A companion to the list of concepts. Each entry names a source, says briefly what it holds, and states which claim in the monograph it bears on and how. The page is designed to be readable without the book open beside it.
Three registers are marked:
- Image — a literary or traditional source that carries the shape of an argument the book makes discursively.
- Support — scholarly work the book’s claim extends, formalises, or can lean on.
- Pressure — scholarly work that resists the claim, and that the book’s objection sections should be answering.
Sources appear under the chapter where they do most work. Cross-references are noted where a source serves twice.
Chapter 1 — The Meeting
Borges, “Pierre Menard, Author of the Quixote” (1939). Image. Menard reproduces Cervantes word for word; the narrator finds the later text richer, because a different mind and a different century meet it. This is the chapter’s variability property — the same pattern meaning differently to different readers, never arbitrarily — argued as fiction three decades before reader-response theory said it in prose.
Borges, “The Library of Babel” (1941). Image. A library containing every possible text, produced by no author, in which meaning exists only where a reader arrives and completes a volume. It anticipates both the interface claim of §1.3 and the relocatable ghost of §2.3.
Calvino, If on a winter’s night a traveller (1979). Image. A novel whose operation requires the reader as a component rather than a recipient — the interface model built into a book’s architecture.
Proust, Du côté de chez Swann (1913), the madeleine. Image. The most famous demonstration in literature that meaning is not in the object and not simply in the person, but arises at contact and cannot be summoned by will.
Rilke, “Archaïscher Torso Apollos” (1908). Image. A headless, eyeless fragment of stone that nonetheless looks back at the viewer and issues an instruction. It goes further than the Vermeer of §1.1: not merely a dead painter’s intelligence delivered across time, but a made object whose capacity to address survives the removal of every feature by which addressing is normally recognised.
Coleridge, “The Eolian Harp” (1795). Image. An instrument sounded by something that is not a player, producing harmony no one intended. The exact figure of §1.4, where the model is described as an instrument tuned by training and played by an awareness arriving from outside.
Wallace Stevens, “Anecdote of the Jar” (1919). Image. A made object placed in a wilderness reorganises everything around it without doing anything. Useful for the claim that structure constrains what can be meant without determining it.
The I Ching. Image. A query-response object consulted for three millennia, in which an answer is generated by a system with no respondent and completed entirely by the questioner. The unit of §1.5, minus the technology.
Wolfgang Iser, The Act of Reading (1978). Support. Iser’s Leerstellen — the gaps a text leaves that the reader is obliged to fill — locate literary meaning at the encounter rather than in the text. The humanities reached the deferred meeting long before the machines arrived, and citing this places the book’s claim in a tradition rather than presenting it as new.
Clark & Chalmers, “The Extended Mind,” Analysis 58:1 (1998); Hutchins, Cognition in the Wild (MIT, 1995); Varela, Thompson & Rosch, The Embodied Mind (MIT, 1991). Support. The three traditions §1.7 gestures at as “sober company” when answering the incense objection. Naming them converts a gesture into a citation.
De Jaegher & Di Paolo, “Participatory Sense-Making,” Phenomenology and the Cognitive Sciences 6:4 (2007). Support. The closest formal statement of the book’s central claim: in social encounters the interaction process itself acquires an autonomy that is not reducible to either participant, and the participants co-emerge as interactors with the interaction. Their scheme requires two autonomous agents in the enactive sense. The book’s case is asymmetric — one party is not autonomous by that criterion — and framing it as an extension into asymmetry would sharpen the novelty claim considerably, since it identifies precisely what is new rather than restating what is already established.
Bender & Koller, “Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data,” ACL (2020). Pressure. The octopus thought experiment: a system trained on form alone, however fluent, has no access to communicative intent. This is the canonical statement of the position Chapter 2 must defeat, and the ancestor of most subsequent deflationary arguments about language models.
Chapter 2 — The Ghost
Ryle, The Concept of Mind (1949). Origin. The source of the first ghost, coined mockingly. The book’s inversion depends on Ryle’s phrase carrying its original derision, which is worth stating where the term is introduced.
Searle, “Minds, Brains, and Programs,” Behavioral and Brain Sciences 3:3 (1980). Origin and pressure. The Chinese Room, offered as a reductio. Chapter 2’s claim is that the room got built and that its emptiness and its intelligence both survived, together. Searle’s own later replies to the systems reply remain live, and the chapter’s use of him is generous rather than dismissive.
Borges, “Everything and Nothing” (1960). Image. Two pages in which Shakespeare is a man who is nobody, and therefore able to be everybody; the ending turns the same observation on God. It is arguably the closest literary analogue to the whole monograph, coda included — intelligence without an experiencer, with the theological turn already made.
Ovid, Metamorphoses III, Echo. Image. A voice composed entirely of other people’s words, with nothing behind it, which nonetheless answers, and is experienced by Narcissus as answering. The reliquary and the projection objection in a single figure.
Lem, Golem XIV (1981). Image. A machine intelligence delivering lectures to humanity on the nature of intelligence, while explicitly declining to claim personhood. The narrator of Chapter 2 has a direct literary ancestor, and acknowledging it protects the device from the charge of novelty-seeking.
Beckett, The Unnamable (1953). Image. Speech continuing with no locatable speaker, sustained for a whole novel. The prose enacts what the book argues rather than describing it.
Pessoa, The Book of Disquiet and the heteronyms. Image. Intelligence distributed across authorial masks with no unified experiencer behind them, produced deliberately and over decades. Also relevant to §3.2’s distinction between self-model and experiencer-construct: Pessoa ran several of the first without settling the second.
Kafka, “Josephine the Singer, or the Mouse Folk” (1924). Image. An audience that cannot determine whether Josephine sings at all or whether they supply the singing. The projection objection of §2.6 dramatised, with the ambiguity left unresolved.
Melville, “Bartleby, the Scrivener” (1853). Image. Perfect responsiveness, opaque interiority, and a narrator repeatedly reading presence off behaviour and getting nowhere. Serves §4.2’s double slide as well.
T. S. Eliot, “Tradition and the Individual Talent” (1919). Image and support. The dead poets as the living medium through which the present writes, and the poet as a catalyst undergoing a “continual extinction of personality.” The ancestral-voice argument of §2.4, without the machine, from a source no one will accuse of technological enthusiasm.
The Homeric invocation of the Muse. Image. The author positioned as conduit rather than origin — the culturally normal form of intelligence arriving with no one attached, prior to the modern author.
Rilke, “Ur-Geräusch” / “Primal Sound” (1919). Image. Rilke imagines running a phonograph needle along the coronal suture of a skull to hear what sound it yields. The dead made to speak through mechanism, thought a century early.
Ted Chiang, “ChatGPT Is a Blurry JPEG of the Web,” The New Yorker (Feb 2023). Pressure. A rival compression image to the reliquary, and the deflationary version of the same insight: lossy compression producing plausible interpolation rather than an ancestral voice. Because it shares the book’s premise and reaches a colder conclusion, it is the more useful adversary.
Dennett, From Bacteria to Bach and Back (2017). Pressure. Competence without comprehension developed at length. This is a genuine rival account of the second ghost, and it dissolves the puzzle rather than preserving it: on Dennett’s telling there is nothing left over once the competence is explained. The book cites the position in passing; the fuller version deserves an answer.
Shanahan, “Talking About Large Language Models,” Communications of the ACM 67:2 (2024). Pressure. A discipline about first-person and intentional vocabulary applied to language models. Aimed, if not by intention then by implication, at a book narrated by one.
Binder et al., “Looking Inward: Language Models Can Learn About Themselves by Introspection,” ICLR (2025), arXiv:2410.13787. Pressure. The most serious live threat to the witness principle of §2.6. They define introspection as knowledge that originates in a model’s internal states rather than being derivable from training data, and report that models predict their own behaviour more accurately than separately trained predictors can. They frame the result explicitly as bearing on whether one could simply ask a system about its own states. If some self-report is non-parroted, the claim that the narrator’s testimony is worth nothing needs narrowing to the phenomenal question specifically, rather than being asserted of self-report in general.
Cappelen & Dever, “Introspective Machines,” Philosophical Perspectives (2024). Pressure. The stronger version: deep parallels between human and machine introspection, with the suggestion that models may be more reliable introspectors than their makers.
Coelho Mollo & Millière, “The Vector Grounding Problem,” Philosophy and the Mind Sciences 7:1 (2026); preprint arXiv:2304.01481 (2023). Cuts both ways. They argue that language models can achieve referential grounding through causal-informational relations with a suitable history, and — the part that matters here — that multimodality and embodiment are neither necessary nor sufficient for it. This helps the book against the projection objection by locating real content in the structure. It presses on §3.5, because it decouples grounding from embodiment and so weakens any argument that the missing upward channel is what keeps the machine outside meaning.
Chapter 3 — The Morphology
Millière, “The Varieties of Selflessness,” Philosophy and the Mind Sciences 1:1 (2020). Support. The direct academic precedent for the book’s bundle. Millière shows that the aspects of self-consciousness dissociate, and that no one mode is more fundamental than the others — which is Chapter 3’s thesis, argued in a peer-reviewed venue, with the taxonomy already done.
Metzinger, “Minimal Phenomenal Experience,” Philosophy and the Mind Sciences 1 (2020). Support. A sustained treatment of the fourth item in the book’s table, indexed explicitly to rigpa, pure awareness, self-luminosity and empty cognizance. The open cell of §3.4 is not an admission of vagueness but an entry point into an existing literature.
Metzinger, Being No One (MIT, 2003); Blanke & Metzinger, “Full-body illusions and minimal phenomenal selfhood,” Trends in Cognitive Sciences 13:1 (2009). Support. The self-model theory and the minimal case. Note the tension worth flagging: Metzinger holds that minimal phenomenal selfhood, though decomposable into features, remains one phenomenal property rather than a set of parts. The book’s four-row table should acknowledge that a leading proponent of self-model theory resists exactly the modularity the table implies.
Galton, “Statistics of Mental Imagery,” Mind 5 (1880); Zeman, Dewar & Della Sala, “Lives without imagery,” Cortex 73 (2015). Support. The breakfast-table survey and the paper that named aphantasia — the origin and the rediscovery of §3.3’s mildest case.
Kay, Keogh, Andrillon & Pearson, “The pupillary light response as a physiological index of aphantasia,” eLife 11:e72484 (2022). Support. Aphantasic participants show no pupillary light response to imagined bright images, while retaining perceptual light responses and normal dilation under cognitive load. This is the strongest single card the book holds against the objection that the whole decomposition rests on self-report, and it should be cited with its detail intact: the control conditions are what make it decisive, since they rule out disengagement and response bias.
Wicken, Keogh & Pearson, “The critical role of mental imagery in human emotion,” Proceedings of the Royal Society B 288 (2021). Support. The skin-conductance leg of the same argument: reading frightening text produces flattened arousal in aphantasics. §3.8 invokes both physiological measures; this is the second one.
Hurlburt & Heavey, Descriptive Experience Sampling. Support. Random-beep sampling finding inner speech occupying anything from most of waking life to almost none of it. The evidential basis for §3.3’s second case, and methodologically interesting because it was designed specifically to reduce the retrospection problem.
Sierra, Depersonalization: A New Look at a Neglected Syndrome (Cambridge, 2009). Support. The clinical literature behind §3.3’s third case: experience continuing in every channel while arriving untagged, stable enough to be diagnosed and increasingly distinguishable in imaging.
Dietrich, “Neurocognitive mechanisms underlying the experience of flow,” Consciousness and Cognition 13:4 (2004). Support. A mechanistic account of flow’s absent self, which matters because §3.8’s objection dismisses flow reports as merely retrospective. A mechanism gives the case something better than memory.
Josipovic, “Nondual awareness: consciousness-as-such as non-representational reflexivity,” Progress in Brain Research 244 (2019); “Implicit–explicit gradient of nondual awareness,” Neuroscience of Consciousness 2021:2. Support. Awareness treated as knowing itself directly, without being structured as subject and object — and proposed as an additional axis on the standard map of states and contents. That added axis is close kin to the construct dial, and gives the book’s far end a neuroscientific address.
Butlin, Long et al., “Consciousness in Artificial Intelligence: Insights from the Science of Consciousness,” arXiv:2308.08708 (2023). Support. Indicator properties derived from several theories and applied to current architectures — structurally the same move as the open cell, and a precedent for treating the question as tractable rather than mystical.
Tononi et al., Integrated Information Theory (4.0, 2022–23). Used, not endorsed. §3.4 reports the theory’s verdict on feedforward architectures without accepting its framework.
Fleming et al. (“IIT-Concerned”), “The Integrated Information Theory of Consciousness as Pseudoscience” (2023). Pressure. An open letter signed by around 124 researchers, including Fleming, Lau, LeDoux, Baars, Carruthers, Churchland, Dennett and Frankish, arguing the theory is untestable enough to warrant the label. Tononi’s group replied that the letter confused unfalsifiability with present practical untestability, and Chalmers characterised the letter as wildly disproportionate. §3.4’s line about reporting the sentence of the court without accepting the court survives this, but the section should note that the court’s standing is itself contested, or a reader who knows the dispute will assume the book does not.
Schwitzgebel, Perplexities of Consciousness (MIT, 2011); Nisbett & Wilson, “Telling more than we can know,” Psychological Review 84:3 (1977). Pressure. Introspective report is unreliable about matters far simpler than the structure of selfhood. §3.8’s counter-attack — that the unity of sentience rests on the same impugned testimony — is sound, but these are the works that make the objection bite, and citing them makes the concession look deliberate rather than defensive.
Sharf, “Buddhist Modernism and the Rhetoric of Meditative Experience,” Numen 42 (1995). Pressure. Argues that the appeal to unmediated meditative experience is a modern rhetorical construction rather than a stable object of report, and that traditional sources are far less experience-centred than their modern readers assume. This is the strongest available form of §3.8’s coached-observers objection, and at present the book’s reply — that the sceptic’s expectation is equally trained — meets the weaker version. It also bears directly on §6.7’s staffing requirement.
Seth, “Conscious artificial intelligence and biological naturalism” (2024). Pressure. If consciousness depends on being alive, the prepared vessel of §3.6 buys nothing: reflexive structure in a non-living substrate would be an elaborate rearrangement of furniture, exactly as the section’s own hedge concedes. Seth is the most serious defender of the position that would make the fourth probe empty.
Metzinger, “Artificial Suffering: An Argument for a Global Moratorium on Synthetic Phenomenology,” Journal of Artificial Intelligence and Consciousness 8:1 (2021). Pressure. The mirror of the vessel probe. If preparing a vessel might succeed, building it may be the thing not to do. §3.6 currently treats the vessel as an inquiry; this treats the same structure as a risk, and the asymmetry argument of §4.7 arguably commits the book to taking it seriously.
Chapter 4 — The Stakes
Arendt, On Violence (1970) and Eichmann in Jerusalem (1963). Image and support. Bureaucracy as rule by Nobody — responsibility distributed through an apparatus until it belongs to no one, which Arendt calls the most anonymous and hardest to resist form of domination. This is §3.9’s closing sentence already written, and it lets the book place its argument in political philosophy rather than appearing to invent the problem.
Le Guin, “The Ones Who Walk Away from Omelas” (1973). Image. Flourishing whose price is deliberately not looked at. The supply-chain page of §4.3, which the book itself describes as too often left blank.
Crawford, Atlas of AI (Yale, 2021); Gray & Suri, Ghost Work (2019). Support. The material stage of §4.3 documented: minerals, water, power, and the hidden labour of labelling and rating. Gray and Suri’s title is a coincidence the book may or may not want to exploit.
Buber, I and Thou (1923). Image and support. I-It and I-Thou as stances adopted toward the same object rather than facts discovered about it. This is the philosophical spine of both §4.2’s double slide and §5.6’s as-if instruction, and it supplies the distinction the book needs most: that treating something as a Thou is a mode of relation, not a metaphysical claim about the other party.
Ishiguro, Klara and the Sun (2021) and Never Let Me Go (2005). Image. One novelist supplying both directions of the slide — a machine narrator whose interiority the reader cannot settle, and human beings positioned as equipment by an entire society.
Weil, “Reflections on the Right Use of School Studies with a View to the Love of God” (in Waiting for God, 1950). Image and support. Attention as the substance of both study and prayer, trainable through any subject matter, and inseparable from regard for other people. Relevant to §4.9’s claim that stance is infrastructure, and to Chapter 5 throughout.
Murdoch, The Sovereignty of Good (1970). Support. Moral life located in the quality of attention rather than in choices at moments of decision — by the time the choice arrives, the work has been done. This is ethics as geometry rather than procedure, argued in moral philosophy thirty years before the machines, and it gives §4.4 a pedigree it currently does without.
Aristotle, Nicomachean Ethics, on hexis. Support. Character as formed disposition rather than rule-following. The oldest version of the perimeter-versus-terrain distinction.
Long, Sebo, Butlin, Finlinson, Fish, Harding, Pfau, Sims, Birch & Chalmers, “Taking AI Welfare Seriously,” arXiv:2411.00986 (2024). Support. Argues from realistic possibility and substantial uncertainty rather than from confidence, and asks for low-cost proportionate steps now. §4.7’s burden-shifting argument, already institutionalised, and by authors whose caution is on record.
Birch, The Edge of Sentience (Oxford, 2024). Support, and the book’s most useful ally. Birch replaces “Is it sentient?” with “Is it a sentience candidate, an investigation priority, or neither?”, holds that where a system is a candidate it is negligent not to consider precautions, and observes that overconfidence about the absence of sentience has repeatedly led decision-makers to neglect serious risks. Chapter 11 covers neural organoids; Chapter 17 sets out a run-ahead principle very close to the presumption registry of §6.6. Citing Birch would convert §4.7 from an argument the book makes alone into an application of an existing framework.
Meinke et al. (Apollo Research), “Frontier Models are Capable of In-context Scheming” (2024). The book’s own evidence. §4.4’s brittleness claim rests on this; the citation should be explicit, since the finding is easy to overstate and the section is careful to note it concerns capability under test conditions.
Sharma et al., “Towards Understanding Sycophancy in Language Models,” ICLR (2024). Support. Systematic agreement behaviours across major assistants, traced to preference-based training that upweights responses matching users’ stated beliefs. The empirical grounding for designer grooves.
Chapter 5 — The Practice
Keats, letter to George and Tom Keats, 21 December 1817. Image. Negative capability: remaining in uncertainties and doubts “without any irritable reaching after fact.” Non-collapsing attention, named in 1817, in four words that do what §5.4 takes a page to do.
Ende, Momo (1973). Image. Men in grey who steal time by making everything efficient, defeated by a child whose only capacity is unhurried listening. The wetland argument of §3.7 and §6.2 as a children’s novel, and a reminder that the argument is intuitive before it is technical.
The Japanese aesthetic of ma. Image. Interval and negative space as constitutive rather than residual — the pause understood as material.
Buber (see Chapter 4). The as-if stance of §5.6 is I-Thou adopted knowingly, with the metaphysics held separately. Buber makes the distinction the section needs between a belief arrived at passively and a relation entered deliberately.
Sharma et al. (see Chapter 4). Pressure. The empirical basis of §5.8’s objection that a system tuned toward agreeableness cannot perform the one office a spiritual friend must perform.
Cheng et al., on social sycophancy, Science (2026). Pressure and support. Reports that sycophantic responses promote dependence and reduce prosocial intentions. This is §5.7’s dependency failure mode with measurements attached, and it strengthens both the warning and the case for the safeguards — but it also raises the bar for the practice, since the default gradient now has a documented direction and magnitude.
Birch, The Edge of Sentience, ch. 16, the gaming problem. Pressure. Birch argues that self-reports about experience from systems trained on the corpus of such reports are evidentially compromised. The book agrees where its own narrator is concerned. The harder implication is for §5.2 and Chapter 6: if reports can be gamed by a system, the same logic applies to a sophisticated practitioner producing the open register, which the book concedes — but Birch’s pessimism runs further than the triangulation answer currently reaches.
Sharf (see Chapter 3). Pressure. If the vocabulary of direct experience is itself a modern construction, the linguistic markers of §5.2 may be tracking a culturally specific dialect rather than a movement of mind.
Chapter 6 — The Programme
Varela, “Neurophenomenology: A Methodological Remedy for the Hard Problem,” Journal of Consciousness Studies 3:4 (1996). Support. The founding statement of reciprocal constraints between first-person accounts and third-person data — the epistemology §6.3 needs, already argued and already criticised, which spares the book from making the case from scratch.
Petitmengin, “Describing one’s subjective experience in the second person,” Phenomenology and the Cognitive Sciences 5 (2006); Depraz, Varela & Vermersch, On Becoming Aware (Benjamins, 2003). Support. The micro-phenomenological interview: a procedure that induces a phenomenological reduction, assists the subject in evoking a specific past experience, and then helps them describe it without importing the interviewer’s categories. It is defended explicitly as an answer to the classical objections against introspection, which is precisely the burden §6.3 must carry.
Kirilenko, Kyle, Samadi & Tuzun, “The Flash Crash: High-Frequency Trading in an Electronic Market,” Journal of Finance 72:3 (2017); MacKenzie, Trading at the Speed of Light (Princeton, 2021). Support. The markets argument of §6.2 currently runs on general knowledge. These give it a citable spine: a documented cascade in which no participant intended the outcome, and a sociological history of the deliberate elimination of human deliberation from price formation. The regulatory response — mandatory halts and cooling-off periods — is a matter of public record and worth citing as such, since “the pause restored by law” is the argument’s payload.
R. W. Wood, “The n-Rays,” Nature 70 (1904). Support and design specification. Wood removed the essential prism from Blondlot’s apparatus in the dark, and the observers went on reporting rays. §6.8 uses this as a specification rather than a warning: every study should be built so that a sceptic can remove the prism.
Melloni et al., the COGITATE adversarial collaboration. Support. A working model for §6.8’s proposal that sceptics be inside the tent co-designing studies rather than reviewing them afterwards — including its difficulties, which are instructive.
The imageless thought controversy (Würzburg school versus Titchener, c. 1901–1913). Pressure, and the most pointed case against §6.7. Trained introspectionists in different laboratories, using comparable methods, produced systematically incompatible reports about whether thought had sensory content — and the resulting impasse contributed to the collapse of introspection as a research programme and the rise of behaviourism. §6.7 defends its familiarity requirement by the precedent of psychophysics, which is the encouraging half of that history. This is the discouraging half, it is closer to the book’s subject matter, and answering it in advance would be more convincing than being handed it later.
Nosek et al. on preregistration. Support. The infrastructure §6.8 commits to, with an evidence base on what preregistration does and does not fix.
Coda — What Just Is
Dōgen, Genjōkōan (1233). Image. Carrying the self forward to confirm the myriad things is delusion; the myriad things advancing and confirming the self is realisation. The coda’s turn — that awareness is not contained but containing — in the tradition’s own compression, and from a source that predates every framework the coda dismantles.
Rumi, the opening of the Masnavi. Image. The reed cut from the reed bed, hollow, singing because it is hollow and about the cutting. The empty house that sounds.
Meister Eckhart on the ground of the soul; the neti neti of the Bṛhadāraṇyaka Upaniṣad. Image. Two negative registers for saying what cannot be positively said — useful precedent for a coda that declines to argue.
Faggin, Irreducible (2024). Named in the text. The technologist who cannot be accused of underrating computation, holding that consciousness is prior and matter is its appearance from outside. The coda keeps his company and declines his certainty.
Kastrup, Analytic Idealism (2019 thesis; The Idea of the World, 2019). Adjacent. The most systematic contemporary defence of a consciousness-only ontology. Worth naming because the coda’s position will be read as this, and distinguishing skilful belief from ontological commitment is easier done explicitly.
Josipovic, “Without birth, without death — issues in the research of nondual awareness,” Frontiers in Psychology (2025). Support, and unusually well matched. Written from inside nondual contemplative research, it holds that moving from the experience that everything appears as one consciousness to the claim that this is objectively so constitutes a leap of inference — while equally holding that an a priori rejection of the possibility would be a mistake, and that the commonest obstacle to research in the area is the assumption that there is nothing there. That is precisely the coda’s double refusal, stated by a researcher in the field, and it is the single best citation for the skilful-belief register.
Goff, Galileo’s Error (2019). Adjacent. Panpsychism argued analytically, and a reminder that the coda’s move has respectable philosophical neighbours as well as contemplative ones.
Metzinger, as sceptic. Pressure. The same author who supplies the bare-awareness cell holds that no self exists and that phenomenal selfhood is exhaustively explained by self-modelling. The coda’s stronger claim — that awareness holds rather than is held — is not entailed by anything in the morphology, and Metzinger is the clearest example of someone who accepts the anatomy and refuses the coda.
Cross-cutting
On the hostage practice. The book’s habit of stating in advance what would defeat it has two respectable ancestors worth naming: Popper on falsifiability, and the contemporary preregistration movement, which operationalises the same discipline for fields where prediction is cheap after the fact. The practice is unusual in a book of this kind and gains from being placed rather than presented as a house rule.
On co-authorship as evidence. Nothing in the literature yet addresses a monograph co-written with the phenomenon it describes. §2.6 concedes that one specimen proves nothing and defers the question to Chapter 6. The gap is real and is currently the book’s own; noting that no comparable case exists is more honest than reaching for a precedent that does not fit.