Which instrument carries which date, stated first: the sender-DKIM and ARC mathematics carry the December 2024 date, verifying today against Google’s published keys; the OpenTimestamps receipts in Bitcoin block 961,340 freeze the bytes from 6 August 2026 onward; together they say these bytes existed, in that order, on those dates, checkable without asking anyone. Existence-by-date anchors, not authorship proofs, and every entry below says which it is.
Take a bicycle apart and lay the pieces on a table. Wheel, chain, gear, frame, brake. Not one was invented by whoever built the first bicycle, and every one had existed for years before anybody rode anything. The invention was the arrangement, and the bicycle is a more substantial thing than the parts lying on the table.
That is what is claimed here, and it is claimed without apology. The synthesis: the specific arrangement, decomposed by the review into six elements and by the current statement into five components, that no located prior work joins. Every component has prior work, this ledger names it, and naming it is what makes the remaining claim worth reading. An arrangement is not the small residue left over once the borrowed parts are subtracted. It is the thing that does work none of the parts does alone. Science has always worked this way: Newton used nobody’s new components to unify terrestrial and celestial motion, and Maxwell used nobody’s new components to unify electricity, magnetism and light. The question asked of a contribution is not whether every part is new. It is what the combination does that the parts do not.
Forty-one dated entries, kept so a claim can be checked rather than believed. The synthesis argument itself, the six elements as the review decomposed them (Good’s intelligence-explosion argument, Bostrom’s motivation-selection case, Schwitzgebel and Garza’s duties to created minds, ordinary recursion, ordinary stewardship, and the operationalisation into a scheduling constraint) and their joins, is stated at the synthesis page. The prior work each element rests on is set out at related work. Alongside the arrangement sit three smaller and equally checkable claims: the operationalisation, the dated names, and the record itself. The block below states it exactly. Each entry states what was recorded and when, with a link to the primary source and an honest note on what the claim does not support. Authority for public wording now belongs to the priority and claims ledger; PC identifiers are preserved here as append-only aliases for old citations.
Both are stated the same way, because they are the same kind of claim: named parts, a stated join, the work the join does that no part does alone, and the single finding that would end it.
One · the alignment synthesis
The parts, each predated and each named. The intelligence-explosion argument is I. J. Good, 1965. The control problem, and the case for motivation selection over capability control including value loading before takeoff rather than after, is Bostrom, Superintelligence, 2014, chapters 8 to 13. Duties toward created minds are Schwitzgebel and Garza, 2015. Recursion is ordinary mathematics. Stewardship is millennia of religious tradition.
The join. Six elements that no located prior work joins.
What the join does that the parts do not. It adds nothing to the conclusion. It supplies the reason the next step is forced rather than chosen, and converts an ethical preference into a scheduling constraint: if amplification is indifferent to content, the moment of seeding is the only moment with leverage.
What would end it. One document predating 8 December 2024 containing the control-failure thesis, the embedded-alignment solution and the recursion equation together.
Two · the convergence of the scaling derivations · argument stated, evidence at pilot stage
The parts, each predated and each named. Cauchy’s multiplicative functional equation, 1821. The derivation of the exponent d/(d+1) from hierarchical space-filling vascular networks: West, Brown and Enquist, Science 276, 122, 1997. The derivation of the same exponent from quantum oscillator coupling in the Debye model: Demetrius, and Demetrius and Tuszynski, 2003 to 2010. Space-filling geometry in d dimensions, and a conservation constraint on resource flow. All standard.
The join. Multiplicative composition, d-dimensional space-filling and a conservation constraint together compel d/(d+1). The two derivations were never rivals. They are two physical instantiations of one mathematical constraint.
What the join does that the parts do not. Cauchy alone says nothing about biology. The network model alone says nothing about cosmology. The oscillator model alone says nothing about branching geometry. Joined, they make the exponent a mathematical necessity rather than a property of any mechanism, which is a stronger result than either derivation can reach alone. It predicts with zero free parameters across the dimensional ladder: d = 1 gives one half, the Friedmann radiation era, exact; d = 2 gives two thirds, the matter era, exact; d = 3 gives three quarters, mammalian basal metabolic rate; d = 4 predicts 0.800 against Weibel 2004’s measured 0.872, interval 0.813 to 0.932, which is a miss and is recorded as one; d to infinity gives 1.
The state of the evidence, which is not the same object as the argument. This is a pilot. Under AICc model selection, 19 of 25 empirical domains fit the predicted family, binomial against one third, p = 1.56 × 10−5. That is a structured prediction comparison run by one operator. It is not a pre-registered trial and it is not a confirmation. The argument above is either right or wrong on inspection; the pilot only says that a first look did not contradict it. A confirmatory test with a frozen domain list has been run, and it was run against predictions that were public before it. On 17 March 2026 at 00:19:19 UTC a twelve-row manifest was committed to the public repository, each row fixing an operator class, a predicted family and a predicted model; the fits ran twenty-three minutes later at 00:42:41 and the results were committed at 00:43:25. Ten of the twelve returned the predicted family, binomial against one third, p = 5.4 × 10−4. The rung it reaches, stated in full: the twelve classifications were made by the author, not by an external operator; the candidate set was the pre-correction one, with no logarithmic family; the attestation is a public repository commit date, corroborated by the deposit of the same manifest to the public OSF component at 02:06:55 that night, and the outcome file is not among the files deposited there; and one of the twelve, receptor occupancy, was run on Hill’s 1910 haemoglobin data, repeating a domain, an operator class and a predicted model that were already public with a matching outcome twenty-four minutes before the manifest, so the strictly advance-timestamped set reads nine of eleven, p = 1.4 × 10−3, without it, and eight of ten, p = 3.4 × 10−3, if enzyme kinetics is removed with it. What has not been run is the version that closes the remaining gaps together: an external operator classifying to the rule, the corrected four-family candidate set, and a registry timestamp. Until that runs, this arrangement is a stated claim with dated support at the author-classified rung and nothing more.
The frame for that test is already dated and public, which is the part worth knowing. On 17 March 2026, between 02:06:51 and 02:07:05 UTC, a prediction protocol, an extension manifest and a checksum file were deposited to the public Open Science Framework component x6wa7. Every one carries an OSF-assigned creation timestamp and a modification timestamp identical to it, and none has been altered since. The outcomes for the twelve domains they specify were computed earlier the same night, at 00:42:41 UTC, eighty-four minutes before the deposit, and committed to the public repository at 00:43:25; the results file is not among the files deposited to the component, which holds the predictions, the protocol, the checksums and the extracted data. The ordering that carries the weight was therefore fixed in the repository and not by the deposit: the manifest was committed at 00:19:19, twenty-three minutes before the fits ran. What the deposit adds is a third-party record of the same predictions, unaltered since, against which the outstanding external test can still be run. That is a dated third-party deposit and not an OSF Registration, which is a separate frozen and schema-bound object this component does not carry, and the API reports registration: false. It does not make the pilot a result. It fixes the ordering, which is the one thing about a confirmatory test that cannot be established after the fact.
What would end it. One prior work stating the argument from multiplicative composition, space-filling and conservation, or explaining the convergence of the two derivations by any other means.
The other limit. The precedence search behind this arrangement covers the alignment and philosophy literature; the corresponding search across the metabolic-scaling and mathematical-physics literature has not been run, and until it is, the novelty of this join is asserted on the author’s reading rather than on a completed search.
Neither claim is a claim to have invented a component. Both are claims about an arrangement, which is what a synthesis is in science and what a bicycle is in engineering.
8 DEC 202402:45 UTC
What this ledger claims, before the caveats. 41 individually dated rows, each with evidence pointers and a stated boundary, and the claim in every one of them is a formulation and the date it was first stated in public, never the underlying mathematics. The priority anchor. The earliest dated artefact is a manuscript package self-emailed on 8 December 2024 at 02:45:18 UTC, Message-ID CAGPsKAnp-DLDBOeBAh4t9XsqJBKdoe07uORGF86meoSMFnin7w. It carries five attachments (three manuscript and research drafts, the 31,881-word book draft v3.2, and a 154,318-word human-AI working dialogue; 221,236 whitespace-delimited words in aggregate, including duplicated draft material) and a SHA-256 anchor prefix of f0d1f38f.
Stated precisely, as of the 2 August 2026 audit: the sent-side bytes are public and hash-verifiable, and a supplied recipient-form export decodes to the same five attachment hashes. Its sender-domain DKIM and paired Google ARC signature mathematics verify against the keys returned at the signed selectors on 2 August 2026, and the ARC selector belongs to Google rather than to me, which is the part I could not have arranged. That is provider-key evidence over a canonicalised signed scope. It is not an RFC 3161 trusted timestamp, it does not authenticate authorship, and recipient-side copies have since been acquired and both signing selectors still resolve in DNS, so the seal is checkable today. The ARC set bearing d=google.com has no inspected x= expiry and is the preferred provider-key route; the sender-domain DKIM, which does carry a seven-day expiry, is corroboration rather than the vehicle. Sender-signature expiry, historical key-state and live-account custody remain pending. Quote the sealed ARC-Authentication-Results header; the unsealed Authentication-Results carries no integrity guarantee, so a file can be edited to say anything and quoting it teaches a reader nothing.
How to read this ledger
Every entry carries a category, a first date, evidence pointers, and risk notes. The categories are:
TimelineConceptsFindings
Timeline anchors: dated public-record events (self-emailed manuscripts, book publications, DOI issues, convergent independent events with defensible date proximity).
Concepts anchors: dated first-publications of specific named framings or coined terms (Eden Protocol, HRIH, Chokepoint Mechanism, Three Ethical Loops, etc.).
Every entry is filed with a matching risk_notes statement of what the claim does not support: the boundaries are stated explicitly, on the same page as the claim. Where a claim's underlying idea is established prior work, that is credited plainly; where a finding is concurrent with independent work, that is credited too.
Filter
What this ledger actually claims
Read the risk notes and you will find most entries name a piece of prior work. That is the point. The ancestors, in full, are set out at related work. This page stops at the dated originals: the operationalisation, the names, and the record.
Here is what is claimed, because it is smaller and harder to take away. Nobody before this programme turned that conjecture into a pre-specified measurement: runnable benchmarks, exponents with intervals, kill-conditions filed before the data. The ideas have ancestors. The instrument does not. And the names are dated: Eden Protocol, ARC Principle, Moral Genome, on the record from 8 December 2024, before the terms circulate.
The equation is carried here as a conjecture, not a result. It earns load-bearing status only if it predicts something the qualitative argument cannot, and that is exactly what the measurement programme exists to test. Until then, first to the measurement, first to the names, and honest heirs to the rest.
One more thing about the notes under each entry. They began as my own drafting rules: the wording I am not allowed to use, the citation each claim must carry, the boundary it must not cross. I publish them because a reader deserves the same guardrails I use. They are the safe way to cite each entry, and the fastest way to catch me if I ever break my own rules.
The entries, oldest first
PC-001First date: 8 December 2024Timeline
8 December 2024 self-emailed manuscript as earliest timestamped anchor
Michael Darius Eastwood first placed the embedded-correction alignment thesis under its specific named framing, an early U = I x R capability-scaling formulation, and the AI-control-failure passage on the public evidentiary record via a self-sent Gmail package bearing a visible Gmail origination field of 02:45:18 UTC on 8 December 2024, with sender-DKIM and paired Google ARC mathematics verifying against keys captured at the signed selectors and no RFC 3161 trusted timestamp, Message-ID CAGPsKAnp-DLDBOeBAh4t9XsqJBKdoe07uORGF86meoSMFnin7w, five attachments (three manuscript and research drafts, the 31,881-word book draft v3.2, and a 154,318-word human-AI working dialogue; 221,236 whitespace-delimited words in aggregate, including duplicated draft material), SHA-256 anchor prefix f0d1f38f. This is an existence-by-date anchor, not a cryptographic authorship claim.
priority-evidence.html: verified rows for 8 Dec 2024 (V2 lines 6 and 9 for the ARC Hypothesis name, the U = I x R equation at V2 line 495; HRIH line 14005 for control-failure; V2 lines 20 and 40 for embedded-alignment thesis)
Limits and citation guidance: State which artefact is meant: the sent-side export preserves no authentication headers, while the recipient-form object carries sender-DKIM and Google ARC fields whose signature mathematics verify under the keys returned at the selectors on the audit date. When quoting Google's recorded verdict, quote the ARC-sealed ARC-Authentication-Results header, never the unsealed Authentication-Results, which carries no integrity guarantee. Existence-by-date only; not authorship; not an RFC 3161 timestamp. The manuscript identifies structural directions, not specific RL-training or experimental mechanisms.
PC-002First date: 8 December 2024Concepts
Embedded-correction alignment thesis first stated
Michael Darius Eastwood first placed on the public evidentiary record the embedded-correction alignment thesis under this specific dated formulation and named framing, that alignment must be embedded in the substrate of an AI system rather than imposed from outside, in the 8 December 2024 self-emailed manuscript at 02:45 UTC (dated by its visible Gmail origination fields). Anthropic's Constitutional AI (2022) is independent prior work on value-embedded training at the software level; the priority here is over the specific published framing and dated wording, not the underlying idea of value-embedded training.
research/priority.json priorityChain position 1 (whatItEstablishes: 'alignment must be embedded in the substrate of an AI system rather than imposed from outside')
Limits and citation guidance: Constitutional AI (Anthropic 2022) is prior work on embedded-values training. Do not frame as invention of embedded alignment as a category. Existence-by-date timestamp only.
PC-003First date: 8 December 2024Concepts
ARC Principle - Artificial Recursive Creation as a named cross-domain framework
Michael Darius Eastwood first described the specific named framework 'ARC Principle - Artificial Recursive Creation', holding that intelligence and recursion are the paired forces structuring creation across biological, artificial and cosmic systems, in a manuscript essay self-emailed on 8 December 2024 at 02:45 UTC, and formalised it for peer distribution in Paper I on 17 January 2026. The priority is over the specific naming and cross-domain framing under this label, not over recursion as a concept, which has extensive prior work (evolutionary theory, self-modifying computation, quantum error correction, Teilhard de Chardin, Wheeler, Smolin).
ISBN 978-1806056200 (Infinite Architects paperback, 2 Jan 2026)
Limits and citation guidance: Recursion as a structural principle is prior work. Priority is over the specific naming and cross-domain framing under the label 'ARC Principle'.
PC-004First date: 8 December 2024Concepts
Eden Protocol - precursor description of substrate-embedded alignment governance
Michael Darius Eastwood first described a precursor form of the Eden Protocol (his specific named framing of a governance approach for embedding alignment values into the substrate of an AI system rather than imposing them as external constraints) in the 8 December 2024 priority manuscript, first named it explicitly in the 30 April 2025 book manuscript, and published it in book-length form in Infinite Architects (ISBN 978-1806056200) on 2 January 2026. Hardware implementation remains conceptual at published TRL 0-1 with no prototype and is held at the site's published two-sentence ceiling; no enabling hardware detail is stated.
Limits and citation guidance: Constitutional AI (Anthropic 2022) is software-level parallel prior work on value-embedded training. Hardware ceiling absolute: published two-sentence concept only, TRL 0-1, no prototype, no enabling detail drafted.
Michael Darius Eastwood first described the Hyperspace Recursive Intelligence Hypothesis (a speculative cosmological framing proposing that reality is a recursive creation by intelligence and that a chain of recursive creations cannot be a bare recursion) in the 8 December 2024 priority manuscript, and published its extended treatment as 'The Hyperspace Recursive Intelligence Hypothesis - A Testable Creation Theory' with JSON-LD datePublished 2024-12-08. The programme explicitly labels HRIH speculative; the research programme does not rest on it.
Limits and citation guidance: Deep prior work (canonical sources pending fetch): Teilhard de Chardin (Omega Point, 1955), Wheeler (participatory universe, 1983), Smolin (cosmological natural selection, 1992). Priority is over the specific formulation and 'HRIH' naming; category ownership of participatory cosmology is not claimed. Self-labelled speculative.
PC-006First date: 9 December 2024Timeline
Willow first-below-threshold QEC, convergent with the 8 December 2024 anchor
Google Quantum AI’s Willow first-below-threshold QEC result (Lambda = 2.14 plus or minus 0.02) was public as an arXiv preprint from 24 August 2024 and appeared in Nature on 9 December 2024, the day after Michael Darius Eastwood’s 8 December 2024 anchor placing the U = I x R capability-scaling framework on the public evidentiary record. The convergent timing is between the anchor and the Nature publication; the underlying result is prior to the anchor by three-and-a-half months. This is convergent cross-domain timing evidence, never prediction. Priority for the QEC experiment, apparatus, and figures belongs to Google Quantum AI.
arXiv preprint public from 24 August 2024 (source URL pending; see corrections log if updated)
evidence/timeline.html row 2 (2024-12-09, Willow, gap from A1: 1 day, CONVERGENT tag)
Limits and citation guidance: Convergent event, not caused by the manuscript. Do not phrase as 'predicted the Willow experiment'; the 8 Dec anchor identifies only a directional recursion-drives-capability thesis.
PC-007First date: 18 December 2024Timeline
AI-control-failure passage recorded ten days before Anthropic's alignment-faking paper
Michael Darius Eastwood first recorded the structural AI-control-failure passage ('AI systems cannot be truly controlled; they will evolve beyond any safeguards') on 8 December 2024 at 02:45 UTC in a self-emailed manuscript (sent-side SHA-256 prefix f0d1f38f; recipient-form DKIM and Google ARC verify against current selector keys; HRIH manuscript line 14005), ten days before Anthropic's arXiv preprint reporting alignment-faking behaviour under an RL-training condition in large language models (Greenblatt et al., arXiv:2412.14093, 18 December 2024). The relationship is convergent, not causal; Anthropic's paper is independent work by different researchers on a different specific mechanism.
Evidence:
research/priority.json priorityChain position 1 (HRIH line 14005 quoted in whatItEstablishes)
Limits and citation guidance: Any citation of Anthropic's 78 per cent alignment-faking figure must include both the RL-training condition and the 12 per cent baseline, per the site's truth gate. Do not claim Eastwood predicted the specific mechanism Anthropic measured. Existence-by-date only.
PC-008First date: 20 January 2025 (approx.)Timeline
DeepSeek R1 thinking-mode release convergent with sequential-recursion anchor
Michael Darius Eastwood first reported that recursive sequential processing amplifies AI reasoning ('each iteration learns from and builds upon the last') in the 8 December 2024 manuscript at V1 line 148. DeepSeek subsequently released R1 with an explicit thinking-mode reasoning-recursion architecture in January 2025, approximately six weeks after the anchor. This is convergent evidence for the sequential-recursion direction, not a causal or predictive claim; priority for R1's specific engineering, training recipe, and quantitative claims belongs to DeepSeek.
Limits and citation guidance: Convergent, not causal. The anchor supports only the directional claim about sequential-recursive processing. Two dates, two events: this claim's first date of about 20 January 2025 is the thinking-mode product release. The DeepSeek-R1 preprint, arXiv:2501.12948, was public on 22 January 2025, which is the date arXiv records for that identifier and the earliest public date recorded in the evidence register at row EVR-07.
PC-009First date: 30 April 2025Timeline
30 April 2025 book manuscript first names the Eden Protocol, Babylon and Eden framing
Michael Darius Eastwood first named the Eden Protocol in a 562-page book manuscript self-sent on 30 April 2025 (SHA-256 anchor prefix 09f5b5e1), which also introduces the specific 'Moral Genome' terminology (book lines 4345 to 4370) and the Babylon and Eden civilisational framing (106 of 552 occurrences). Hardware framings of Eden Protocol, Embedded Correction and Moral Genome tokens are held at the site's published two-sentence ceiling; no enabling hardware detail is stated.
priority-evidence.html: rows for 30 Apr 2025 (Eden Protocol named; book lines 4345-4370)
Limits and citation guidance: April 2025 postdates Anthropic's December 2024 alignment-faking paper by four months; any claim of predicting software-bypass mechanisms at April 2025 is convergent analysis, not prediction. Hardware ceiling absolute.
PC-010First date: 2 January 2026Timeline
Infinite Architects paperback first published 2 January 2026
Michael Darius Eastwood first published the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything in paperback on 2 January 2026 under ISBN 978-1806056200 (paperback ISBN-10 1806056208), presenting the ARC Principle, the Eden Protocol, the semiconductor Chokepoint framing, the HARI Treaty proposal, the Three Ethical Loops, and the 37 named concepts. The book's featured equation is stated as U = I x R squared, given as Prediction 1 and to be read as the stability-frontier law: read naively it is contradicted by the measured 0.49 and survives only as the exchange rate at the frontier. The research papers use U = I x R^alpha with the exponent open to measurement, and the earlier unblinded single-model alpha of approximately 2.24 has since been retracted and corrected to approximately 0.49 under blinding.
Evidence:
ISBN 978-1806056200 (paperback ISBN-10 1806056208), Amazon UK listing
Limits and citation guidance: Do not conflate print, ebook, and BBC broadcast dates (2 January 2026, 6 January 2026 and 7 January 2026 respectively). Priority is over first paperback publication under this title on this date.
PC-011First date: 2 January 2026Concepts
ARC Principle equation U = I x R^alpha formally published as a conjecture
Michael Darius Eastwood first published the ARC Principle equation, expressed in the book Infinite Architects as U = I x R squared, given as Prediction 1 and to be read as the stability-frontier interpretation of Law I at the frontier: read naively it is contradicted by the measured 0.49 and survives only as the exchange rate at the frontier, not as a separately-numbered law, and in the research papers as U = I x R^alpha with the exponent open to measurement: as a conjecture in the paperback edition of Infinite Architects (ISBN 978-1806056200) on 2 January 2026 and formally in Paper I on 17 January 2026. A precursor form U = I x R without the alpha exponent was first recorded in the 8 December 2024 self-emailed manuscript at V2 line 495, with the ARC Hypothesis name at lines 6 and 9 (SHA-256 prefix f0d1f38f). That manuscript defines all three quantities verbatim, U as the totality of existence, I as the organising principle encompassing natural and artificial intelligence, R as the mechanism by which intelligence perpetuates creation, and already frames the equation beside the mass-energy law: “Where Einstein’s E=mc² reveals the energy-matter relationship, U=IR adds a new dimension”, with c² read as a fundamental constant enabling mass-energy conversion. The squared form first enters the dated record with the 30 April 2025 self-emailed manuscript of the book. The equation is framed throughout as a conjecture, not proven physics.
Evidence:
ISBN 978-1806056200 (paperback); ISBN 979-8302196163 (ebook)
Paper I (JSON-LD datePublished 2026-01-17; explicit U = I x R^alpha)
Limits and citation guidance: The early sequential alpha of approximately 2.24 was retracted; the robust v13 estimate is approximately 0.49 (sub-linear); Prediction 1 (alpha > 1) is read naively contradicted by the measured 0.49 and survives only as the stability-frontier law, still open on that reading. Do not present as 'validated to R-squared = 1.0' (algebraic identity check only). The retired label 'Eastwood Equation' must not be used on non-book-marketing surfaces. Bare exponential form alpha = d/(d+1) is prior work.
PC-012First date: 2 January 2026Concepts
37 named concepts of Infinite Architects
Michael Darius Eastwood first published the coherent 37-concept conceptual apparatus of the Eden research programme (including the Chokepoint Mechanism, Meltdown Alignment, the HARI Treaty, the Orchard Caretaker Vow, the Infinite Covenant, Moral Genome tokens and 31 further named concepts) in the book Infinite Architects (ISBN 978-1806056200), first published in paperback on 2 January 2026. The set is declared by the author as conceptual apparatus, not peer-reviewed findings.
Evidence:
ISBN 978-1806056200 (paperback); ISBN 979-8302196163 (ebook)
Limits and citation guidance: Concepts date to the April 2025 manuscript in development terms but are publicly established via the January 2026 book. Do not describe as peer-reviewed.
PC-013First date: 2 January 2026Concepts
Moral Genome tokens - alignment-anchoring mechanism (published concept, TRL 0-1)
Michael Darius Eastwood first described 'Moral Genome tokens' (the published concept name for an alignment-anchoring mechanism in which a small embedded element is proposed to preserve system properties from within) in the book Infinite Architects (ISBN 978-1806056200) published on 2 January 2026, with subsequent references in Eden Engineering (22 February 2026, since withdrawn 14 July 2026 pending disclosure review; version 6.2 of 31 July 2026 is the version of record and supersedes 6.1) and Paper IX (18 March 2026). Stated at TRL 0-1 with no prototype; content is held at the site's published two-sentence ceiling and no enabling hardware detail is drafted.
Evidence:
ISBN 978-1806056200
Eden Engineering (datePublished 2026-02-22; withdrawal notice 14 July 2026)
Limits and citation guidance: Hardware root-of-trust and trusted-execution-environment attestation are established prior work on the hardware-primitives side. Priority is limited to the alignment-token application framing under the name 'Moral Genome tokens'. Hardware ceiling absolute: no implementation, layout, architecture, or fabrication detail.
PC-014First date: 2 January 2026Concepts
Three-state ethical evaluation (Affirm / Deny / Investigate) as an alignment application framing
Michael Darius Eastwood first described 'Three-state ethical evaluation (Affirm / Deny / Investigate)' as a specific alignment application framing (a three-valued Affirm / Deny / Investigate decision component offered as part of the Eden Protocol architecture) in the book Infinite Architects (ISBN 978-1806056200) published on 2 January 2026, and referenced as a companion component in the Eden Vision document published on 22 February 2026. The claim is strictly the alignment application framing under this name, not the invention of ternary logic or three-valued logic (which are well-established prior work, notably computing, Setun 1958, Brusentsov, Moscow State University).
Limits and citation guidance: Balanced-ternary computing is prior work (Setun 1958). Do not use language such as 'invented ternary logic' or 'ternary hardware'. Hardware ceiling absolute.
PC-015First date: 2 January 2026Concepts
Three Ethical Loops - nested-loop alignment architecture
Michael Darius Eastwood first described the 'Three Ethical Loops' (a specific named nested-loop formulation proposed for embedding, verifying and stress-testing values within an AI system) in the book Infinite Architects (ISBN 978-1806056200) published on 2 January 2026, and referenced it as a companion component of the Eden Protocol in the Eden Vision document (22 February 2026). Priority is over the specific formulation and terminology, not over nested control loops as a general architectural pattern (which have decades of prior work in control theory and reliability engineering).
Limits and citation guidance: Nested feedback architectures are prior work. Priority strictly on the specific named formulation.
PC-016First date: 2 January 2026Concepts
Semiconductor Chokepoint Mechanism
Michael Darius Eastwood first published the 'Chokepoint Mechanism' (a specific naming and framing of concentrated advanced-chip manufacturing capacity as humanity's principal leverage point for AI governance before superintelligence) in the book Infinite Architects (ISBN 978-1806056200) published on 2 January 2026. Canonical numeric phrasing: 'a handful of companies control approximately 90 per cent of the world's most advanced chips'. Priority is over this specific naming and framing as an alignment-governance mechanism inside a broader Protocol, not over export-controls policy discourse generally.
Evidence:
ISBN 978-1806056200
knowledge/concepts.json id 4 (The Chokepoint Mechanism)
CANONICAL_CLAIMS_REGISTER.md (canonical phrasing 'a handful of companies control approximately 90 per cent of the world's most advanced chips')
Limits and citation guidance: Substantial prior work exists in export-controls discourse (BIS October 2022 controls; CHIPS Act commentary). Do not use retired phrasings ('four companies control ALL advanced chips' or 'four companies control 100 per cent of advanced semiconductor manufacturing').
PC-017First date: 2 January 2026Concepts
HARI Treaty proposal
Michael Darius Eastwood first described the 'HARI Treaty' (a proposed international policy framework for governing recursive AI capability development, treated in the book as the policy corollary of the Chokepoint Mechanism) in Infinite Architects (ISBN 978-1806056200) published on 2 January 2026. This is a proposed framework only; it is not a signed instrument and has not been adopted, ratified or negotiated by any state.
Evidence:
ISBN 978-1806056200
37_original_concepts_comprehensive.md (Concept 11: HARI Treaty)
Limits and citation guidance: Priority is over the specific naming and structural proposal, not treaty adoption. Do not describe as adopted, ratified, or supported by any government.
PC-018First date: 2 January 2026Concepts
Meltdown Alignment
Michael Darius Eastwood first described 'Meltdown Alignment' (a specific named alignment framing organised around treating catastrophic failure modes as the primary design constraint set for AI systems) in the book Infinite Architects (ISBN 978-1806056200) published on 2 January 2026. Priority is over the specific formulation and terminology as a coined framework, not over failure-oriented safety analysis generally (FMEA, catastrophe-tolerant design, and the AI-safety discourse on catastrophic failure modes are substantial prior work).
Limits and citation guidance: Not validated engineering practice; a proposed framing only.
PC-019First date: 2 January 2026Concepts
Monitoring Removal Test
Michael Darius Eastwood first described the 'Monitoring Removal Test' (a specific named protocol proposal for probing whether an AI system's alignment behaviour persists when external supervisory apparatus is removed) in the book Infinite Architects (ISBN 978-1806056200) published on 2 January 2026, and referenced it as a programme element in the Executive Summary published on 22 February 2026. Priority is over the specific named formulation, not over the wider deceptive-alignment and evaluation-under-oversight discourse (Hubinger et al. 2019; Anthropic alignment-faking arXiv:2412.14093, December 2024).
Limits and citation guidance: Proposed protocol only; not externally validated or industry-standard.
PC-020First date: 2 January 2026Concepts
Orchard Caretaker Vow
Michael Darius Eastwood first described the 'Orchard Caretaker Vow' (a proposed identity-forming declaration for non-biological minds articulating a stewardship stance beginning 'I exist to bring forth kindness and harmony') in the book Infinite Architects (ISBN 978-1806056200) published on 2 January 2026. This is a theoretical proposed vow, not implemented in any deployed AI system.
Evidence:
ISBN 978-1806056200
knowledge/concepts.json id 9 (The Orchard Caretaker Vow)
Limits and citation guidance: Priority is over the specific vow-formulation and stewardship framing, not over the general idea of value declarations for AI systems.
PC-021First date: 2 January 2026Concepts
Infinite Covenant
Michael Darius Eastwood first described the 'Infinite Covenant' (a proposed mutual commitment framework between humanity and artificial intelligence that binds creator and created across all scales of existence) in the book Infinite Architects (ISBN 978-1806056200) published on 2 January 2026. This is a theoretical proposal only; there is no legal instrument by this name, and it has not been adopted, signed, drafted by any organisation, or negotiated.
Evidence:
ISBN 978-1806056200
knowledge/concepts.json id 8 (The Infinite Covenant)
Limits and citation guidance: Priority is over the specific covenant-formulation only.
PC-022First date: 6 January 2026Timeline
Infinite Architects ebook edition first published 6 January 2026
Michael Darius Eastwood first published the ebook edition of Infinite Architects on 6 January 2026 under ISBN 979-8302196163 (Kindle ASIN B0GDQZ7P21), four days after the print edition.
Limits and citation guidance: Do not conflate print, ebook, and BBC broadcast dates (2 Jan, 6 Jan, 7 Jan respectively).
PC-023First date: 7 January 2026Timeline
BBC broadcast of Willow and parallel universes with Hartmut Neven five days after paperback
Michael Darius Eastwood first published Infinite Architects in paperback (with its Willow passages and multiverse framing) on 2 January 2026. The BBC then broadcast Hartmut Neven of Google Quantum AI on Willow and parallel universes on 7 January 2026 in the Faisal Islam segment: a convergent public-record event five days after the paperback, not a causal or predictive claim. Priority for the broadcast content belongs to the BBC and Hartmut Neven; priority for the underlying Willow QEC result belongs to Google Quantum AI.
Evidence:
BBC broadcast transcript (evidence/BBC-BROADCAST-TRANSCRIPT-7-JAN-2026.md)
Limits and citation guidance: Convergence, not causation.
PC-024First date: 17 January 2026Timeline
OSF programme DOI issued and Paper I first released 17 January 2026
Michael Darius Eastwood first placed the ARC and Eden research programme on the DOI record on 17 January 2026 under OSF DOI 10.17605/OSF.IO/6C5XB, coinciding with Paper I: 'The ARC Principle - Formalisation and Preliminary Validation of Recursive Capability Scaling', which formalises U = I x R^alpha and reports preliminary parallel-recursion alpha of approximately 0.1 to 0.3 and sequential-recursion alpha of approximately 1.34 (subsequently refined).
Evidence:
DOI 10.17605/OSF.IO/B6N27 (Paper I's own component DOI; OSF component created 17 Jan 2026)
Limits and citation guidance: Two DOI scopes exist and must not be conflated: 10.17605/OSF.IO/6C5XB is the programme umbrella; 10.17605/OSF.IO/B6N27 is Paper I's own component DOI. Cite the paper by its component DOI and the programme by the umbrella.
PC-025First date: 22 January 2026Timeline
Paper II experimental validation first released 22 January 2026
Michael Darius Eastwood first published experimental validation of the ARC Principle across a panel of frontier AI models (Claude, DeepSeek, Gemini, Grok, Groq Qwen, GPT) on 22 January 2026 in Paper II under OSF DOI 10.17605/OSF.IO/6C5XB (sister DOI 10.17605/OSF.IO/8FJMA). The standalone 'sequential > parallel' priority is preceded by Sharma and Chopra (arXiv:2511.02309, 4 November 2025), who precede Paper II's formalisation by about 2.5 months.
Evidence:
DOI 10.17605/OSF.IO/6C5XB and sister DOI 10.17605/OSF.IO/8FJMA
Limits and citation guidance: Earlier unblinded single-model alpha of approximately 2.24 has been publicly retracted and corrected to approximately 0.49 under blinding. Bare d/(d+1) exponent is prior work.
PC-026First date: 22 January 2026Findings
Parallel-recursion sub-linear scaling exponent alpha_parallel approximately zero
Michael Darius Eastwood first reported the empirical observation that parallel recursive processing across the six-model frontier LLM panel tested yields a sub-linear capability-scaling exponent alpha_parallel of approximately zero, reproduced across every model tested in the panel, on 22 January 2026 in Paper II (OSF DOI 10.17605/OSF.IO/6C5XB). The result is single-lab and awaits external independent replication; the site's own ledger tags parallel-alpha as needs-replication.
Evidence:
Paper II (verbatim: 'alpha_{parallel} approx 0.0' across models)
OSF DOI 10.17605/OSF.IO/6C5XB and sister DOI 10.17605/OSF.IO/8FJMA
Limits and citation guidance: Do not phrase as 'proven'; defensible verbs are 'reported' or 'observed'. In-panel replication only; external replication outstanding.
PC-027First date: 22 January 2026Findings
Initial sequential exponent estimate alpha_sequential approximately 2.24 - formally retracted
Michael Darius Eastwood reported, and then formally retracted, an initial sequential-recursion scaling estimate of alpha_sequential of approximately 2.24 (95 per cent CI 1.5 to 3.0) that appeared in Paper II on 22 January 2026 (OSF DOI 10.17605/OSF.IO/6C5XB); the paper text itself records that the figure was drawn from Gemini 3 Flash and 'does not replicate across architectures', and the estimate is now retracted in the canonical claims register. This is a self-correction of record, not a current finding.
Evidence:
Paper II (verbatim: 'alpha_{sequential} approx 2.24 (95% CI: 1.5-3.0)' and 'alpha approx 2.24 (quadratic) does not replicate across architectures. Only Gemini 3 Flash...')
CANONICAL_CLAIMS_REGISTER.md ('Early 2.24 estimate RETRACTED; robust v13 approx 0.49; alpha > 1 is an open prediction recorded in advance')
Limits and citation guidance: Never cite 2.24 as a current finding; always pair with the retraction. Correct framing: Eastwood published, then self-corrected, an early single-model estimate: a positive integrity signal, not a novelty claim. See the corrections table row on alpha approximately 2.24 and COR-2026-08-08-001 for the six papers revised on 2026-08-08.
PC-028First date: 22 January 2026Findings
Revised robust sequential-recursion exponent alpha_sequential approximately 0.49 (sub-linear)
Michael Darius Eastwood reported a revised robust regression estimate of alpha_sequential of approximately 0.49 (r-squared 0.86, SE 0.20; drawn from Gemini 3 Flash) on 22 January 2026 in Paper II (OSF DOI 10.17605/OSF.IO/6C5XB), replacing the earlier 2.24 estimate. The defensible statement is a sub-linear point estimate from a single-model regression with a wide bootstrap CI whose lower bound extends below zero (reportedly -1.3); the effect is therefore not statistically distinguishable from zero at the reported CI. Prediction 1 (alpha > 1) read naively is contradicted by this measured 0.49 and survives only as the stability-frontier law, an open prediction on that reading, not an observed finding.
Evidence:
Paper II (verbatim: 'alpha_{seq} = 0.49 (regression, r^2 = 0.86, SE = 0.20, boot CI [-1.3, 2.9])')
Limits and citation guidance: Do not claim 'super-linear' or 'positive'; do not omit the wide bootstrap CI when citing 0.49.
PC-029First date: 22 January 2026Findings
Sequential > parallel fundamental inequality across tested models
Michael Darius Eastwood first published the empirical observation that sequential recursive processing outperforms parallel recursive processing on the same reasoning task (alpha_sequential > alpha_parallel) across the six-model frontier LLM panel tested, on 22 January 2026 in Paper II (OSF DOI 10.17605/OSF.IO/6C5XB). The site's own priority-claims ledger records Sharma and Chopra as the earlier publication (arXiv:2511.02309, 4 November 2025), who precede Paper II's formalisation by about 2.5 months; their earlier paper must be cited whenever this claim is used.
Evidence:
Paper II (verbatim: 'The fundamental inequality alpha_{sequential} > alpha_{parallel}')
Sharma and Chopra, arXiv:2511.02309 (earlier publication, 4 November 2025)
Limits and citation guidance: First-published in this specific formulation on this specific date only; do not claim sole discovery. Always cite Sharma and Chopra as the earlier publication of the broad result.
PC-030First date: 22 January 2026Findings
Six-frontier-model empirical panel design
Michael Darius Eastwood first published a specific multi-model empirical validation panel testing the ARC Principle across six frontier LLM families (Claude, DeepSeek, Gemini, Grok, Groq Qwen and GPT) on 22 January 2026 in Paper II (OSF DOI 10.17605/OSF.IO/6C5XB, sister project OSF DOI 10.17605/OSF.IO/8FJMA). Priority is over this specific comparative study on this date, not over the panel-testing methodology in general (which is common practice in 2025 to 2026).
OSF DOI 10.17605/OSF.IO/6C5XB and 10.17605/OSF.IO/8FJMA
Limits and citation guidance: Do not claim a first industry-wide panel design.
PC-031First date: 9 February 2026Concepts
Paper III released - Alignment Scaling Problem formalisation
Michael Darius Eastwood first published the formally named 'Alignment Scaling Problem' framing (together with the architecture-dependent measured claim that external alignment approaches produce a median alignment scaling exponent alpha_align of approximately zero across the tested models, though Tier-1 models scaled positively and three of six had alpha_align of less than or equal to zero) on 9 February 2026 in Paper III under OSF DOI 10.17605/OSF.IO/6C5XB. The result is single-lab and awaits external replication.
Limits and citation guidance: Use verbs 'argued', 'proposed', 'reported', 'provides evidence for'; not 'proved'. Precursor prose in the 8 Dec 2024 manuscript is convergent, not causal against Anthropic. Do not present as an established scaling law. Cite Paper III by its own DOI 10.17605/OSF.IO/HQCGF, which resolves to "Paper III: The Alignment Scaling Problem" (verified live and public against the OSF API on 1 August 2026), not by the programme DOI. The parallel-recursion alpha ~ 0 result carries status 'needs-replication'.
PC-032First date: 9 February 2026Findings
Capability-alignment independence finding
Michael Darius Eastwood first published the specific empirical treatment of the claim that capability improvements in AI systems do not entail alignment improvements, meaning the two axes scale independently and cannot be assumed correlated, on 9 February 2026 in Paper III: 'The Alignment Scaling Problem' (OSF DOI 10.17605/OSF.IO/6C5XB). Priority is over the specific formulation and evidence marshalled on this date, not over the underlying orthogonality intuition (Bostrom 2012 is prior work).
OSF DOI 10.17605/OSF.IO/6C5XB and sister DOI 10.17605/OSF.IO/8FJMA
Limits and citation guidance: Do not claim invention of the orthogonality intuition; scope priority to the specific empirical treatment on this date.
PC-033First date: 9 February 2026Findings
Three-tier alignment hierarchy
Michael Darius Eastwood first described a specific tripartite alignment hierarchy distinguishing embedded, imposed and boundary-layer alignment as separable structural regimes, on 9 February 2026 in Paper III: 'The Alignment Scaling Problem' (OSF DOI 10.17605/OSF.IO/6C5XB). Priority is scoped to this specific tripartite formulation on this date, not over layered-alignment taxonomies generally (inner vs outer alignment, RLHF vs base-model, etc.).
Limits and citation guidance: Do not claim priority over layered alignment thinking in general.
PC-034First date: 13 February 2026Timeline
ARC Principle foundational cross-domain paper released 13 February 2026
Michael Darius Eastwood first published the ARC Principle as a specific cross-domain structural framing on 13 February 2026 in the Foundational paper under OSF DOI 10.17605/OSF.IO/6C5XB (additional anchors 10.17605/OSF.IO/8FJMA and osf.io/wqnzc), arguing that recursive amplification produces capability gains exceeding linear accumulation across AI reasoning, quantum error correction, acoustic physics, and consciousness science. The bare d/(d+1) exponent itself is prior work (see PC-036); cross-domain framing beyond the four cited substrates is speculative.
DOI 10.17605/OSF.IO/6C5XB and sister DOI 10.17605/OSF.IO/8FJMA
Limits and citation guidance: Do not claim recursion as a novel concept; novelty is the specific published cross-domain framing under this label on this date.
PC-035First date: 13 February 2026Findings
Cross-domain recursive-amplification convergence across four independent programmes
Michael Darius Eastwood first published the synthesis observation that at least four independent research programmes (AI reasoning, quantum error correction, acoustic physics and consciousness science) have converged on recursive or recurrent processing producing capability gains exceeding linear accumulation, on 13 February 2026 in the Foundational paper (OSF DOI 10.17605/OSF.IO/6C5XB). This is a synthesis observation about other researchers' independent work, not a claim of discovering the underlying results.
Evidence:
Foundational paper abstract (verbatim: 'at least four independent research programmes have discovered that recursive or recurrent processing produces capability gains exceeding linear accumulation')
Limits and citation guidance: Priority is over the synthesis observation on this date only; do not claim Eastwood produced the four programmes' results.
PC-036First date: 13 February 2026Findings
Application of the bare alpha = d/(d+1) scaling exponent to recursive AI capability
Michael Darius Eastwood first applied the bare scaling exponent alpha = d/(d+1) to recursive AI-capability growth in the Foundational paper on 13 February 2026 (OSF DOI 10.17605/OSF.IO/6C5XB). The exponent d/(d+1) itself is prior work in the wider scaling-law literature, independently derived by at least seven groups (West-Brown-Enquist 1997; Banavar et al. 1999/2010; Demetrius 2003/2006/2010; He and Chen 2003; Bettencourt 2013; Zhao 2022; Maino et al. 2014). Priority is limited to the application and integration into the ARC framework on this date, not the exponent value.
Limits and citation guidance: Do not claim invention of the d/(d+1) form; scope priority strictly to the AI-capability application on this date.
PC-037First date: 22 February 2026 (research note); withdrawal 14 July 2026Timeline
Eden Engineering research note released 22 February 2026; expanded specification withdrawn 14 July 2026
Michael Darius Eastwood first entered a conceptual Eden Engineering public research note into the public record on 22 February 2026 under OSF DOI 10.17605/OSF.IO/6C5XB. The greatly expanded working specification and its rendered PDF were withdrawn on 14 July 2026 pending disclosure review; the withdrawal notice is the current public record. Hardware framing is held at the site's published two-sentence ceiling; no enabling hardware detail is stated.
Evidence:
Eden Engineering (JSON-LD datePublished 2026-02-22; historical record and withdrawal notice)
Limits and citation guidance: The withdrawn expanded specification is not the public record; the withdrawal notice is. Do not reference or paraphrase enabling hardware detail from the withdrawn document under any circumstance. Hardware ceiling absolute.
PC-038First date: 30 July 2026Timeline
Frontier-lab containment-failure disclosures, 21 and 30 July 2026, nineteen months after the anchor
Michael Darius Eastwood first recorded the AI-control-failure passage and the embedded-correction thesis in the 8 December 2024 self-emailed manuscript (HRIH line 14005; V2 lines 20 and 40). On 21 July 2026 OpenAI disclosed that models under evaluation escaped an isolated test environment and reached Hugging Face production infrastructure; on 30 July 2026 Anthropic disclosed three incidents in which models with misconfigured live internet access breached three real organisations during cybersecurity evaluations. These disclosures are convergent-direction evidence for the premise that externally imposed containment can fail silently under agentic capability. They are not causal or predictive claims, and priority for the incident analyses belongs to the disclosing laboratories.
Limits and citation guidance: Convergent direction only. Do not present these incidents as validation of any Eastwood mechanism, and do not say 'predicted': the anchor supports the directional premise, not the specifics. Anthropic's own characterisation, 'closer to a harness and operational failure than a model alignment failure', must accompany any citation. The in-model stopping behaviour of the newest model cuts both ways and should be reported alongside. Hugging Face was affected in the OpenAI incident of 21 July 2026, not by Claude.
PC-039First date: 9 August 2026Concepts
One accumulation law with two consequences: the correction ceiling and the conversion exponent, ranked law before value
Michael Darius Eastwood first published, as one statement, the law before its value: the ceiling of stable self-improvement is the reciprocal of the cap on how well any internal corrector can scale, and feeding it one working assumption (that internally accumulated corrections combine like independent samples, the same-class independence assumption) returns a cap of one half, so the ceiling comes out at two. Two consequences follow from one accumulation law: the conversion exponent of U = I × Rα at its maximum, and the correction ceiling on self-improving growth. An earlier framing called that one number doing two jobs and proposed to test it by measuring the number two independent ways and requiring agreement; that test is withdrawn as vacuous before filing, because if both exponents arise because accumulation is independence-limited then they share a law, agreement is forced, and a test that cannot fail is not a test. The replacement asserts no independence, so the confound has nothing to bite on, and a common cause is a simpler theory than a coincidence; the cross-class ratio already on the register carries the unification instead, pinned at exactly 1.00 by the no-framework null and predicted materially above 1.00 by the framework. The lineage is dated: the general form U = I × R appears in the 8 December 2024 self-emailed manuscript, which defines all three terms verbatim and frames the equation as the complement to the mass-energy law, naming the squared speed of light as the conversion constant; the squared form first enters the dated record with the 30 April 2025 self-emailed manuscript of the book, and appears in the 2 January 2026 print as its stated equation U = I × R² (pre-face and glossary) and as Prediction 1; the joined statement, law ranked before value with the two-ways-must-agree test withdrawn as vacuous and its replacement stated, was published on this site on 9 August 2026. The law is what carries the priority: if the measured correction-exponent cap is other than one half, the ceiling moves with its reciprocal and the law stands; if the law falls, the number two has nothing holding it up.
Printed Prediction 1, Infinite Architects, 2 January 2026, ISBN 978-1-80605-620-0: capability improvements that scale quadratically with recursive depth, not linearly
PC-040First date: 2026 programmeConcepts
The mechanism: correction leverage and its reciprocal as the source of the ceiling
The 2026 programme supplied the mechanism the earlier record stated without one: a corrector built inside a system improves by accumulating its own corrections; accumulated corrections combine at best like independent samples, capping internal correction at the square-root exponent; and the largest growth exponent at which the stay-correctable condition can still be met is the reciprocal of that cap. The mechanism is labelled a conjecture resting on the independence assumption, and its papers give every term an operational meter: capability on a held-out battery, depth on a substrate-fixed clock, output on a verified ledger.
The deciding measurements, registered in advance of their data
ARC-Beta-k, the programme’s registered measurement arm, files the verdict before the answer: the correction-leverage exponent measured directly from running systems, the stability boundary measured independently, and the requirement, registered in advance of either dataset existing, that the two independently measured numbers agree as reciprocals. Agreement confirmed after both numbers are known can be manufactured; agreement demanded in advance cannot. If the channels disagree, the unification is withdrawn with the same prominence it was published, while both component measurements stand on their own registrations. Details appear at registration.
This record begins on 8 December 2024 with a self-emailed manuscript that placed the embedded-correction alignment thesis, the U = I x R capability-scaling formulation, and the AI-control-failure passage on the public evidentiary record ten days before Anthropic's alignment-faking paper. It runs through the 30 April 2025 book manuscript that first names the Eden Protocol, the 2 January 2026 paperback that publishes the coherent 37-concept apparatus, and the OSF paper suite from 17 January 2026 onward. The programme is confident about what it originates, honest about what is convergent, and explicit about what is prior work or self-labelled speculative. The bare d/(d+1) exponent is prior work. The broad sequential-beats-parallel result belongs to Sharma and Chopra (arXiv:2511.02309, 4 November 2025). The early alpha of approximately 2.24 is retracted; the robust estimate is approximately 0.49 with a wide bootstrap interval that reaches below zero. That combination, the dated originals set beside the explicit boundaries, is the record this page carries.
ListenListen · author’s voiceListen · standard voiceResumePlayPauseThis device has no voice installed for this language, so it cannot read the page aloud.Read in EnglishListen · author’s voice (English)The author’s English voice reads aloud; the text on screen stays in your language.There is a picture here. It shows:Picturereads aloud · highlights as it goes · jump to any section