Theory-level predictions preregistered on OSF: DOI 10.17605/OSF.IO/P8CKQ, registered 8 September 2026 (version 1.100), updated 13 September 2026 (version 1.102), twenty-two propositions.
Companion: Register · The ARC Theory · Operational Definitions, version 1.9.1, 5 October 2026.
A working paper outside the ARC Theory and the experiments series, load-bearing for no law; it bears on Law V only as a dated record.
This paper sets out the Polymathic Neurodivergent Profile (PNP) as a descriptive, non-diagnostic framework for a rare but practically important pattern: unusually strong cross-domain learning, structural pattern transfer between fields, intense self-directed research, and high output in complex, high-interest domains, coexisting with clinically significant difficulty in routine administration, pacing, timekeeping, task-switching, and low-stimulation procedural work. The central proposition is the Capability-Adjustment Fallacy: the institutional inference that visible high performance is evidence that reasonable adjustments are unnecessary. That inference is unsafe, and is contradicted by existing clinical guidance. NICE CG142 already directs professionals to recognise the discrepancy between intellectual functioning and adaptive functioning, and modern strengths-based assessment treats the same neurodivergent trait as advantage or impairment depending on context.
Beyond naming the pattern, the paper does three further things. First, it presents a documented live case: the author's own profile, evidenced across three independent documentation streams (clinical assessment, a court record in which adjustment requests and high-quality self-represented output coexist, and a dated research corpus, whose signature-time assertions are the mail provider's rather than an RFC 3161 timestamp, spanning AI alignment, scaling theory, and mathematics). Second, it sets out a priority and convergence analysis: the framework's central claim about neurodivergent cognition found adjacent territory in Hernández-Espinosa and colleagues (PNAS Nexus, 14 April 2026), whose subject is artificial agents: they prove full AI-human alignment mathematically impossible for Turing-complete systems and propose diversity among competing agents, under a neurodivergence analogy, as the counterbalance, an argument that in turn served as the stated conceptual foundation for a formal undecidability result (Gumbau Mezquita, June 2026); the author's timestamped manuscripts precede both. Third, it states a falsifiability standard: pre-committed abandonment conditions, operational measures, and the incremental-validity test the construct must pass or be discarded.
The aim is neither to medicalise polymathy nor to use ability as immunity from ordinary process, but to prevent a recurrent error: the conversion of a person's strengths into evidence against their needs.
Keywords: autism spectrum condition; ADHD; twice-exceptionality; monotropism; executive function; reasonable adjustments; strengths-based assessment; institutional misrecognition; access to justice; AI alignment; cognitive diversity.
This paper documents a PROFILE, not a theory: the Polymathic Neurodivergent Profile and the Capability-Adjustment Fallacy as an analytical construct sitting alongside the theory's organs rather than inside them. §12's incremental-validity check is drafted as the intended future evaluation, preregistered by timestamp and awaiting human submission, and every quantitative first-mention is labelled exploratory. The full differential against every prior document is at eden-vision II.A.8.
The contemporary neurodiversity literature has exposed a long-standing weakness in diagnostic and institutional reasoning: a tendency to recognise difficulty without adequately recognising form, pattern, and unevenness (Manalili et al., 2023). Autism and adjacent neurodevelopmental profiles are still often described primarily through deficits, even as a growing body of work argues that assessment should also identify strengths, skills, and interests, because these shape both lived experience and support needs (Woods and Estes, 2023; Russell et al., 2019).
This matters most for a particular kind of adult: one whose profile is neither global impairment nor smooth giftedness, but radical asymmetry: advanced performance in selected domains alongside marked dysfunction in others. This paper names one version of that asymmetry the Polymathic Neurodivergent Profile (PNP) and advances a single governing principle:
Capability describes what can be done under favourable cognitive conditions. Disability describes where a person is substantially disadvantaged by the environment, the process, or the demand structure. Both can be true at once.
That sentence is the functional heart of the paper. It is also the proposition that institutions, courts, clinics, universities and employers must understand before they can interpret such a person fairly. The paper goes further than naming the pattern: it documents a live instance across three independent record types, and it situates the pattern inside a set of 2026 results in AI alignment that have made cognitive diversity a research question rather than a slogan.
One biographical fact orients everything that follows. The documented case of Section 9 did not begin as an AI-safety researcher: he began as a creation theorist. The December 2024 manuscripts are, before anything else, a theory of why created systems persist: recursion amplifies, and only embedded, co-scaling correction prevents amplification from becoming destruction. The AI-safety programme, the scaling-law papers, the stability criterion and the book all derive from that single creation-theoretic move, and the derivation chain is timestamped end to end (HRIH states the hypothesis; Papers I to XI operationalise it). This matters for the present paper because it is the cleanest possible exhibit of PNP component 2, recursive pattern transfer: one structural idea, extracted once, applied across theology, cosmology, error correction, biology and machine learning, on checkable dates. The profile did not produce scattered interests; it produced one theory with many testable faces.
Credibility requires the boundary of the claim to be stated plainly and kept.
It claims: that there exists a recurring, describable pattern of cross-domain high ability combined with disability-linked impairment in routine procedural functioning; that this pattern is currently under-described by deficit-only and credential-centric models; that describing it improves the accuracy of clinical formulation and the fairness of institutional treatment; that the supporting components are individually grounded in existing literature; and that one fully documented instance exists whose records are open to qualified scrutiny.
It does not claim: that "polymathy" is or should become a diagnosis; that PNP is a validated clinical entity (it is a hypothesis); that every highly able neurodivergent person fits the profile; that the profile confers superiority rather than unevenness; that documenting it guarantees any legal outcome; that the author's cognition caused any external research result; or that artificial intelligence creates the underlying cognitive pattern. PNP is offered as a descriptive construct to be tested, refined, or refuted, not as a discovery to be announced.
The strongest single foundation for this paper is that good autism reporting is no longer expected to describe deficits alone. NHS adult-assessment guidance states that a report should record whether the person is autistic, what the person is good at, what they may need help with, and whether other conditions were found. NICE CG142 frames adult assessment as a comprehensive process covering childhood and adult presentation, functioning, coexisting conditions, sensory sensitivities, and attention to detail (NICE, 2021).
Woods and Estes (2023) developed the Survey of Autistic Strengths, Skills, and Interests precisely because clinicians and autistic people experience current approaches as overly deficit-centred and insufficiently informative for support planning; the neurodiversity paradigm reframes assessment toward what autistic people can do, not only what they cannot (Ferreira, 2025). Empirically, Russell et al. (2019) found that the most frequently described autistic traits, the ability to hyperfocus, attention to detail, strong memory, and creativity, "could be experienced either as advantageous or disadvantageous dependent on moderating influences." Cope and Remington (2022) similarly catalogue cognitive advantages in focus, memory, and creativity reported by autistic adults in the workplace.
This is the descriptive logic PNP sits inside. The point is not that strengths cancel difficulties; it is that strengths and difficulties are frequently intertwined, and that both are necessary to understand how a person functions. Neuro-affirming reporting guidance makes the same case at the level of practice, recommending that reports avoid crude functioning labels and instead describe strengths, difficulties, thinking style, and concrete adjustments (NAIT, 2024).
If the strengths literature establishes that unevenness occurs, monotropism offers a mechanism for why. Murray, Lesser and Lawson (2005) describe autistic cognition as the tendency for attention to be pulled strongly, cognitively and emotionally, into a narrow attention tunnel, producing intense engagement with a small number of interests and difficulty engaging with stimuli outside the tunnel, with transitions experienced as costly. A self-report measure, the Monotropism Questionnaire, found both autism and ADHD associated with higher monotropism, with combined-profile respondents scoring highest (Garau et al., 2023; at the time of writing this measure was a preprint and is cited as preliminary).
Closely related, Ashinoff and Abu-Akel (2021) review hyperfocus, complete absorption in a task to the point where a person appears to tune out everything else, noting it is under-researched and absent from diagnostic criteria despite being ubiquitous in autism and ADHD self-report. In ADHD clinical writing, the same phenomenon is framed as an interest-based rather than importance-based pattern of attentional regulation; this framing is clinically useful but rests largely on clinical observation, and is presented here as explanatory rather than as hard evidence.
Together these accounts predict exactly the signature PNP describes: sustained, high-output engagement with self-directed, high-interest, high-complexity material, and disproportionate difficulty with externally paced, low-stimulation, low-reward procedural tasks. The complex task is not necessarily harder; for a monotropic mind it may be more regulating.
The gifted-education literature independently describes the same shape. "Twice-exceptional" refers to individuals who are simultaneously gifted and neurodivergent, whose spiky cognitive profiles show peaks in the gifted range and valleys in the impaired range, and who exhibit asynchronous development: an internal mismatch among cognitive, emotional, and procedural domains (Reis, Baum and Burke, 2014; Foley-Nicpon et al., 2011). A recent systematic review notes that asynchrony intensifies precisely when high intelligence coexists with significant weakness.
Critically, twice-exceptionality is explicitly a profile, not a diagnosis: a direct precedent for the way PNP is framed here. The literature also names the social consequence PNP is concerned with: such individuals can be too able to be seen as disabled, and too disabled to be recognised as able, and are therefore frequently misread by systems that expect ability to be uniform. Much of this literature is child-focused and US-based; its extension to UK adults is offered as a reasoned generalisation, not as direct adult evidence.
A Polymathic Neurodivergent Profile is a descriptive construct, stated here as a hypothesis, characterised by six interacting features:
| # | Component | Description | Evidence strength |
|---|---|---|---|
| 1 | Cross-domain depth | Not casual interest in many subjects, but repeated acquisition of serious competence across distinct fields | Observational; polymathy literature (Araki, 2018; Root-Bernstein, 2009) |
| 2 | Recursive pattern transfer | Extracting relational structure from one domain and using it to accelerate learning or reasoning in another; "recursive" in the ordinary structure-mapping sense (Gentner, 1983; Gick and Holyoak, 1983). Genuine cross-domain transfer is rare and effortful, which is part of what makes a transfer-rich profile distinctive | STRONG mechanism literature; profile-level evidence observational |
| 3 | Autonomous research intensity | Deep self-directed study driven by intrinsic interest, often outside formal pathways | Consistent with special-interest depth (Grove et al., 2018) |
| 4 | Hyperfocus and immersion | Periods of unusually sustained attention, high output, and rapid synthesis in high-stimulation domains | STRONG (Ashinoff and Abu-Akel, 2021; Russell et al., 2019) |
| 5 | Executive asymmetry | Disproportionate impairment in routine administration, timekeeping, correspondence, filing, scheduling, task-switching, and procedural housekeeping | STRONG (NICE CG142 IQ-adaptive gap; Hill, 2004; Demetriou et al., 2018) |
| 6 | Institutional misrecognition | The tendency of organisations to misread the profile because its strengths and impairments violate ordinary expectations of consistency | STRONG (underemployment, masking and burnout literature; Buckland, 2024; ONS, 2021) |
The framework is descriptive, not taxonomic. It implies no new clinical subtype, no biomarker, and no diagnostic threshold. Its purpose is narrower: to improve formulation where standard labels capture only fragments of the picture. In some cases "autism" or "ADHD" may name the neurodevelopmental substrate while "PNP" names the strengths-based manifestation and the misunderstandings that follow from it.
The obvious question is whether this is twice-exceptionality (§5) under a new name. It is not, and the difference is what the construct has to earn its keep on. Twice-exceptionality pairs general giftedness with a co-occurring disability, is overwhelmingly a child-and-education construct, and is silent about what follows institutionally. PNP specifies three things 2e does not: that the ability is cross-domain structural transfer rather than high ability in a single area (components 1 and 2); that the population is adults navigating administrative, employment and legal systems rather than schoolchildren; and that the pattern generates a named, correctable institutional error, the Capability-Adjustment Fallacy of §7, whose refutation is the paper's practical payload. Whether that specificity survives measurement is exactly the incremental-validity test §12 commits the construct to passing or failing in public.
In one sentence: advanced capacity in complex, self-directed domains may coexist with serious impairment in externally sequenced, low-stimulation, administratively rigid systems.
This section is stated in the author's own voice; the disclaimer that opens it binds every clause that follows. Section 4 described the pattern from the outside, as monotropism. What follows describes the same pattern from the inside, because how the mechanism feels to its subject is part of what the profile has to explain.
The author does not claim superior intelligence. What is claimed is a different attentional architecture: perseverative, recursive iteration on the same problem or subject-matter until the outcome-space is substantially exhausted. Section 4 describes this pattern from the outside as monotropism; this subsection describes it from the inside. Every scenario is considered; every third-, fourth- and fifth-order influence is traced; every question is asked, answered, and then re-asked from a different angle. The loop does not release its object until the object stops yielding. This is not brilliance. It is a wiring, and a costly one, whose costs are the executive-function impairments recorded in §5 and evidenced across the case in §9.
The pattern produces the second mechanism: expertise as trained, not innate. Recursive questioning applied over decades to many domains accumulates by ordinary compounding. Since generative AI became widely available in 2023, the loop has been massively accelerated: answers that once required months of library work now return in minutes. More questions asked, more answers returned, more domains partially mastered, deeper transfer between them, expertise reinforcing expertise. The author was not born with cross-domain competence. The competence was ingrained, trained, and built up over decades of the loop running, and the polymathy documented in §9 is the accumulated output of that loop, not a starting condition.
The convergence is the sharpest new point in this subsection. A study in the field now finds, on its own quantitative evidence, that sequential scaling, in which reasoning chains build on their previous attempts, outperforms parallel self-consistency at matched compute in 95.6% of the configurations it evaluated, across five open-source models and three reasoning benchmarks. Sharma and Chopra published the quantified result in November 2025 (Aman Sharma and Paras Chopra, "The Sequential Edge: Inverse-Entropy Voting Beats Parallel Self-Consistency at Matched Compute", arXiv:2511.02309, 4 November 2025). The author's independent formalisation appears in Paper II (now titled Paper II · The ARC Theory · Law I, the ARC Principle · Blinded Cross-Architecture Replication and the Retraction of a Super-Linear Estimate; first published January 2026), and the 8 December 2024 self-emailed manuscript bundle (SHA-256 f0d1f38f) carries the one-sided directional statement that recursion amplifies capability, with the comparison itself first formalised in Paper II.
The standing three-layer crediting discipline applies here without exception. Publication priority on the quantified 95.6% number belongs unreservedly to Sharma and Chopra. The 8 December 2024 timestamp anchors a one-sided directional statement, not the comparison; the comparison was nonetheless the equation's to make, since parallel width has no term in U = I x R^alpha, and that derivation was completed in 2026 by the operational definition of R, carrying the date of its youngest premise. Independent formalisation is acknowledged in both directions, with no influence claimed in either; the author was unaware of the Sharma and Chopra paper when he formalised the January 2026 experiments and encountered it months afterwards (biographical attestation, offered as such).
In reasoning-token architectures that make chain-of-thought visible at scale, the field is engineering into machines the iterate-question-iterate loop a monotropic mind already runs natively. DeepSeek's R1, released January 2025 [DeepSeek-AI, 2025], counts as one of the first prominent open-weights systems of that kind; the hedge is deliberate, because reasoning-model work descends from more than one lineage. From this convergence a compact hypothesis follows, rung-tagged accordingly. HYPOTHESIS The neurodivergent cognitive style documented in this paper may be the biological instance of the reasoning architecture that Sharma and Chopra find outperforms parallel self-consistency. If the hypothesis holds, it would explain both the polymathy (compounding cross-domain training) and the priority record recorded in §10: the loop was running in a single mind for years before the machines had it.
The claim is the polymathic method, never the stature. The book from which this paper descends (Eastwood, Infinite Architects: Intelligence, Recursion, and the Creation of Everything, print edition 2 January 2026, ebook 6 January 2026, ISBN 978-1806056200) already stated this position: its Author's Note (pp. 43-63) sets out the biographical basis and closes with a passage that binds the register.
The ARC Principle, the Eden Protocol, the framework of caretaker doping: these are my contributions. I make no claim to have invented the components. Recursion is well understood in mathematics. The alignment problem has been articulated by minds far more credentialed than mine. Religious traditions have explored stewardship for millennia. What I offer is the synthesis: the recognition that these separate streams are describing the same river, and that seeing them as one changes how we must act. Michael Darius Eastwood, Infinite Architects, Author's Note, p. 62. Quote reproduced verbatim.
The book's Introduction names Rumi, Teilhard de Chardin and Leibniz as the three individual polymaths whose independent architectures converge on the same account. They were chosen precisely because they were cross-domain synthesisers, not as any claim to their stature. Claiming the stature of Rumi, Teilhard or Leibniz would be absurd, and this paper does not; borrowing their method, which is what a research posture actually does, is a different act altogether. This subsection is the formal statement in the profile paper of the position the book had already taken in prose: the polymathic method is claimed; the stature is not.
Offered here strictly for orientation, and clearly outside the paper's evidential apparatus. A linear-thinking professional (the author's wife, a dentist) once agreed to role-play a student of the cross-domain framework. Her first question was which subject the lecture belonged to. The question has no answer, and that is the point. Cross-domain work has no home discipline, which is precisely why single-domain credentialled specialists were unlikely to have produced it, and why the Capability-Adjustment Fallacy's institutional home, "which box does this person fit?", is the same question in another costume. Section 7 states the fallacy formally; this vignette is only the domestic version.
This subsection is placed strictly beside §6A because it is often the next question a careful reader asks: is the profile PNP names ever found in the historical record of cross-domain synthesisers? Two safety rules govern everything that follows.
The first safety rule is that this paper does not itself apply PNP, or any diagnostic category, to any named historical figure. The analytical work of connecting historical evidence to modern categories is done by others in credentialled venues, and this paper cites their work rather than reproducing it. The second safety rule is linguistic: the operative verb here is "consistent with", not "was". PNP itself takes no support from any individual retrospective case; the paper's interest in this literature is the pattern that credentialled retrospective analyses keep finding the same trait-cluster in history's cross-domain synthesisers. All crediting flows outward.
The retrospective literature is credentialled and it exists. Catani and Mazzarello (2019), a peer-reviewed analysis in Brain by clinicians at King's College London and the University of Pavia, examined the documented pattern in Leonardo da Vinci's records (chronic non-completion of commissioned work, relentless procrastination and mind-wandering, productive bursts across day and night) and concluded that attention deficit hyperactivity disorder is the most scientifically plausible modern explanation of the recorded pattern (M. Catani and P. Mazzarello, "Leonardo da Vinci: a genius driven to distraction", Brain, 142(6), 1842-1846, DOI 10.1093/brain/awz131, 28 May 2019).
Two further monographs sit alongside it. Fitzgerald's Autism and Creativity: Is There a Link between Autism in Men and Exceptional Ability? (2004, Routledge) surveys retrospective trait-pattern assessments for a broader sample of historical figures. James's Asperger's Syndrome and High Achievement: Some Very Remarkable People (2006, Jessica Kingsley Publishers) develops the same register for a set including Isaac Newton and Albert Einstein. Neither monograph diagnoses; each documents a recurring trait-pattern in the historical record and cites the biographical evidence that carries it.
Underneath the retrospective register sits a mechanism claim at population scale. Baron-Cohen's hypersystemising theory identifies the drive to detect if-and-then causal patterns as one of the engines of invention and cross-domain analytical pursuit (S. Baron-Cohen, The Pattern Seekers: How Autism Drives Human Invention, 2020, Basic Books, developed from a longer research programme on empathy and systemising in autism).
Genome-wide association work by Warrier, Baron-Cohen and colleagues reported a positive genetic correlation between autism and systemising of rg = 0.26 (standard error 0.06, p = 3.35 × 10-5), independent of educational attainment and genetically dissociable from the social-communication component of autism (V. Warrier et al., "Social and non-social autism symptoms and trait domains are genetically dissociable", Communications Biology, 2, 328, 2019). The reported correlation does not diagnose any individual and does not settle prevalence for the profile this paper names. It does place the systemising drive characterising cross-domain analytical pursuit on a genetic axis linked to autism at the population level, rather than a purely biographical curiosity, and it is the kind of finding a reader can check.
The author's own working notes in the research corpus explored the mapping from DSM-5 ASC and ADHD criteria to the documented behaviours of celebrated historical polymaths, and contained their own section on the limits of retrospective methodology; those notes predate this paper but are cited here as working notes, not as evidence, and are consistent with the polymathic-method claim recorded in the book (§6A closing block).
Methodological caveat (stated prominently). Retrospective assessment of historical figures is contested in the clinical literature. The dead cannot be diagnosed. Historical records carry period bias, translation loss, and selection: the records that survived may not be representative of the person's ordinary functioning, and the categories being applied are twenty-first-century constructs applied to lives lived under earlier ones. This paper accordingly treats the Catani-Mazzarello, Fitzgerald and James analyses as convergent observations by qualified authors, not as clinical facts, and PNP itself draws no support from any individual historical case. HYPOTHESIS The connection between the trait-cluster that credentialled retrospective analyses keep finding in history's cross-domain synthesisers, and the profile named in §6, is a hypothesis, not an established finding.
The closing tie is short. The book from which this paper descends (Eastwood, Infinite Architects, 2026a) chose Rumi, Teilhard de Chardin and Leibniz as its exemplars in the Introduction. The choice was principled, not decorative: those three were selected precisely because they were documented cross-domain synthesisers, and because the polymathic method (§6A closing block) is what they had in common. This subsection is that principle stated more carefully. The individual figures remain the property of the retrospective literature that examines them; the method is what is claimed, and it is claimed as method, not as stature.
The Capability-Adjustment Fallacy is the inference that because a person can perform a high-level task, they cannot require adjustments for lower-level procedural demands. It appears reasonable only on the assumption that cognitive difficulty scales linearly: that anything "easier" than the hard thing a person can do must also be within reach. In neurodivergent life, that assumption frequently fails. A high-complexity research problem may be easier than a simple administrative task, because the former is meaningful, patterned, self-directed, and intrinsically regulating, while the latter is fragmented, externally paced, low-reward, and executive-function heavy.
This is not a rhetorical move. It is anchored in clinical guidance. NICE CG142 directs all health and social care professionals to understand "the potential discrepancy between intellectual functioning as measured by IQ and adaptive functioning as reflected, for example, by difficulties in planning and performing activities of daily living including education or employment" (NICE, 2021). DSM-5-TR records that autistic difficulties "may be masked by learned strategies in later life," so that high apparent functioning can conceal, rather than refute, support needs (APA, 2022). The executive-function literature shows that differences in autism are real but domain-specific rather than globally uniform (Hill, 2004; Demetriou et al., 2018), and that everyday dysexecutive difficulties can be present even where laboratory task performance is intact. A person can therefore be cognitively advanced and adaptively impaired at the same time; in PNP that discrepancy may be extreme and institutionally confusing.
The fallacy has a practical cost: it converts evidence of competence into evidence against need. A litigant who files a sophisticated written argument may be denied extra time, filing support, or procedural clarification because the decision-maker sees capability rather than disability. That is precisely backwards. The sophistication of the substantive work shows that the person's cognitive strengths are real; it says little about whether they can manage routine procedural burdens without disadvantage.
The proper test is functional and demand-specific: What is the task? What cognitive, sensory, social, administrative, and time demands does it impose? Which parts of the person's profile make that task disproportionately difficult? Which reasonable changes would reduce the disadvantage while preserving fairness to others and the integrity of the process? That is a more serious analysis than asking whether the person appears intelligent.
A person may be in top sets for subjects that engage their interests and at the bottom in subjects that do not. Read morally, this becomes laziness or inconsistency; read neurodevelopmentally, it may reflect interest-based attentional regulation, executive asymmetry, and sensory or social load. Uneven attainment is common in neurodivergence, and strengths-based assessment argues that it should be described as a pattern rather than a character flaw.
The outcomes data are stark. Per the Office for National Statistics' first release to identify autistic respondents, autistic adults had the lowest employment rate of any disability group (ONS, 2021). The Buckland Review of Autism Employment found that only around three in ten working-age autistic people are in work, against roughly five in ten for all disabled people and eight in ten for non-disabled people; that autistic graduates are about twice as likely as non-disabled graduates to be unemployed fifteen months after graduating; and that autistic workers face the largest pay gap of any disability group (Buckland, 2024). Capability is plainly not being converted into recognition. Masking compounds this: camouflaging is associated with higher measured ability, later or missed diagnosis, and worse mental-health outcomes (Hull et al., 2017; Cage and Troxell-Whitman, 2019; Livingston et al., 2019), and sustained mismatch between expectation and support is a route to autistic burnout: chronic exhaustion, skill loss, and reduced tolerance, distinct from ordinary occupational burnout (Raymaker et al., 2020).
Legal systems combine high cognitive complexity with intense procedural rigidity: deadlines, bundles, filing sequences, service evidence, transcript requests, order review. Those demands are executive-function intensive, not intellectually simple for everyone. A litigant in person can be strong in legal reasoning and weak in procedural administration; treating the former as proof that the latter needs no support produces unfairness, and is the Capability-Adjustment Fallacy applied to access to justice.
Generative AI can accelerate drafting, research, and synthesis, but it does not originate the judgement, selection, cross-domain integration, and strategic direction that decide what to do with that output. For a person with a strong pre-existing pattern of cross-domain synthesis, AI functions as a throughput multiplier: it can make strengths more visible while leaving executive-function impairments untouched. The presence of AI-assisted work should therefore not be used to deny either the person's own cognitive contribution or their disability-linked support needs. This is the paper's most modest and least evidenced claim, and is offered as a reasoned position rather than a demonstrated finding.
Named cognitive profiles usually rest on aggregated questionnaires or anonymised vignettes. This paper can do something unusual: present a single, fully documented instance in which the profile and the fallacy are recorded in three independent documentation streams, generated by different institutions for different purposes, none of them constructed for this paper. The subject is the author. Section 15 addresses the ethics and the epistemics of that choice; this section states the evidence.
The author holds adulthood diagnoses of autism spectrum condition and ADHD, with independent psychiatric documentation describing precisely the asymmetry the profile predicts: unusually strong self-directed learning, pattern recognition and cross-domain synthesis in high-interest domains, coexisting with clinically significant executive-function impairment in routine administration, timekeeping and procedural work. The clinical documentation is held privately for obvious reasons, and is available to qualified reviewers under appropriate conditions. Nothing in this paper asks the reader to take the clinical stream on trust: the paper's argument survives even if the reader brackets it entirely, because the other two streams are public.
The author has conducted complex multi-track civil litigation in person in the courts of England and Wales. The record of that litigation contains, simultaneously: substantive written work of a standard the opposing parties engaged specialist counsel to answer, and repeated applications for modest procedural adjustments under the Equality Act 2010 and CPR PD1A, disclosed disability, and documented difficulty with exactly the procedural demand classes the profile predicts (deadlines, sequencing, administrative correspondence). The record therefore contains both halves of the fallacy in a single institutional file: output treated as evidence of capability, and adjustment requests assessed against that appearance of capability. The paper takes no position here on the merits of any live proceedings; the record is cited only as a documented instance of the demand-specific asymmetry, and it is a matter of court record rather than self-report.
The third stream is the one no questionnaire could fabricate: a dated body of cross-domain research output, whose signature-time assertions are the mail provider's rather than an RFC 3161 timestamp. On 8 December 2024, the author self-emailed manuscripts totalling roughly 221,236 words; the export carries a published SHA-256 hash, and its date is carried by the provider's cryptographic signature (sender-domain DKIM with Google ARC). The corpus that followed spans AI alignment (an evaluation-blinding protocol in which unblinding reversed the sign of a result), scaling theory (a co-scaling stability condition, demonstrated so far only in simulation, with its measurement on real systems unsolved), mathematics (a functional-equation account of why two independent derivations of the metabolic scaling exponent converge, sent to the leading authors of both derivations), and physics-adjacent classification work on time crystals. Independently sourced external events have since converged on the corpus's structural claims; the programme's graded convergence register sets each one out in its own row, gives that row an evidence class, and publishes no combined figure, since technical experiments, theorems, engineering proposals and policy developments cannot legitimately be summed as one evidential quantity (a snapshot retained from before the grading, dated 2 July 2026, put the number at nineteen, an exploratory count kept for reference; this paper does not itself test the convergence, and the programme’s audit of how many of the convergences are independent is a draft, preregistered by timestamp and awaiting human submission: its data already exist, and its registered analysis has not been run). That register is public: every row names one primary source, and wherever the dates require it, priority is credited to others in plain terms.
Component 1 of the profile (cross-domain depth) and component 2 (pattern transfer) are ordinarily the hardest to evidence: self-report is unreliable and credentials measure institutional access, not cognition. A hash-timestamped corpus solves both problems. The dates cannot be backfilled, the domains are objectively distinct, and the transfer is inspectable in the documents themselves: the same recursion-and-correction structure is applied to quantum error correction, metabolic scaling, and reinforcement-learning drift, with each application dated. Whatever one thinks of the corpus's conclusions, its existence, dates and breadth are checkable facts, and they are exactly the facts the profile predicts and the credential system failed to register.
Provenance note (19 September 2026): what the carrier establishes, and what reading establishes. A domain signature over a message establishes the integrity of the signed bytes as sent, and asserts a carriage time on the signing domain's authority. It does not establish semantic merit, originality, causal origin or construct validity, because none of those is a function of the bytes and the signing time alone. The two properties the paragraph above relies on, that the domains are objectively distinct and that the transfer is inspectable in the documents themselves, are matters for assessment of the documents against stated criteria; the hash and the signature do not deliver them, and reading alone does not settle a substantive cross-domain transfer without those criteria and an assessor applying them. "The dates cannot be backfilled" is read at the strength this paper states earlier: the signature-time assertion is the provider's rather than an RFC 3161 timestamp, so it is a trusted-third-party assertion, and the sender controls both the content and the moment of sending. The same limit governs the next paragraph: the third stream attests carriage and byte integrity, not the asymmetry, which is a property of the content assessed against criteria. Unchanged: the existence, dates and breadth of the corpus as checkable facts; this paper's earlier and governing statement of the carrier's status; and the agreement of the first two streams.
The three streams are independent in the relevant sense: a psychiatrist, a court, and a mail server have no common interest and no coordination. Their agreement on the same asymmetry, high structured output here, impaired procedural functioning there, is the strongest form of documentation available for a single case, and it converts the profile from an introspective claim into an auditable one.
The framework's position in time can be recorded precisely, because 2026 produced independent results that bear directly on its central claim.
| Date | Event | Anchor |
|---|---|---|
| 8 Dec 2024 | Author's manuscripts state the control-failure thesis for AI alignment and the embedded-correction remedy; the corpus that instantiates PNP components 1 to 4 begins its public timestamp record | SHA-256 published; date carried by the provider’s cryptographic signature (DKIM with Google ARC) |
| May 2025 | Hernández-Espinosa, Abrahão, Witkowski and Zenil circulate their preprint proving that full AI-human alignment is mathematically impossible for Turing-complete systems and proposing diversity and mutual influenceability among competing artificial agents, under the title’s neurodivergence analogy, as the counterbalancing mechanism; their subject is artificial agents, and the human-cognition reading is this paper’s own claim, never theirs | arXiv:2505.02581 |
| 14 Apr 2026 | The same argument is published in PNAS Nexus | DOI 10.1093/pnasnexus/pgag076 |
| 26 Jun 2026 | Gumbau Mezquita publishes formal results (an unverifiability theorem and a soundness-completeness-tractability trilemma) whose stated conceptual foundation is the Hernández-Espinosa paper | arXiv:2606.28639 |
| 3 Jul 2026 | This paper names the profile (PNP) and the Capability-Adjustment Fallacy, and places the documented case on the public record | This paper; OSF |
Three points of precision. First, the author's manuscripts precede the preprint by roughly five months and the formal proof by eighteen; the direction of every gap is checkable against published hashes. Second, no causal claim is made in either direction: neither group had any way to know of the author's work, and that mutual ignorance is what gives the convergence its evidential value. Third, the chain has a striking internal structure: a paper arguing that neurodivergent cognition protects alignment research from shared blind spots served as the stated foundation for the theorem that formalised what a neurodivergent researcher had argued structurally eighteen months earlier. The paper about minds like the author's inspired the proof that vindicated the author's thesis. The chain is documented end to end; every link is independently checkable.
Reading note (19 September 2026), governing the three points of precision above. First, what a public record evidences is the absence of any identified citation or connection, which is weaker than the absence of knowledge; a universal negative about what other people knew is not established by a record of what was published. Second, the external preprint's subject is artificial agents, and the human-cognition reading is this paper's own claim and never theirs, exactly as the table above states; the shorter phrase in the paragraph above is read against that table, which governs. Third, the 2026 formal results are an unverifiability theorem and a soundness, completeness and tractability trilemma. A limit on universal verification is relevant to the control-failure half of the 8 December 2024 content under the theorem's own stated assumptions. It does not establish the embedded-correction remedy, because a limit on verification says nothing about whether embedding correction is the remedy; and it does not by itself establish the whole of the control-failure half either, since that would need the theorem's assumptions and the claim's content compared term by term, which the paragraph above does not do. "Vindicated" is therefore read as relevance to the first half under stated assumptions, and not as establishment of either half. Unchanged: the dates and the direction of every gap; the disclaimer earlier in this paper that no external research result is claimed to have been caused by the author's cognition; and the table's own scope statement.
This paper and Laws IV and V, 3 October 2026. The theory has been stated with five laws since 24 September 2026. Law IV, the ARC Persistence Law, is carried by proposition P15, and Law V, the ARC Embedding Law, by propositions P12 and P21. The three propositions are registered; neither law is registered as a law, and neither has been tested. Their mathematics is set out in Laws IV and V: the mathematics, version 1.0, of 3 October 2026. This paper documents a profile and tests no part of the theory. It bears on Law V only as a dated record. Section 1 and the table above report that the author’s manuscripts of 8 December 2024 state the embedded-correction remedy, and the quotation in section 6A names the framework of caretaker doping among the author’s contributions. In the theory registration, the book’s caretaker doping is embedded correction, carried by P21, and P21 is one of the two propositions that carry Law V. The reading note above already states that the 2026 formal results do not establish that remedy. This paper compares no correction that takes part in each revision round with correction that sees only the finished output, varies no build order, and measures no gain installed by external post-training under later capability-directed training. Nothing in it is a test of P15, P12 or P21. Naming the two laws changes nothing printed here: every claim, date and limit of this paper stands as it was.
The convergence chain points at a field-level implication that goes beyond any single person. If Hernández-Espinosa and colleagues are right that uniform compliance cannot deliver alignment and that behavioural diversity is protective, then the composition of the research field itself becomes a safety variable. Uniform minds, trained in the same institutions on the same curricula, tend to miss the same things in the same direction at the same time. That is a monoculture risk, and it is invisible from inside the monoculture.
The research programme this paper belongs to offers three concrete instances (offered as exploratory analysis; none of this paper's own quantitative content has yet been tested). Paper IV.d reports that for two model families the direction of the alignment-depth relationship changed once scoring was blinded; unblinded scoring is a failure mode that persisted across the field partly because evaluators shared assumptions about their own neutrality. One recursive-amplification structure runs through quantum error correction, biological scaling and machine reasoning, and the Foundational paper set it out; drawing that map at all meant handling those literatures as a single subject, which is exactly what disciplinary training discourages. Paper X derived a stability condition for recursive self-improvement (correction must out-scale drift) whose central move, refusing to treat capability and correction as separable, cuts against the field's default decomposition. None of this demonstrates that neurodivergent cognition is necessary for such work. It demonstrates that the work in fact came from outside the monoculture, on documented dates, and that the field's own 2026 results now predict exactly that pattern.
The implication for institutions is uncomfortable and practical: exclusion of atypical cognitive profiles from research, funding and legal participation is not only an equity failure. On the field's own emerging evidence, it is a technical risk. The Capability-Adjustment Fallacy, operating in hiring committees, grant panels and courtrooms, is one of the mechanisms by which that exclusion happens.
Scope note, added 19 September 2026: the conclusion above states a consequent, and is read with its antecedent in place. The claim at the head of this section is printed in conditional form, and the conclusion above inherits that condition. "On the field's own emerging evidence" identifies where the evidence comes from; it does not restore the conditional and it does not supply a comparison. The necessity limit printed above governs here too: this paper states that none of this demonstrates that neurodivergent cognition is necessary for such work. These passages also do not supply a comparison that would establish institutional composition, rather than the documented individual route, as the cause of the findings, so the step from the diversity result about artificial agents to a claim about human institutional composition is read as a proposed bridge and not as a validated implication. Unchanged: the conditional statement at the head of this section; the exploratory-analysis marker and the necessity limit above, both of which are already printed and both of which govern; and the account of the Capability-Adjustment Fallacy in the paragraph above.
For the construct to be scientifically serious rather than self-descriptive, it must be operationalised and exposed to refutation.
Measurable components. PNP would be indicated by: (a) documented cross-domain output over time; (b) evidence of self-directed learning; (c) demonstrable pattern transfer between fields; (d) a measurable asymmetry between performance on preferred, high-interest versus non-preferred, low-stimulation tasks; (e) clinical evidence of executive-function impairment in routine administration; and (f) evidence of institutional under-recognition, where outputs and impairments are treated as mutually inconsistent.
Conditions under which PNP should be abandoned. The hypothesis fails if it cannot be reliably distinguished from (i) ordinary high openness to experience, (ii) ordinary giftedness, (iii) narcissistic or grandiose self-description, or (iv) general executive dysfunction; or if naming it adds no predictive or practical value beyond existing autism and ADHD formulations. These are real tests, and the paper welcomes them. The decisive empirical question is whether PNP incrementally predicts the preferred versus non-preferred task asymmetry and the institutional under-recognition outcome beyond what existing instruments already capture. If a validated measure showed PNP variance collapsing into those existing constructs, the construct should be discarded, and this paper's own falsification register commits to saying so publicly.
Research questions. Can clinicians and researchers reliably identify a cross-domain strengths-and-needs profile? How often does it appear among autistic and ADHD adults with histories of self-directed expertise? Does naming it improve clinical reporting, workplace support, educational accommodation, or court participation? The documented case in Section 9 supplies a template for the documentation standard such research would need: independent record streams rather than self-report alone.
The clinically defensible request is not that a psychiatrist diagnose "polymathy." It is that the psychiatrist provide a full adult autism assessment and if clinically justified, describe the person's strengths, support needs, thinking style, executive-function profile, and functional consequences in real settings. That is consistent with NHS and NICE reporting principles and with neuro-affirming practice (NICE, 2021; NAIT, 2024).
A report could properly state, in substance, that the individual presents with autism-spectrum features and ADHD-related executive-function difficulties, together with an unusually strong profile of self-directed learning, pattern recognition, cross-domain synthesis, and productivity in selected high-interest domains; and that the same neurodevelopmental profile produces marked functional disadvantage in routine administration, timekeeping, task-switching, and low-stimulation procedural work. The clinically important addition is the explicit statement that the asymmetry itself is the finding: high-level output should not be read as evidence that support needs are absent.
This formulation is more powerful than grand claims precisely because it remains sober, observable, and demand-specific. It asks the clinician to explain functioning, not to certify genius or legal correctness.
The court-facing formulation should be precise and restrained, and should ask the court only to act consistently with its own access-to-justice obligations and with the functional evidence in any diagnostic report.
The statutory architecture supports this directly. The Equality Act 2010 imposes a duty to make reasonable adjustments where a provision, criterion, or practice places a disabled person at a substantial disadvantage (s.20), and in the exercise of public functions that duty is anticipatory. The Public Sector Equality Duty requires due regard to advancing equality of opportunity (s.149), and s.149(6) expressly provides that compliance "may involve treating some persons more favourably than others": statutory authority that adjustment is not unfair advantage. In civil litigation, CPR 1.1 and Practice Direction 1A require the court, so far as practicable, to ensure that parties are on an equal footing, can participate fully, and can give their best evidence, and recognise that vulnerability, including mental impairment or disability, may both impede participation and diminish the quality of evidence (CPR PD1A, 2021).
Within that architecture, the Capability-Adjustment Fallacy has a precise legal form: it substitutes a global impression of ability for the statutory question, which is always demand-specific disadvantage. A tribunal that reasons "this litigant writes able submissions, therefore no substantial disadvantage arises from these deadlines, this correspondence load, this hearing format" has answered a question the statute does not ask, and left unanswered the one it does. The correct sequence is the functional one set out in Section 7: identify the demand, identify the disadvantage on the evidence, identify the adjustment, and test it against fairness to the other parties. Nothing in that sequence turns on how clever the person's best work looks.
Single-case evidence. Section 9 documents one instance, however thoroughly. A documented case can instantiate a construct and refute the claim that no such profile exists; it cannot establish prevalence, and it cannot by itself validate the construct. The paper's population-level claims rest on the cited literatures, not on the case.
Author as subject. The author is the documented case, which creates an obvious interest. Three mitigations are structural rather than rhetorical: the evidence streams are institutional records not authored for this paper; the falsifiability conditions in Section 12 are pre-committed and public; and the research programme this paper belongs to has already retracted one of its own quantitative claims when replication failed, with the retraction kept on the public record. Readers should nonetheless weight the case study as what it is: an unusually well-documented n of 1.
Privacy and dignity. Publishing one's own diagnostic status invites both stigma and its mirror image, inspiration narrative. The paper intends neither. The clinical stream is disclosed at the minimum level of detail the argument requires; the clinical documents themselves remain private. No inference should be drawn from this paper about any other person's profile, and the profile itself, as Section 2 states, confers no superiority: its costs are clinical, documented, and lifelong.
Live litigation. The court record is cited as an instance of a documented asymmetry, not as commentary on the merits of any proceedings, which remain for the courts that hear them.
What would change this paper. A validated instrument showing PNP variance fully absorbed by existing constructs; adult-population studies contradicting the asymmetry pattern; or failures of the documentation standard in Section 9 under qualified scrutiny. Any of these would be recorded on the programme's public falsification register.
The Polymathic Neurodivergent Profile is offered as a working hypothesis, not an established finding. It is falsifiable, it is open to correction, and it makes no claim to diagnostic status. It is a descriptive framework for a pattern that is real, institutionally consequential, and systematically misread; and it now has something such frameworks rarely have: a fully documented live case across independent clinical, legal and cryptographic records, and an independent 2026 literature, from formal logic to PNAS Nexus, converging on its central implication that cognitive diversity is not a concession but a resource.
The Capability-Adjustment Fallacy, the inference that visible high performance is evidence that adjustments are unnecessary, is unsafe. It is contradicted by clinical guidance, incompatible with the statutory architecture of the Equality Act 2010 and CPR PD1A, and on the field's own emerging evidence, technically costly to the one research problem that can least afford shared blind spots. The correction is not special treatment. It is the recognition that capability in one domain does not remove support needs in another, and that the same cognitive architecture which produces extraordinary output in favoured domains may be unable to produce consistent output in routine ones without structural accommodation.
The researcher who wrote this paper is the case study. The manuscript hashes are published. The court record exists. The clinical documentation exists. The convergence register is public, with priority credited to others wherever the dates demand it. The question this paper leaves with its readers is the one the fallacy prevents institutions from asking: not "how capable does this person seem?" but "what does this specific demand cost this specific mind, and what would fairness actually require?"
Note (3 October 2026): what ten corrected passages and three note headings printed before. The working-status box at the head of this paper said that no claim in it was on trial through a programme draft study. The incremental-validity test that section 12 sets out is drafted as a programme follow-on study and preregistered by timestamp, so the box now gives that state, and the reading key adds it. In section 6A, the quotation from the book’s Author’s Note left out two words of its first sentence, the name of a framework, and now gives the sentence as the book prints it. The same section said that the field had demonstrated that sequential, iterative, self-questioning reasoning outperforms parallel sampling on the same problems, and its hypothesis spoke of the reasoning architecture the field had just proved superior. Both passages now state Sharma and Chopra’s result in their own verb: they find that sequential scaling outperforms parallel self-consistency at matched compute in 95.6% of the configurations they evaluated. In section 9, the account of the third stream described its co-scaling stability condition as having a first real-model demonstration. Paper X withdraws its real-model drift run of 2 July 2026 as mechanism evidence and states that the condition is demonstrated only in simulation, with its measurement on real systems unsolved, and the passage now says so. The same account said that no programme draft study covers the convergence. The programme’s audit of how many of the convergences are independent is a draft study, preregistered by timestamp, and the account now gives its state. The evidence box that follows said that the same structure was applied to quantum error correction, metabolic scaling and reinforcement-learning drift years before independent groups published parallel structures in each domain; that clause is withdrawn, and the box now says that each application is dated. In section 11, the exploratory-analysis marker said that none of this paper’s own quantitative content is tested by a programme draft study. The test of section 12 is drafted as such a study, so the marker says instead that none of that content has yet been tested. The same section said that Paper IV.d showed that unblinded AI evaluation can reverse the sign of an alignment result; it now gives Paper IV.d’s own finding, that for two model families the direction of the alignment-depth relationship changed once scoring was blinded. The epistemic-status line said that what the programme names Laws are conjectures under registered adversarial test; it now gives Laws I to III and Laws IV and V their separate status. The headings of the three notes of 19 September 2026, in sections 9, 10 and 11, said that each note was added after an owner review; they now give the date without that clause. Unchanged: every other claim, date and limit of this paper, and the text of its three notes of 19 September 2026.
American Psychiatric Association (2022). Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, Text Revision (DSM-5-TR). APA Publishing. DOI 10.1176/appi.books.9780890425787.
Araki, M. E. (2018). Polymathy: A new outlook. Journal of Genius and Eminence, 3 (Fall 2018), 66-82. DOI 10.18536//jge.2018.04.3.1.06, as Crossref registers the article. No official online source: on 3 October 2026 the DOI resolved to a domain that no longer serves the journal.
Ashinoff, B. K., and Abu-Akel, A. (2021). Hyperfocus: the forgotten frontier of attention. Psychological Research, 85, 1-19. DOI 10.1007/s00426-019-01245-8.
Baron-Cohen, S. (2020). The Pattern Seekers: How Autism Drives Human Invention. Basic Books. Publisher: Hachette (Basic Books). The publisher’s page, read on 3 October 2026, gives ISBN 9781541647138 (ebook) and lists a trade paperback; no other ISBN could be read there. Developed from a longer research programme on empathy and systemising in autism, hypersystemising as an engine of cross-domain analytical pursuit.
Buckland, R. (2024). The Buckland Review of Autism Employment: report and recommendations. Department for Work and Pensions. GOV.UK publication.
Cage, E., and Troxell-Whitman, Z. (2019). Understanding the reasons, contexts and costs of camouflaging for autistic adults. Journal of Autism and Developmental Disorders, 49, 1899-1911. DOI 10.1007/s10803-018-03878-x.
Catani, M., and Mazzarello, P. (2019). Leonardo da Vinci: a genius driven to distraction. Brain, 142(6), 1842-1846. DOI 10.1093/brain/awz131, published 28 May 2019. Peer-reviewed retrospective trait-pattern analysis by clinicians at King's College London and the University of Pavia; cited as credentialled retrospective analysis, not clinical fact.
Civil Procedure Rules, Part 1: Overriding Objective, rule 1.1 (justice.gov.uk, updated 1 October 2024).
Civil Procedure Rules, Practice Direction 1A: Participation of Vulnerable Parties or Witnesses (justice.gov.uk, updated 31 May 2022).
Cope, R., and Remington, A. (2022). The strengths and abilities of autistic people in the workplace. Autism in Adulthood, 4(1), 22-31. DOI 10.1089/aut.2021.0037.
DeepSeek-AI (2025). DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning. arXiv:2501.12948 (version 1 of 22 January 2025; the arXiv record gives no release date for the model).
Demetriou, E. A., et al. (2018). Autism spectrum disorders: a meta-analysis of executive function. Molecular Psychiatry, 23, 1198-1204. DOI 10.1038/mp.2017.75.
Eastwood, M. D. (2026a). Infinite Architects: Intelligence, Recursion, and the Creation of Everything. ISBN 978-1806056200. Print 2 January 2026; ebook 6 January 2026. Author's Note pp. 43-63.
Eastwood, M. D. (2026b). Paper X · The ARC Theory · Law II, the ARC Co-Scaling Law · Correction That Out-Scales Drift. OSF: 10.17605/OSF.IO/BSE2Q.
Eastwood, M. D. (2026c). Paper IV.d · The ARC Theory · The Effect of Blinding on AI Alignment Evaluation. OSF: 10.17605/OSF.IO/2S3E6.
Eastwood, M. D. (2026d). Paper II · The ARC Theory · Law I, the ARC Principle · Blinded Cross-Architecture Replication and the Retraction of a Super-Linear Estimate. OSF: 10.17605/OSF.IO/8FJMA. Independent formalisation, January 2026; the 8 December 2024 self-emailed manuscript bundle (SHA-256 f0d1f38f) carries the one-sided directional statement.
Eastwood, M. D. (2026e). Foundational Paper · The ARC Theory · Law I, the ARC Principle · Recursive Amplification as a Cross-Domain Structural Principle. OSF: 10.17605/OSF.IO/Y7QGD.
Eastwood, M. D. (2026f). Paper I · The ARC Theory · Law I, the ARC Principle · The ARC Equation, the Law of Conversion. OSF: 10.17605/OSF.IO/B6N27.
Eastwood, M. D. (2026g). Paper III · The ARC Theory · Law II, the ARC Co-Scaling Law · The Alignment Scaling Problem: Why External AI Safety Approaches Cannot Scale With Recursive Capability. OSF: 10.17605/OSF.IO/HQCGF.
Eastwood, M. D. (2026h). Paper XI · The ARC Theory · Law I, the ARC Principle · Convergent Evidence for Recursive Amplification: a Graded Cross-Domain Register. OSF: 10.17605/OSF.IO/DC9GW.
Eastwood, M. D. (2026i). Monograph · The ARC Theory · The Hyperspace Recursive Intelligence Hypothesis: a Testable Creation Theory (HRIH). OSF: 10.17605/OSF.IO/UYDXQ.
Eastwood, M. D. (2026j). Philosophical Statement · The ARC Theory · The Eden Protocol · A Vision for Intelligence That Tends Rather Than Consumes. OSF: 10.17605/OSF.IO/9M3DG.
Eastwood, M. D. (2026k). Laws IV and V: the mathematics, version 1.0. First published 3 October 2026.
Equality Act 2010, 2010 c. 15. UK Public General Acts, legislation.gov.uk.
Ferreira, G. J. (2025). Commentary: Toward a more comprehensive autism assessment: the Survey of Autistic Strengths, Skills, and Interests. Frontiers in Psychiatry, 16, 1552960. DOI 10.3389/fpsyt.2025.1552960.
Fitzgerald, M. (2004). Autism and Creativity: Is There a Link between Autism in Men and Exceptional Ability? Routledge. DOI 10.4324/9780203491591. Retrospective trait-pattern assessment across a set of historical figures; cited as credentialled retrospective analysis, not clinical fact.
Foley-Nicpon, M., Allmon, A., Sieck, B., and Stinson, R. D. (2011). Empirical investigation of twice-exceptionality: where have we been and where are we going? Gifted Child Quarterly, 55(1), 3-17. DOI 10.1177/0016986210382575.
Garau, V., et al. (2023). Development and validation of a novel self-report measure of monotropism in autistic and non-autistic people: the Monotropism Questionnaire. OSF preprint (cited as preliminary). DOI 10.31219/osf.io/ft73y.
Gentner, D. (1983). Structure-mapping: a theoretical framework for analogy. Cognitive Science, 7(2), 155-170. DOI 10.1207/s15516709cog0702_3.
Gick, M. L., and Holyoak, K. J. (1983). Schema induction and analogical transfer. Cognitive Psychology, 15(1), 1-38. DOI 10.1016/0010-0285(83)90002-6.
Grove, R., Hoekstra, R. A., Wierda, M., and Begeer, S. (2018). Special interests and subjective wellbeing in autistic adults. Autism Research, 11(5), 766-775. DOI 10.1002/aur.1931.
Gumbau Mezquita, J. P. (2026). The Undecidability of Artificial General Intelligence (AGI) Alignment. arXiv:2606.28639.
Hernández-Espinosa, A., Abrahão, F. S., Witkowski, O., and Zenil, H. (2026). Neurodivergent influenceability in agentic AI as a contingent solution to the AI alignment problem. PNAS Nexus, 5(4), pgag076. DOI 10.1093/pnasnexus/pgag076 (preprint: arXiv:2505.02581).
Hill, E. L. (2004). Executive dysfunction in autism. Trends in Cognitive Sciences, 8(1), 26-32. DOI 10.1016/j.tics.2003.11.003.
Hull, L., et al. (2017). "Putting on my best normal": social camouflaging in adults with autism spectrum conditions. Journal of Autism and Developmental Disorders, 47, 2519-2534. DOI 10.1007/s10803-017-3166-5.
James, I. (2006). Asperger's Syndrome and High Achievement: Some Very Remarkable People. Jessica Kingsley Publishers. ISBN 9781843103882. No official online source: the publisher’s catalogue does not list the book (searched 3 October 2026). Retrospective trait-pattern assessment for a set including Isaac Newton and Albert Einstein; cited as credentialled retrospective analysis, not clinical fact.
Livingston, L. A., Shah, P., and Happé, F. (2019). Compensatory strategies below the behavioural surface in autism: a qualitative study. The Lancet Psychiatry, 6(9), 766-777. DOI 10.1016/S2215-0366(19)30224-X.
Manalili, M. A. R., et al. (2023). From puzzle to progress: how engaging with neurodiversity can improve cognitive science. Cognitive Science, 47(2), e13255. DOI 10.1111/cogs.13255.
Murray, D., Lesser, M., and Lawson, W. (2005). Attention, monotropism and the diagnostic criteria for autism. Autism, 9(2), 139-156. DOI 10.1177/1362361305051398.
NAIT (2024). Neuro-affirming Reports: Guidance for health practitioners. National Autism Implementation Team, Scotland, March 2024. Its own suggested citation names the authors: Day, H., Rutherford, M., Johnston, L., Utley, I., MacIver, D., Kourti, M., Srireddy, P., and Meff, T. Official PDF.
NHS (2026). Autism assessments. nhs.uk, page last reviewed 6 May 2026.
NICE (2021, updated 2025). Autism spectrum disorder in adults: diagnosis and management. Clinical guideline CG142. nice.org.uk/guidance/cg142.
Office for National Statistics (2021). Outcomes for disabled people in the UK: 2020. ONS release.
Raymaker, D. M., et al. (2020). "Having all of your internal resources exhausted beyond measure and being left with no clean-up crew": defining autistic burnout. Autism in Adulthood, 2(2), 132-143. DOI 10.1089/aut.2019.0079.
Reis, S. M., Baum, S. M., and Burke, E. (2014). An operational definition of twice-exceptional learners: implications and applications. Gifted Child Quarterly, 58(3), 217-230. DOI 10.1177/0016986214534976.
Root-Bernstein, R. (2009). Multiple giftedness in adults: the case of polymaths. In International Handbook on Giftedness, 853-870. Springer. DOI 10.1007/978-1-4020-6162-2_42.
Russell, G., Kapp, S. K., Elliott, D., Elphick, C., Gwernan-Jones, R., and Owens, C. (2019). Mapping the autistic advantage from the accounts of adults diagnosed with autism: a qualitative study. Autism in Adulthood, 1(2), 124-133. DOI 10.1089/aut.2018.0035.
Sharma, A., and Chopra, P. (2025). The Sequential Edge: Inverse-Entropy Voting Beats Parallel Self-Consistency at Matched Compute. arXiv:2511.02309, 4 November 2025. Independent quantification of the sequential-over-parallel finding; publication priority on the quantified number credited unreservedly per the three-layer discipline.
Warrier, V., Toro, R., Won, H., et al. (2019). Social and non-social autism symptoms and trait domains are genetically dissociable. Communications Biology, 2, 328. DOI 10.1038/s42003-019-0558-4. Reports positive genetic correlation between autism and systemising rg = 0.26 (standard error 0.06, p = 3.35 × 10-5), independent of educational attainment and genetically dissociable from the social-communication component of autism.
Woods, S. E. O., and Estes, A. (2023). Toward a more comprehensive autism assessment: the Survey of Autistic Strengths, Skills, and Interests. Frontiers in Psychiatry, 14, 1264516. DOI 10.3389/fpsyt.2023.1264516.
Paper C v1.10 - 5 October 2026: the title now follows the programme’s uniform order (paper, its law or framework, its own title), Paper C · Standalone · Polymathy and Neurodivergent Cognition: the Polymathic Neurodivergent Profile (PNP) and the Capability-Adjustment Fallacy; until this version it read Polymathy and Neurodivergent Cognition: the Polymathic Neurodivergent Profile (PNP) and the Capability-Adjustment Fallacy, under the heading Polymathy and Neurodivergent Cognition. Mentions of the programme’s papers in this paper now carry their uniform titles. No registered law, prediction, kill condition or dependency changes; no result is added; no date of any claim moves.
Paper C v1.9 - 3 October 2026: ten passages are corrected where they stand, and a dated note before the references records what each printed before. The working-status box said that no claim in the paper was on trial through a programme draft study; the incremental-validity test of section 12 is drafted as a programme follow-on study, preregistered by timestamp and awaiting human submission, and the box and the reading key now say so. In section 6A, the quotation from the book’s Author’s Note now gives its first sentence whole, as the book prints it. The same section and its hypothesis now state Sharma and Chopra’s result in their own verb, that they find sequential scaling outperforms parallel self-consistency at matched compute, where they said that the field had demonstrated and proved it. In section 9, the third stream’s scaling-theory work no longer carries a first real-model demonstration: Paper X withdraws its real-model drift run of 2 July 2026 as mechanism evidence and states that its condition is demonstrated only in simulation. Section 9 also said that no programme draft study covers the convergence; the programme’s audit of how many of the convergences are independent is a draft study, preregistered by timestamp, and the section now gives its state. The evidence box in section 9 no longer says that the corpus applied its structure years before independent groups published parallel structures in each domain; it says that each application is dated. In section 11, the exploratory-analysis marker now says that none of this paper’s own quantitative content has yet been tested, where it said that none is tested by a programme draft study, and the sentence on Paper IV.d gives its own finding, that for two model families the direction of the alignment-depth relationship changed once scoring was blinded, where it said that unblinded evaluation can reverse the sign of an alignment result. The epistemic-status line now gives Laws I to III and Laws IV and V their separate status. The headings of the three notes of 19 September 2026 no longer say how the notes came to be added; their dates and text are unchanged. A dated note in section 10 states how this paper stands to Laws IV and V: the paper reports what the manuscripts of 8 December 2024 state about embedded correction, and it tests neither law. The abstract now says that the paper sets out the Polymathic Neurodivergent Profile, and section 6 that the profile is a descriptive construct stated as a hypothesis; both said that it was proposed. Twelve reference entries are corrected from their official records: the Araki entry printed a DOI registered to another article and now gives the one Crossref registers for it, which no longer leads to the journal, with the issue as Crossref gives it; five titles printed in short form are given in full; the NAIT guidance, the DeepSeek-R1 report and the Warrier author list follow their own records, the NAIT entry with the authors its own citation names and the DeepSeek-R1 entry with the date of its first arXiv version; the James entry no longer presents a bookseller’s page as the publisher’s record, since the publisher’s catalogue does not list the book; the Baron-Cohen entry gives the ISBN that the publisher’s page shows; and the Fitzgerald entry, with its mention in section 6B, gives the imprint its record names. Ten works cited in the text and missing from the references are added from their official records: Part 1 and Practice Direction 1A of the Civil Procedure Rules, the Equality Act 2010, the NHS page on autism assessments, and six of the author’s own works, Papers I, III and XI, the HRIH paper, the Eden Protocol’s philosophical vision and the report on the mathematics of Laws IV and V. The works the paper cites are linked to their official sources where one can be reached, and each of the author’s own papers to its DOI. Substantive: the ten corrected passages and the note on Laws IV and V. Editorial: the headings of the three dated notes, the verb in the abstract and section 6, the references, the links, the dateline, the footer and the page’s structured and citation records, which carry the new version and date. No registered law, prediction, kill condition or dependency changes; no result is added; no date of any claim moves.
Paper C v1.8 - 21 September 2026: three dated notes are added beside passages that already stood, and no sentence of the paper is changed. In section 9, a provenance note below the third-stream evidence box separates what a carrier establishes, the integrity of the signed bytes as sent and a carriage time asserted on the signing domain’s authority, from what assessment of the documents against stated criteria establishes, and carries the same limit into the paragraph on the three streams below it. In section 10, a reading note below the three points of precision states that a public record evidences the absence of an identified citation or connection rather than the absence of knowledge, that the external preprint’s subject is artificial agents exactly as the table above it already says, and that the 2026 formal results are relevant to the control-failure half of the 8 December 2024 content under their own stated assumptions rather than establishing it. In section 11, a scope note below the institutional conclusion carries the conditional at the head of that section and the necessity limit already printed above it into the conclusion, and reads the step from a diversity result about artificial agents to a claim about human institutional composition as a proposed bridge rather than a validated implication. All three are substantive qualifications of how existing passages are to be read, and none is editorial. No claim is withdrawn and no recorded date, result or status changes; the revision dateline and the version fields move with the version, as they do at every revision. Each note keeps the 19 September 2026 label of the frozen owner-decision text it reproduces, preserved unchanged as that file’s own historical label; the owner decision that settled the three notes is dated 20 September 2026, and this revision is published on 21 September 2026. The date of first publication, 3 July 2026, is unchanged.
v1.7: The provenance note at the end has been corrected for accuracy. No claim, date, result or status has changed.
Paper C v1.6 - 2 September 2026: in the cognitive-monoculture section the cross-domain mapping now credits the Foundational paper rather than Paper III, and the Foundational paper joins the reference list; the third evidence stream states its convergence under the register’s no-total policy, keeping the nineteen as the dated snapshot of 2 July 2026; Cage and Troxell-Whitman (2019) and DeepSeek-AI (2025) are cited in the running text as well as listed; Paper X appears under the title it now bears; the citation block carries this paper’s own DOI (10.17605/OSF.IO/RBHDP) in place of the programme umbrella; a structured-data link pointing to another author’s preprint is gone; the footer version matches the version shown on the first page.
Epistemic status. What this programme names Laws are conjectures. Laws I to III are under registered adversarial test. Laws IV and V were added on 24 September 2026; the predictions that carry them are registered, the laws themselves are not yet registered, and neither has been tested. Every quantity in this paper is operationally defined, and established-law standing is claimed nowhere. The registered programme exists to earn that standing, or lose it, by measurement, replication and survived refutation.
© 2026 Michael Darius Eastwood. Full human authorship and moral rights are asserted under the Copyright, Designs and Patents Act 1988 and consistently with United States Copyright Office guidance on works containing AI-generated material; any novel technical contribution described in this work was conceived by the human author. Full statement: michaeldariuseastwood.com/authorship.
Standing covenant. Prove this paper wrong, and I will publish the refutation myself. Falsification conditions are stated in this paper; the standing challenge: github.com/MichaelDariusEastwood/arc-scaling-challenge.
Michael Darius Eastwood conceived and directs this research programme and is the author of this work. Across the programme, he has used more than six AI systems in parallel, under his own instructions, to stress-test his arguments, identify possible errors, and assist in preparing draft text from his own outlines. He determines what is adopted, revised or rejected and takes responsibility for the published content. These systems are tools, not authors.