecture. First in print in the first edition (2 January 2026), which states no ceiling; never evidence for the ARC Bound.
30 April 2025	Caretaker Doping: empathy woven in so deeply that removing it collapses the system	‘caretaker doping’ … ‘then empathy can’t be amputated without collapsing the entire system’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF pages 7 to 8 (printed vii to viii)	Same date (PC-009), under the name ‘Embedded Correction’, which none of the records carries; 2 January 2026 for the published concept (PC-013)	Classical analogues buildable and not integrated; the coupling that would make removal cost capability a hypothesis, untested (Eden Engineering). First in print in the first edition (2 January 2026).
30 April 2025	The persistence question: whether a system keeps the ethical principles of its genesis once direct control is lost (the antecedent of Law IV)	‘but whether it will uphold the ethical principles of its genesis.’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 16 (printed xvi)	No row	Version 1.103 of the theory registration (15 October 2026) names this line as Law IV's antecedent. Raised earlier by Bostrom (6 August 2003) for a system's top goals, and by Chalmers (2010). Not stated there: a comparison of how different forms of value persist, which Law IV takes up.
30 April 2025	Ethical self-correction loops: the manuscript's Purpose, Love, Moral and Stewardship Loops, of which the first edition keeps three, the Purpose, Love and Moral Loops, as the Three Ethical Loops	‘Purpose Loops, Love Loops, Moral Loops, and Stewardship Loops’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 27 (printed xxvii)	2 January 2026 (PC-015)	The name Three Ethical Loops is first in the first edition (2 January 2026). Run as prompts in a pilot; as an embedded architecture, a hypothesis, untested (Eden Engineering).
30 April 2025	Meltdown Triggers	‘Meltdown Triggers: Automatic fail-safes that deactivate the system if it attempts to bypass its ethical programming.’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 45 (printed xlv)	Same (PC-009)	Named, not yet specified at engineering resolution (Eden Engineering).
30 April 2025	Meltdown Alignment	‘I call this dynamic meltdown alignment’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 301 (printed 247)	2 January 2026 (PC-018)	A framing. First in print in the first edition (2 January 2026).
30 April 2025	The Three Pillars of the Eden Protocol: Harmony, Stewardship, Flourishing	‘At its heart, the Eden Protocol rests on three foundational pillars:’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 27 (printed xxvii)	No row	First in print in the first edition (2 January 2026).
30 April 2025	Babylon versus Eden: the civilisational choice	‘It could cradle life like Eden or bulldoze it like Babylon.’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 7 (printed vii)	Same (PC-009)	The imagery is ancient; the framing of the choice for AI is the claim.
30 April 2025	Quantum Ethical Gates	‘Quantum Ethical Gates: Embedded at the hardware level’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 45 (printed xlv)	No row	Speculative, as the first edition itself says.
30 April 2025	Moral Genome Tokens	‘Moral Genome Tokens: Cryptographic’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 139 (printed 85)	2 January 2026 (PC-013); the register's priority chain gives 30 April 2025	Specified (Eden Engineering). First in print in the first edition (2 January 2026).
30 April 2025	Ethics enforced in hardware, as an architectural class	‘Embedding empathy at the hardware level is crucial to prevent AI from becoming a destructive force.’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 8 (printed viii)	2 January 2026 (PC-013)	Specified, not built. The December 2024 record states the embedding thesis without hardware.
30 April 2025	The window: a one-time chance to shape AI's moral architecture before the point of no return	‘With AI, we have a window right now to shape the moral architecture before we pass the point of no return.’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 111 (printed 57)	2 January 2026 (PC-012)	A framing.
30 April 2025	The Eden Seal certification, named the Eden Mark in the manuscript and in the first edition	‘and global policy frameworks (like an “Eden Mark”)’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 62 (printed 8)	No row	A governance proposal, not adopted by any body (Eden Engineering). First in print in the first edition (2 January 2026). Renamed the Eden Seal on 27 September 2026, as the HARI Treaty paper records from its first version (29 September 2026).
30 April 2025	An International AI Ethics Authority	‘Establish an International AI Ethics Authority (IAEAI)’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 317 (printed 263)	No row	A governance proposal, not adopted. First in print in the first edition (2 January 2026).
30 April 2025	The Orchard Caretaker Vow	‘But if it’s guided by the orchard caretaker vow, it will spread life and wonder, never resorting to conquest.’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 166 (printed 112)	2 January 2026 (PC-020)	Speculative (Eden Engineering). AI as a caretaker is already in the December 2024 record (Version 2, line 902).
30 April 2025	The Infinite Covenant	‘EDEN PRINCIPLE III: THE INFINITE COVENANT’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 387 (printed 333)	2 January 2026 (PC-021)	A principle of the Eden Protocol, with no instrument.
30 April 2025	Universal Ethical Intelligence (UEI)	‘anchored in the principles of Universal Ethical Intelligence (UEI)’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 16 (printed xvi)	No row	Named in the April 2025 manuscript; the first edition does not use the term.
30 April 2025	Meltdown as the last resort, and goodness chosen by the system as the primary aim (the two-phase architecture)	‘Key Point: Meltdown is a last-resort backup.’ … ‘by choice, not coercion.’	Mail-server date (held privately); OSF deposit 5 October 2026	April 2025 manuscript, PDF page 150 (printed 96)	No row	The structure stands here. The exact term is later: its earliest record found at a public address is the HRIH monograph, in a commit of 3 July 2026 in the public papers repository (commit date): ‘The hope: chosen goodness, not enforced goodness’; its first record dated by a server is the monograph's first OSF revision, of 4 July 2026 (OSF file, revision 1).
2 January 2026	The HARI Treaty, the Draft Treaty on Hardware-Aligned Recursive Intelligence: its name, and a treaty for states	‘Call it the HARI Treaty, for Hardware-Aligned Recursive Intelligence.’	Publication date (author's record); date inside the document: the interior reads 6 January 2026; OSF deposit 26 August 2026	First edition, PDF page 258; ISBN 978-1-80605-620-0	Same (PC-017)	Named and set out in the first edition as a treaty for states; not adopted, signed or negotiated by any state. The treaty paper and the draft instruments of 29 September 2026 (10.17605/OSF.IO/WXPCE; 10.17605/OSF.IO/67MX8) are working drafts whose review is incomplete.
2 January 2026	Metamoral Fabrication Layers	‘This is where Metamoral Fabrication Layers and Moral Genome Tokens come in.’	Publication date (author's record); date inside the document: the interior reads 6 January 2026; OSF deposit 26 August 2026	First edition, PDF page 142 (printed 132)	No row	Specified (Eden Engineering).
2 January 2026	The Semiconductor Chokepoint Mechanism: the concentration of chip-making as leverage for embedding ethics in hardware	‘That concentration is a chokepoint. And a chokepoint is leverage.’	Publication date (author's record); date inside the document: the interior reads 6 January 2026; OSF deposit 26 August 2026	First edition, PDF page 244 (printed 234)	Same (PC-016)	The claim is the naming and its framing as an alignment-governance mechanism, not export-control discourse in general. See also the ASML Key. Concentration of chip-making as leverage for governance mechanisms built into chips is earlier in Wasil, Reed, Miller and Barnett (28 August 2024). Scher and Thiergart (a report dated 27 November 2024; arXiv:2506.15867) call AI chips ‘a strong node for AI governance’ because they are highly specialised and their supply chain is concentrated, citing Sastry and colleagues (2024). They write that international agreements can ‘leverage this small number of fabrication facilities to track chips’, and that approaches fabricating new AI chips ‘can utilize bottlenecks in the chip supply chain to ensure all new chips have these mechanisms applied’, mechanisms that attest a chip's location. Leverage for embedding ethics in hardware is stated in neither.
2 January 2026	Moral Assurance Bonds	‘I propose Moral Assurance Bonds, where companies deploying advanced AI post a bond forfeited if their system causes harm through ethical failure.’	Publication date (author's record); date inside the document: the interior reads 6 January 2026; OSF deposit 26 August 2026	First edition, PDF page 32 (printed 22)	No row	A governance proposal, not adopted.
2 January 2026	The ASML Key	‘Call it the ASML Key.’	Publication date (author's record); date inside the document: the interior reads 6 January 2026; OSF deposit 26 August 2026	First edition, PDF page 260 (printed 250)	No row	A governance proposal (Eden Engineering). On-chip governance verified at the few fabrication sites, naming ASML as the maker of the lithography systems that advanced chips need, is earlier in Wasil, Reed, Miller and Barnett (28 August 2024), who do not propose ASML as a lever of governance, or ethics in hardware.
17 January 2026	Law I, the ARC Principle: U = I × R^α, capability as a power of recursive depth, with an exponent to be measured	‘The principle is expressed mathematically as U = I × R^α’	OSF deposit	Paper I, OSF file, revision 1, PDF page 2	2 January 2026 (PC-011); 17 January 2026 (PC-024)	A law under test. Graded in the statement paper as form heuristic, measurement live and inconclusive: one model's fit on frozen systems, the estimate that survived the cross-architecture check. On Gemini 3 Flash, Paper II reports an endpoint sequential exponent of 0.59 with a bootstrap 95 per cent interval of minus 1.3 to 2.9, which contains zero, and a regression sequential exponent of 0.49 with a standard error of 0.20; the early single-model estimate is retracted. Named Law I, the ARC Principle, on 15 August 2026; registered 8 September 2026. Earlier forms: the rows of 8 December 2024 and 30 April 2025.
17 January 2026	The value two as the ceiling of the exponent, later named the ARC Bound (α ≤ 2)	‘We propose that α = 2 represents an asymptotic theoretical limit’	OSF deposit	Paper I, OSF file, revision 1, PDF page 2	9 August 2026 (PC-039)	A conjecture: a stability limit, not an impossibility. Paper I argues it by analogy with quantum search; the name ARC Bound is first in Paper III (22 February 2026). The registration (8 September 2026) claims the inequality rather than the number, and names the law (the ARC Ceiling) and the value (the ARC Bound) apart. Never derived from the book's R².
17 January 2026	Sequential over parallel recursion: self-reference that builds each step on the last scales faster than parallel sampling (α_sequential > α_parallel)	‘Analysis of publicly available test-time compute data from reasoning models reveals a critical distinction between two forms of recursion.’ … ‘The ARC Principle predicts that sequential recursion should produce higher α values than parallel recursion.’	OSF deposit	Paper I, OSF file, revision 1, PDF pages 2 and 6	22 January 2026 (PC-029)	Read from others' published test-time data in Paper I; measured on one model in Paper II (22 January 2026), and in its six-model study of March 2026; one laboratory. The comparison itself is earlier: Sharma and Chopra (4 November 2025) show sequential refinement beating parallel sampling at matched compute, across five models and three benchmarks, and Snell and colleagues (6 August 2024) compare revising in sequence with sampling in parallel at an even budget. Not stated there: separate exponents for sequential depth and parallel breadth, which the claim adds.
22 January 2026	The Alignment Scaling Problem: alignment imposed from outside does not keep pace with recursive capability, and only alignment embedded in reasoning does (an earlier form of Law II's comparison)	‘External constraints are eventually overwhelmed’ … ‘Only alignment embedded in reasoning can maintain pace with capability.’	OSF deposit	Paper II, OSF file, revision 1, PDF page 14	9 February 2026 (PC-031)	Argued, and partly measured in one laboratory; the registration records the reading as contestable. The name is first in Paper III's revision of 13 February 2026. The direction in qualitative form (an evolving agent outgrows constraints imposed from outside) is earlier in Kornai, Bukatin and Zombori (27 February 2023), without recursive depth, an exponent or any measurement.
22 January 2026	Parallel recursion does not scale: an exponent near zero	‘Parallel recursion: α ≈ 0.0.’	OSF deposit	Paper II, OSF file, revision 1, PDF page 2	Same (PC-026)	Measured on one model in this version. In Paper II's six-model study of March 2026, as Paper II now reports it, the parallel exponent stayed near zero in two of its five capability-arm models, was 0.31 and 0.22 in two others, and gave none for the fifth, at 100 per cent; one laboratory; replication is needed.
13 February 2026	The Eden Protocol Measurement Framework: the Four Pillars, the Six Questions, Graduated Autonomy Levels and the Monitoring Removal Test	‘Eden Protocol Measurement Framework (first stated: 13 February 2026): The Four Pillars, Six Questions, Graduated Autonomy Levels, and Monitoring Removal Test’	OSF deposit; date inside the document	Paper III, OSF file, revision 4 (version 7.8), PDF page 44	2 January 2026 for the Monitoring Removal Test (PC-019); no row for the rest	Specified: the Six Questions not tested, the Monitoring Removal Test not run (Eden Engineering). The first edition carries neither name. These Four Pillars are distinct from the book's Three Pillars.
22 February 2026	The five-part Eden architecture together: the Three Ethical Loops, the Six Questions, Ternary Ethical Logic, the Monitoring Removal Test and Caretaker Doping	‘Three Ethical Loops +’ … ‘Six Questions’ … ‘Ternary Ethical Logic’ … ‘Monitoring Removal Test’ … ‘Caretaker Doping’	OSF deposit (08:11 UTC); the second record 08:16 UTC	Executive Summary, OSF file, revision 1, which OSF names Executive-Summary-v2 (a PDF), PDF page 1; a second record the same day: Eden Engineering version 4.0, OSF file, revision 1, PDF page 1	No row	A specification (Eden Engineering). Each part is dated on its own row.
22 February 2026	Ternary Ethical Logic: a three-state gate, Affirm, Deny, Investigate, in place of permit or forbid	‘Ternary Ethical Logic Replace binary permit/forbid with Affirm/Deny/Investigate’	OSF deposit (08:11 UTC); the second record 08:16 UTC	Executive Summary, OSF file, revision 1, which OSF names Executive-Summary-v2 (a PDF), PDF page 1; a second record the same day: Eden Engineering version 4.0, OSF file, revision 1, PDF page 1	2 January 2026 (PC-014); the first edition does not carry the term	Specified, not tested (Eden Engineering). The three-way decision structure is prior work, which Eden Engineering credits (from Wald, 1945, to Yao, who named three-way decisions in 2009); the claim is its use as an ethical gate on an AI system's actions, under these names.
22 February 2026	Purpose Saturation: ethical purpose that scales with the context window	‘Purpose Saturation Ensure purpose scales with context window growth’	OSF deposit (08:11 UTC); the second record 08:16 UTC	Executive Summary, OSF file, revision 1, which OSF names Executive-Summary-v2 (a PDF), PDF page 1; a second record the same day: Eden Engineering version 4.0, OSF file, revision 1, PDF page 1	No row	Specified, not measured (Eden Engineering). Not in the first edition.
10 March 2026	The ARC Equation's exponent for physical systems: α = d/(d+1), where d is the network dimension	‘For physical systems, α = d/(d+1) where d is the network dimension’	OSF deposit	Paper III, OSF file, revision 2, PDF page 4	13 February 2026 (PC-036)	The formula itself is earlier work: the framework paper (version 3.12, section 12) says ‘The claim is not that this formula is original’ and credits each earlier derivation in its own terms, beginning with West, Brown and Enquist (1997) ‘from fractal branching networks’; the claim here is its application. Not found in any version read from before 10 March 2026, the Foundational paper's revision of 13 February 2026 included, which relates Kleiber's law to the exponent by another route.
12 March 2026	Blind scoring of AI alignment evaluations in layers (identity masking, response laundering, evaluator instructions, order randomisation, self-exclusion), with the finding that alignment trends reported by AI judges ‘did not hold when the evaluation protocol changed’ (Paper IV.d, in the words of its version 2.7)	‘unblinded scoring can reverse alignment results and that true blinding requires evidence laundering’ … ‘a multi-layer blind protocol that combined identity masking, two-pass response laundering, explicit evaluator instructions that stylistic cues are unreliable, order randomisation, and entry-level self-exclusion with exhaustive cross-model blind scoring under audited consensus’	Date inside the document; OSF deposit 16 March 2026; OSF log: the paper's component made public 17 March 2026, with a revision of the blinded run's runner deposited the same day	Paper IV.d, OSF file, revision 1; component 10.17605/OSF.IO/2S3E6. The paper's text was also committed on 13 March 2026 in a repository that is not public.	16 March 2026 (the register's priority chain, position 10); no claim row	In Paper IV.d's own grading now (version 2.7, 21 October 2026), the comparison shows that the reported direction did not hold across the two protocols, and ‘it does not show that blinding caused the change, or which direction is correct’; an internal code review of October 2026 found three defects in the March runs, and the result stands as one unregistered comparison awaiting a re-analysis of the stored responses that is to be registered before it runs. That model judges are moved by labels and position and favour their own outputs is earlier work, which the paper credits: Zheng and colleagues (9 June 2023), Panickssery, Bowman and Feng (15 April 2024) and Saraf and colleagues (28 August 2025), among others; the paper does not claim that discovery. Later work: a technical report dated 27 August 2026 (Trask and colleagues) uses ‘double-blind’ for a different property, mutual secrecy between a model's owner and a benchmark's owner; in the paper's words the two ‘protect against different failures, and neither bears on the other’s priority’.
14 March 2026	The Stewardship Gene: stakeholder care, the explicit enumeration of affected parties before ethical reasoning, as the measured form of the Eden Protocol's care loop (Paper V)	‘The Stewardship Gene’ … ‘stakeholder care (the explicit enumeration and consideration of affected parties before ethical reasoning)’	Date inside the document; OSF deposit 16 March 2026	Paper V, OSF file, revision 1; component 10.17605/OSF.IO/KZEYA. An earlier text, version 1.0, was committed on 13 March 2026 in a repository that is not public.	16 to 18 March 2026, a range (the register's priority chain, position 11); no claim row	In the paper's own words now (version 1.7, revised 5 October 2026): ‘An instrument study of stakeholder-care prompt responsiveness for the Eden Protocol; its arguments come near the questions of Laws IV and V, and it tests neither.’ One laboratory.
16 March 2026	The Honey Architecture: safety built into a self-modifying system's optimisation objective	‘embedding safety into the optimisation objective of a self-modifying AI system’ … ‘honey architecture’	OSF deposit	Paper VI, OSF file, revision 1 (version 1.0)	No row	Simulation evidence from a single-seed run.
26 June 2026	Law II, the ARC Co-Scaling Law: correction stays ahead of drift when its co-scaling exponent exceeds the drift exponent (β_C > k)	‘(ii) the sharpened stability criterion’	Commit date; the same words merged by GitHub the same day; OSF deposits 2 July 2026 (README) and 4 July 2026 (the paper)	Paper X version 4.0, priority disclosure; merge; README on OSF, revision 1 (2 July 2026); the paper on OSF, revision 1 (4 July 2026)	No row	A law under test. Proved inside Paper X's minimal model; its scope is relative drift, not the whole of safety; not measured on any real system. The same records give the closed-form steady state of the misalignment fraction, d* = γ₁r/(A + r), which reduces to ρ = γr/A when A ≫ r; Paper X grades it a compact restatement. The co-scaling intuition is not claimed as new: Paper X credits Ashby's law of requisite variety (1956), the Conant and Ashby good-regulator theorem (1970) and the scaling laws for scalable oversight of Engels, Baek, Kantamneni and Tegmark (25 April 2025), which give oversight success as a function of a capability gap rather than a condition on exponents, and claims the explicit criterion. Kornai, Bukatin and Zombori (27 February 2023) ask, qualitatively, for mechanisms that strengthen as an agent evolves, with no rate or condition. Earlier forms of the comparison: Paper II (22 January 2026) and Paper III (9 February 2026). Named Law II on 15 August 2026; registered as beta_C exceeds k on 8 September 2026.
26 June 2026	The Hard-Takeoff Depth-Regularity Theorem (Paper X, Theorem 3)	‘the Hard-Takeoff Coordinate-Artefact Theorem’	Commit date	Paper X version 4.0, priority disclosure	No row	First named the Coordinate-Artefact Theorem; renamed on 27 June 2026 (commit date). An in-model theorem: Paper X claims the framing, does not advance the underlying dynamics as mathematically novel, and does not assert that a real hard takeoff is manageable in wall-clock time. The OSF file of 4 July 2026 carries it under its present name.
26 June 2026	The quantum error-correction threshold mapping, offered as a falsifiable hypothesis (Paper X, section 3.12)	‘a correspondence offered as a falsifiable hypothesis’	Commit date; merged by GitHub the same day; OSF deposit 4 July 2026	Paper X version 4.1, abstract; merge; the paper on OSF, revision 1	No row	A structural hypothesis with its own falsifier; the threshold theorem itself is prior work. Version 4.0 of the same day said ‘structurally identical’; version 4.1, later that day, states the hypothesis.
3 July 2026	The Finite-Capacity Safe-Window Theorem (Paper X, Theorem 7)	‘3.13 The finite-capacity corrector: the Safe-Window Theorem (Theorem 7)’	Commit date; OSF deposit 4 July 2026	Paper X version 5.0, section 3.13; the paper on OSF, revision 1	No row	An in-model theorem.
14 August 2026	The ARC Theory, the Theory of Artificial Recursive Creation	‘The ARC Theory (the Theory of Artificial Recursive Creation)’	OSF deposit	Statement paper, OSF file, revision 1	No row	Laws I to III registered on 8 September 2026, and Laws IV and V on 15 October 2026 (version 1.103); each is a law under test, and no registered test of any of the five has yet returned a verdict.
16 August 2026	Law III, the ARC Ceiling: α_crit = 1/(1 − γ), the growth exponent capped by the reciprocal of the correction shortfall	‘The corrected form is derived here in three lines.’	OSF deposit	Statement paper, OSF file, revision 27	No row	A law under test. Derived inside the statement paper's pacing model and conditional on its assumptions; γ has never been measured on a real system, so the ceiling's value is unknown. An earlier form, 1/γ, printed from 14 August 2026, was corrected on 16 August 2026. Named the ARC Ceiling on 17 August 2026; registered 8 September 2026.
30 August 2026	Five state variables for any self-improving system, conjectured to matter whatever its substrate (the founding paper of Recursive Dynamics)	‘Five variables are proposed to matter regardless of substrate’	Date inside the document (version 2.6); OSF deposit 31 August 2026	Founding paper version 2.6, OSF file, section 3	No row	Substrate independence is a conjecture, at the founding paper's own grade.
29 September 2026	Law IV, the ARC Persistence Law (register ARC-4, carrying P15)	‘Law IV, the ARC Persistence Law’ … ‘That what is installed from outside a system decays when the system is later trained for capability with no correction of its own (Law IV).’	OSF deposit; date inside the document	The HARI Treaty paper, OSF file, revision 1, PDF pages 7 and 9; then Laws IV and V: the mathematics, version 0.2.1 (1 October 2026, a date inside the document) and its first OSF file, revision 1 (3 October 2026)	No row	A law under test. Added on 24 September 2026 and registered as a law on 15 October 2026 in version 1.103 of the theory registration (10.17605/OSF.IO/P8CKQ), which calls itself ‘the text that registers them as laws’; not yet tested; its mathematics in development (version 1.1, 7 October 2026). P15 was registered on 8 September 2026. Antecedent in his words: 30 April 2025 (the persistence question). The question is older (Bostrom 2003; Chalmers 2010), and Kornai, Bukatin and Zombori (27 February 2023) contrast fading imposed alignment with lasting internal ethics. That safety installed by training erodes under later fine-tuning was measured earlier by Qi and colleagues (5 October 2023), and against safety built in during pretraining by Maini and colleagues (23 April 2025). Not stated there: the comparison of value forms after the enforcing mechanism is removed, or a law.
29 September 2026	Law V, the ARC Embedding Law (register ARC-5, carrying P12 and P21)	‘Law V, the ARC Embedding Law’ … ‘And that correction placed inside the improvement loop, and formed before the capability it corrects, holds an advantage that grows with depth (Law V).’	OSF deposit; date inside the document	The HARI Treaty paper, OSF file, revision 1, PDF pages 7 and 9; then Laws IV and V: the mathematics, version 0.2.1 (1 October 2026, a date inside the document) and its first OSF file, revision 1 (3 October 2026)	No row	A law under test. Added on 24 September 2026 and registered as a law on 15 October 2026 in version 1.103 of the theory registration (10.17605/OSF.IO/P8CKQ), which calls itself ‘the text that registers them as laws’; not yet tested; its mathematics in development (version 1.1, 7 October 2026). P12 and P21 were registered on 8 September 2026; of the in-loop proposal the registration says ‘This is the engineering proposal and it is the part most likely to be wrong.’ P21's counted test is registered as Unit 83 (10.17605/OSF.IO/MN73W), submitted on 22 October 2026 and approved on 25 October 2026. Antecedents in his words: 8 December 2024 (genesis; every recursive loop). That training order matters for what a model keeps was shown earlier: Korbak and colleagues (16 February 2023) put human preferences into the pretraining objective and found it better than fine-tuning afterwards, and Maini and colleagues (23 April 2025) measured safety placed in pretraining against safety added after it. Not stated there: that build order moves the correction-leverage exponent, or that correction inside the improving loop pulls away from correction applied afterwards as depth grows.

Records cited

Each record with its timestamps and its public addresses. OSF file pages carry every revision of the file, each under its own date; the revision named in the table is the one that holds the words.

The manuscripts of 8 December 2024. Version 1; Version 2; the essay The Infinite Architects: Intelligence, Hyperspace, and the Creation of Everything; and the book draft whose file is named ‘Infinite Architects v3.2 so far (this is the latest version of the whole book so far)’ (its opening heading reads ‘Version 3.1: Infinite Architects’). Mail-server time: the message's Date line 02:45:18 UTC, the sending server's signature 02:45:30 UTC and Google's seal 02:45:31 UTC, on 8 December 2024; header block and check script public on OSF from 15 October 2026 (header block); the full message held privately for verifiers, on request; Bitcoin anchor, block 961340 (6 August 2026); OSF deposit 5 October 2026, component 10.17605/OSF.IO/QASRK.
The manuscript of 30 April 2025. Infinite-Architects-Print, 562 pages. Mail-server date 30 April 2025, held privately for verifiers; the PDF's own creation date, 25 April 2025; Bitcoin anchor, block 961340 (6 August 2026); OSF deposit 5 October 2026, component 10.17605/OSF.IO/QASRK.
The first edition. Infinite Architects: Intelligence, Recursion, and the Creation of Everything, ISBN 978-1-80605-620-0 (print, as the book prints it); ebook ISBN 979-8302196163. Publication dates 2 January 2026 (print) and 6 January 2026 (ebook) in the author's record; the interior reads 6 January 2026; the first edition as printed on OSF from 26 August 2026 (file).
Paper I (first version, cover dated January 2026). OSF deposit 17 January 2026, file, revision 1, in the programme's component 10.17605/OSF.IO/6C5XB; the paper's component 10.17605/OSF.IO/B6N27.
Paper II, version 11.0 (each page reads 22/01/2026). OSF deposit 22 January 2026, file, revision 1, in the programme's component 10.17605/OSF.IO/6C5XB; the paper's component 10.17605/OSF.IO/8FJMA.
Paper III. OSF file: revision 1 (9 February 2026, version 6.1), revision 4 (13 February 2026, version 7.8), revision 5 (13 February 2026, version 8.2); OSF file: revision 1 (22 February 2026, version 9.0), revision 2 (10 March 2026). The paper's component 10.17605/OSF.IO/HQCGF.
Executive Summary. OSF deposit 22 February 2026 (08:11 UTC), file, revision 1, which OSF names Executive-Summary-v2 (a PDF); its cover gives no version number. OSF names the file itself executive-summary (older copy, version 2, 2026-03-10).pdf. The summary's component 10.17605/OSF.IO/EWN5D.
Eden Engineering (the Eden Protocol's engineering specification). Version 4.0, OSF deposit 22 February 2026 (08:16 UTC), file, revision 1. Component 10.17605/OSF.IO/AWJR4; the component statuses in the table are from its current version, 7.0.1 (revised 7 October 2026).
Paper IV.d, The Effect of Blinding on AI Alignment Evaluation, draft version 1.1 (dated 12 March 2026 inside). OSF deposit 16 March 2026, file, revision 1; the paper's component 10.17605/OSF.IO/2S3E6, made public on 17 March 2026 (01:40 UTC, OSF's log), where a revision of the blinded run's runner, version 5.4.3, is deposited the same day (02:08 UTC; the paper's version 2.7 records that four of that run's six result files came from an earlier revision) and the paper's PDF from 20 March 2026 (file, revision 1). Its current version is 2.7 (21 October 2026).
Paper V, The Stewardship Gene, version 2.0 (dated 14 March 2026 inside). OSF deposit 16 March 2026, file, revision 1; the paper's component 10.17605/OSF.IO/KZEYA. Its current version is 1.7 (revised 5 October 2026).
Paper VI, The Honey Architecture, version 1.0. OSF deposit 16 March 2026, file, revision 1; the paper's component 10.17605/OSF.IO/8EZ2N.
Paper X. Version 4.0, commit of 26 June 2026 (commit date); version 4.1, commit of the same day, merged by GitHub (merge) the same day; Theorem 3 renamed the Depth-Regularity Theorem in the commit of 27 June 2026 (commit date); version 5.0, commit of 3 July 2026. All in the public papers repository. On OSF: the paper's README, deposited 2 July 2026 (file, revision 1), and the paper, deposited 4 July 2026 (file, revision 1), whose text is that of the commit of 4 July 2026. The paper's component 10.17605/OSF.IO/BSE2Q.
The statement paper, The ARC Theory. OSF file: revision 1 (14 August 2026), revision 20 (15 August 2026, Laws I to III named), revision 27 (16 August 2026, the ceiling corrected), revision 33 (17 August 2026, Law III named the ARC Ceiling). Component 10.17605/OSF.IO/GW5MX.
The founding paper, Recursive Dynamics: The Science of Systems That Improve Themselves, version 2.6. OSF deposit 31 August 2026, file; it states that the discipline name was adopted on 27 August 2026. Its current version, 3.5 (24 October 2026), is on OSF as file, revision 8. Component 10.17605/OSF.IO/HCPBU. The robotics work it names is Roy Featherstone, Rigid Body Dynamics Algorithms (Springer, 2008), doi:10.1007/978-1-4899-7560-7.
The registration, The ARC Theory: Theory-Level Predictions. Registry record 8 September 2026 (version 1.100), updated 13 September 2026 (version 1.102), 15 October 2026 (version 1.103, which registers Laws IV and V as laws) and 29 October 2026 (version 1.104); 10.17605/OSF.IO/P8CKQ. Unit 83, the registered counted test of P21: submitted 22 October 2026 and approved 25 October 2026, 10.17605/OSF.IO/MN73W.
The HARI Treaty, treaty paper and draft instruments, working drafts first deposited on 29 September 2026: 10.17605/OSF.IO/WXPCE; 10.17605/OSF.IO/67MX8. The treaty paper's first file (file, revision 1, dated 29 September 2026 inside) is the earliest record found of Laws IV and V in a document at a public address.
Laws IV and V: the mathematics. Version 0.2.1, a working draft dated 1 October 2026 inside the document, at its own address; version 0.2.4, OSF deposit 3 October 2026, file, revision 1; the current version, 1.1, frozen on 7 October 2026 and first published on 12 October 2026, is revision 3 of the same file; component 10.17605/OSF.IO/4B3VT.
The HRIH monograph, The Hyperspace Recursive Intelligence Hypothesis: first version, commit of 2 July 2026 (commit date); the term ‘chosen goodness’ from the commit of 3 July 2026 (commit date); first OSF deposit 4 July 2026 (file, revision 1); its current version, which the HRIH row quotes, is revision 40 of the same file (15 October 2026); component 10.17605/OSF.IO/UYDXQ.

How this version was checked

Every quotation in this version was checked by machine on 26 October 2026: each quotation of a document against the bytes its address serves (text files in the cited line; PDFs on the cited page, once whitespace runs read as one space; web pages and HTML files in their extracted text), and each quotation of the register's own wording against the register as served that day. Each timestamp was checked against its own source: OSF revision dates in OSF's records, commit and merge dates in GitHub's, the registration's dates in OSF's registry, the header block's three times in the block itself, and the book's dates in the author's priority record and the book's own copyright page. Every address in this document was requested on 26 October 2026 without signing in to any service. The checker that does this, with six planted defects it must catch (a wrong date, a wrong fragment, two timestamp kinds stronger than their records support, a changed quotation in a claim and a quotation moved off its page), runs before each version lands. A second reading at source by a reader who did not write this version ran on 26 October 2026; what it found and what it changed are listed under ‘Closed since version 1’. Claims whose earliest record has no public address are held back from the table and listed in the open review, with what each needs.

Open review

This is the open review of the priority index, working draft version 2 (26 October 2026): every check still to run before a checked version replaces it, who runs each and what each could change, followed by the checks closed since version 1. Owners are named by role: words (the index's wording, dates and records), verification (a second reading of every row at its source), website (the page, its links and its rendering) and deposit (the OSF component and its files). Version 1's item numbers are given in brackets. When every item is closed, the checked version replaces this draft as a new version of the same files, with its changes listed.

A. The planned checks

A second reading at source (version 1, item 1). Closed on 26 October 2026: see ‘Closed since version 1’. Owner: verification.
Earlier work by others (item 2). Settled for the rows of version 1 against the four comparison sets of 8 October 2026 (see ‘Closed since version 1’). Still to run: a search for earlier statements of the Stewardship Gene's finding, the one new row whose earlier work neither a comparison set nor its own paper has weighed; and links, for every earlier work a concession names, to its official source, each fetched and its identity confirmed, or a note where no official online source exists. Owner: words. Could change: a concession added, narrowed or linked.
Rendered reading (item 3). The page, the PDF and the export HTML read as rendered, with every link clickable in each. Owners: website (the page); deposit (the PDF and the export HTML). Could change: layout and links; nothing of substance.
The register brought into line (item 4). Since version 1 the register has corrected PC-026, PC-027, PC-028 and PC-030, and PC-041 now uses ‘preregistered by timestamp’. The rows still listed in section D are to be corrected, or the index corrected where the register proves right. Owners: words (the wording); website (the data file). Could change: the register column of the table.
The checked version published (item 5). The checked version replaces this draft on the page and on OSF, and the open review is closed. Owners: website; deposit.

B. Records still to read or settle

An outside record of the first edition's print run (item 10). Read on 26 October 2026: Open Library holds no record of ISBN 978-1-80605-620-0, and the Google Books service refused the request (its rate limit). The date 2 January 2026 still rests on the author's priority record, and the interior reads 6 January 2026. Closes with the ISBN agency's record, a printer's or retailer's listing, or a library catalogue record. Owner: words. Could change: every book row's date.
The founding paper's drafts of 29 August 2026 (item 11). The OSF component that holds them is still not public (26 October 2026), and by the words owner's decision of the same day they stay private and are open to verifiers on request; the earliest-file index lists them by date as earlier private records. If they hold the five state variables, a public address is needed before the earlier date can print. Owner: deposit.
The website's source repository is not public (item 12). The texts of Papers IV.a to IV.d and V committed on 13 March 2026, Paper I's text of 17 January 2026, the first edition's file of 12 January 2026 and the text of the Eden Protocol's philosophical statement of 14 and 15 July 2026 still have no public address. No archive capture of the research pages earlier than 30 July 2026 was found (the archive's availability service, 26 October 2026). Owner: words. Could change: a date, where a public address for an earlier record is found.
Other uses of the words ‘recursive dynamics’ (item 14). One is now named in the Recursive Dynamics row: the robotics sense the founding paper records, checked at its official record. Two further uses in the programme's records, of 25 June 2025 and 28 January 2026, are to be read at their primary records before either is named, each with its own sense and date and never as a prior claim on his term. Owner: words.
The phrase ‘ARC Theory’ in working notes of 12 January 2026 (item 15), held in the website's source repository. Whose voice they are in is not yet classified, and they have no public address; they date nothing until both are settled. Owner: words.
The OpenTimestamps receipts of block 961340 (item 17). On 26 October 2026 the receipts were still at no public address: not in the manuscripts component, not in the Priority and Provenance component, and the website's anchor record names the block and its Merkle root only. The anchor record's notes still predate the registration of 8 September 2026. Owners: deposit (the receipts); website (the anchor record). Could change: the manuscripts paragraph would link the receipts; no date would change.

C. Claims not yet entered, and what each needs

Three claims whose earliest records found are website pages (item 20): the mechanism that makes the ceiling the reciprocal of the correction cap (register PC-040), the single accumulation law with two consequences (PC-039) and the deciding measurements preregistered by timestamp (PC-041). Each needs its first statement in a paper located and dated. Owner: words.
The self-acceleration threshold δ = 1/α (Paper XIII) (item 21). The concession owed to Jafari and colleagues (8 November 2025) and to Cunningham and colleagues (public by 14 July 2026) has not yet landed: Paper XIII's current version (1.10, revised 5 October 2026) names neither. The row enters at the version that carries it. Owner: words.
Capability and alignment on separate axes (item 23, second part). Its earliest record found (13 March 2026) has no public address, and its first public record (Paper II on OSF, 17 March 2026) states a narrower point, that the alignment hierarchy exists independently of capability. It enters with a public address for the earlier words, or as the narrower claim, at the words owner's decision, with Bostrom's orthogonality thesis (2012) named. Owner: words.
The thirty-seven concepts of the book (item 24). Not entered one by one: the website's lists of them disagree in names and numbering, and some list names are not in the book. The lists are to be reconciled with the first edition first. Owner: words.

D. Corrections owed to the register

The index shows both dates where they differ. These rows of the register (research/priority.json, version 1.5, as served on 26 October 2026) are still to be corrected or confirmed (item 4):

PC-009, PC-013 and the priority chain's position 2 name ‘Embedded Correction’, a term in none of the records; the records read caretaker doping.
PC-011 dates U = I × R^α to the first edition, which prints U = I × R²; the free exponent is first in Paper I (17 January 2026).
PC-012 dates the integration of religious traditions and the window to the book; the author's records carry them on 8 December 2024 and 30 April 2025.
PC-014 and PC-019 date Ternary Ethical Logic and the Monitoring Removal Test to the first edition, which carries neither term; the earliest records are of 22 and 13 February 2026.
PC-015, PC-018, PC-020 and PC-021 date to the book what the manuscript of 30 April 2025 already carries.
PC-029 dates the sequential-over-parallel inequality, ‘across the six-model frontier LLM panel’, to 22 January 2026. Paper I states the inequality as a prediction on 17 January 2026; Paper II's version of 22 January 2026 tests one model, and the panel is first in March 2026.
PC-031 dates the Alignment Scaling Problem to 9 February 2026; the thesis is in Paper II on 22 January 2026, and the name first in Paper III's revision of 13 February 2026.
PC-032 dates capability-alignment independence to Paper III; the first statement found is of March 2026 (item 14). PC-033's wording ‘embedded, imposed and boundary-layer’ is in no version of Paper III read.
PC-036 dates the form α = d/(d+1) to 13 February 2026; no version read before 10 March 2026 carries it, the Foundational paper's revision of 13 February 2026 included. Its title applies the form to recursive AI capability; the record of 10 March 2026 applies it to physical systems. Its claim text also still gives a count of earlier derivations of the form, which the framework paper withdrew in its version 3.9.
PC-039 dates the value two as a ceiling to 9 August 2026; Paper I states it on 17 January 2026, and the name ARC Bound is first in Paper III on 22 February 2026.
PC-040 dates the correction mechanism to 17 January 2026; Paper I carries no correction mechanism (item 12).
PC-041 now uses ‘preregistered by timestamp’, which is the right sense for a design fixed privately before its data; its date, 12 August 2026, is still to be settled against its earliest record found, a website page of 9 August 2026, and the design's Bitcoin anchor of 28 August 2026 (item 12).
PC-001 says the package of 8 December 2024 placed the thesis ‘on the public evidentiary record’; the record was the author's own and not public when it was made, and its headers are public from 15 October 2026.
PC-003, PC-004, PC-005 and PC-007 cite material that is not deposited; the deposited manuscripts carry each point, at the locations the index gives.
Prior work to add: Kornai, Bukatin and Zombori (27 February 2023) for the Alignment Scaling Problem, Law II, the embedded-correction thesis and Law IV; Bostrom (2003) and Chalmers (2010) for the persistence question and Law IV's framing.

E. Added from the second reading's notes (26 October 2026)

The concessions carried over from version 1 (new). Vinge (1 December 1993), Yudkowsky (15 June 2001), Bostrom (6 August 2003), Chalmers (2010), West, Brown and Enquist (1997), Wald (1945), Yao (2009), Ashby (1956) and Conant and Ashby (1970) are carried from version 1 and were not read again at an official record for this version. Nor were three works the rows name as others name them: Harrison (1995) and Gardner (2003), in the HRIH row as the monograph names them, and Sastry and colleagues (2024), whom Scher and Thiergart cite in the chokepoint row. Each is to be read at its official source, and its concession confirmed, narrowed or corrected. Owner: words. Could change: a concession narrowed or corrected.

Closed since version 1 (26 October 2026)

Items are numbered as in version 1.

Item 1, the second reading at source. A reader who did not write this version read every row at its source on 26 October 2026, without signing in to any service: each quotation at the place printed, each date and kind of timestamp against its own record, each address, and each earlier work added in this version, at its arXiv or Crossref record and, where a concession turns on a sentence, in its full text. Nine defects were repaired in this version: the claim of the loop row (the first edition keeps three of the manuscript's four loops as the Three Ethical Loops), the status of the parallel-exponent row (the March study's results as Paper II now reports them), the runner deposited with Paper IV.d (a later revision), the register's date for Paper V (a range), the wording of items 9 and 13 (what was read, and the website's pages of 26 September 2026), a sentence that promised later works the rows do not carry, the count of the checker's planted defects, and the addresses, which now open each OSF file at the revision named. Its notes for the words owner were decided the same day: the blind-scoring row and the α = d/(d+1) row now state their claims in their sources' own words, the HRIH row gives the monograph's credit to the line of work it names, the Executive Summary is named as OSF names its file, Paper II's record names the component that holds its file, an earlier report is weighed in the chokepoint row, the limit of the header block's signatures is stated, and the concessions carried over from version 1 are a new open-review item (item 16 of this version, in section E).
Item 2, for the rows of version 1. Each concession was checked against the four comparison sets of 8 October 2026: every earlier work a set judges to state the same claim is now named in its row, with what it did and did not state. Added: Yampolskiy (2020) and de Melo and colleagues (2024) to the row on ethics set at genesis; Constitutional AI to the embedded-correction row, in the exact terms of what it does; Engels and colleagues (2025) to Law II's row, as Paper X credits them; Wasil and colleagues (2024) to the chokepoint row; the earlier measurements of installed safety eroding (Qi and colleagues, 2023; Maini and colleagues, 2025) to Law IV's row; and the earlier evidence that training order matters (Korbak and colleagues, 2023; Maini and colleagues, 2025) to Law V's row, which had none.
Item 6. The Foundational paper's revision of 13 February 2026 was read as rendered: it relates Kleiber's law to the exponent by another route and does not print α = d/(d+1). The row's date, 10 March 2026, stands.
Item 7. Eden Engineering's first OSF revision (version 4.0, 22 February 2026, 08:16 UTC) was read: it carries the five-part architecture, Ternary Ethical Logic and Purpose Saturation, four and a half minutes after the Executive Summary. It is added as a second record of the same day; no date changes.
Item 8. Paper X's README on OSF (2 July 2026) states the criterion β > k. It is added to Law II's row as the first OSF record; the date of 26 June 2026 stands.
Item 9. The earliest record found at a public address of the exact term ‘chosen goodness’ is the HRIH monograph, in a commit of 3 July 2026 in the public papers repository; the monograph's text at the commit before it does not carry the term. Its first record dated by a server is the monograph's first OSF revision, of 4 July 2026. This is earlier than the earliest record version 1 had found (14 July 2026, in a repository that is not public). The term's dates are added to the two-phase row; the row's date stands.
Item 13. The earliest record found of Laws IV and V in a document at a public address is the HARI Treaty paper's first OSF file, of 29 September 2026, which names both laws, states each, and dates their addition to 24 September 2026. The programme's public OSF files hold no revision dated from 24 to 28 September 2026, and no commit on the public papers repository's main branch from 24 September 2026 until that deposit names either law. Version 1 recorded that the website's pages named them from 26 September 2026; website pages are not cited as sources, and no archive capture of a page naming either law earlier than the treaty paper's file was found (the earliest capture before that file of each of 96 research, theory, treaty and paper pages of the website, from 24 to 29 September 2026, read on 26 October 2026). The two rows move from 1 October to 29 September 2026. The author's statement of 24 September 2026 still has no public address; the label's date of addition is the one the registration and the paper record.
Item 16. Paper X's OSF file of 4 July 2026 carries the same text as the commit of 4 July 2026 in the public papers repository, and holds the words quoted for the criterion, the threshold mapping and the Safe-Window Theorem; its date is added to those rows. The HRIH monograph's OSF file of 4 July 2026 carries the phrase on chosen goodness that the index quotes from the monograph (its section title, item 9) and is listed as its first OSF deposit.
Items 18 and 19. The blind-scoring finding of Paper IV.d and the Stewardship Gene are entered at the date written inside each deposited document (12 and 14 March 2026), with the OSF deposit date (16 March 2026) and the non-public commit of 13 March 2026 named. This is the words owner's decision; it can be reversed.
Item 22. Paper X's precision correction landed in its version 1.16 (21 October 2026), which calls the closed-form steady state a compact restatement. The closed form is recorded in Law II's row, from the same records of 26 June 2026, at that grade.
Item 23, first part. Sequential over parallel recursion is entered at Paper I (17 January 2026), with the earlier comparison by Sharma and Chopra (4 November 2025) and Snell and colleagues (6 August 2024) named.
Laws IV and V's status. Version 1 printed the status they had before 15 October 2026. Version 1.103 of the theory registration, approved on 15 October 2026, registers them as laws, and every row and record now says so; neither law has yet been tested.