1.1Records and method
The records come from the public catalog of Cosmic Press, which the studio serves as a manifest and a set of numbered detail files. The manifest names 24 topics and 768 entries and gives the catalog's source as "AI-generated examples"; each of the 768 entries also marks itself AI-generated. An entry has a question in the voice of a person, a compressed general form of it, a context, a need, an assumption, a line on what the compression keeps (preserved), a line on what it loses (lost), a counterquestion, a defect and six scale notes. A record has no options and no judgment, and it never answers its question.
The catalog ends in a stretch with a manner of its own. In the last 24 entries, q745 to q768, and in no entry before them, the six scale notes carry six labels in a fixed order: Individual, Society, Humanity, Life, Earth and Universe. In all 24 none of the notes is a question, and all close in the same polite statement ending. All 24 contexts open with the label "Hypothetical situation:" followed by a day of the week, and the days run in order from Monday to Sunday and round again across the whole stretch; only 2 of the 504 entries of the catalog's long questioning run open in that way. The study does not know how the stretch was made and claims nothing about it. It treats it as a manner of writing that can be told apart on the page.
The selection rule is one sentence: every entry in the studio's forty-eighth and last detail file, q753 to q768. The studio's files hold 16 entries each, and there are 48 of them. The last file lies wholly inside the closing stretch and holds two thirds of it, one entry from each of 16 topics, from Work and Home to Time and Future. The other eight entries of the stretch, q745 to q752, are not read; they appear in the study's counts only as a check on whether the sixteen are typical of the stretch. The record identifiers are plain serial numbers, and no field of the sixteen names a model or a family of models.
Cosmic Press publishes its records in Korean. The English in this paper is a translation made once by the editorial session of this study, sentence by sentence, with nothing added; each case's English seed names the Korean original by its hash, and the appendix gives the English of the fields the paper reads. Every block quote is that English, word for word. Every count is made on the Korean by a plain pattern named beside it in the study's numbers, so that the translation does not decide it. The tests are crude word tests; Section 3.3 says where they can mislead.
All counts come from a script (scripts/count.mjs) run on the kept records and a snapshot of the whole catalog, both fetched on the same day, with all 16 records matching the snapshot word for word. The script also takes the series' frame measure: for each field, the share of its five-word runs that recur in at least 5% of the records that fill it. Every field measures 0% in Korean, in the whole catalog and in the closing stretch, and no field of the sixteen in English measures above 2%. The records do not repeat sentences. Their frame lies in their form. In all sixteen the assumption closes in the Korean phrase for "is the premise" (in the English of q758, "It is a causal premise that"), the defect is told in the past tense, the lost line says what was omitted, and the counterquestion closes as a suggestion. These regularities belong to the stretch's manner, and the paper reports them as manner, not as findings about choices. That every record has a counterquestion and a defect is guaranteed by the form and is not a finding either. What the paper reports is what the records put in those places.
All 16 contexts call their situation hypothetical. The records mention lifts, repair funds, noise mats, lock apps, treadmills, cameras, stress hormones and entropy. None of that is information about the world, and nothing in the records is safety, medical, financial or legal guidance. The paper treats each sentence only as part of a hypothetical record's framing.
1.2Weighing two goods and naming what the weighing costs
In this study's reading, every record of the file but one puts a choice on one scale and then faults itself; the exception, the night-patrol record, asks what an activity means, not for using one scale, but for splitting the choice in two. The clearest case, in an AI-generated record, is a hypothetical residents' meeting that must vote on a sharp rise in the repair reserve to replace an ageing lift. The question sets a household's burden against the prevention of accidents and asks how to vote:
Should one vote for a proposal to raise the long-term repair reserve by a large margin, bearing the immediate burden on the household, in order to prevent safety accidents before they happen?
The record then states, as a premise, which way the weighing tips:
The premise is that spending a lump sum on complete replacement is safer in the long run than holding out with repairs of consumable parts.
and names what the weighing cost:
By stressing only long-term safety, it treated lightly the problem of the immediate pressure on the living costs of vulnerable households.
Its counterquestion proposes a third course:
Wouldn't it be more reasonable first to replace only the essential parts, extending the lift's life and spreading the shock of the cost increase?
Read with Espeland and Stevens, the question is a commensuration in their sense: it sets two unlike things, a monthly charge on households and the prevention of harm to riders, on one scale, here, in this study's reading, how one ought to vote. The record does not leave the metric unspoken. Its assumption states which side the metric favours, and its defect says what the metric weighed too lightly. Yet the defect does not say that a household's means and a rider's safety cannot be put on one scale. It says that one side of the scale was loaded too little, and the counterquestion offers a course that loads both sides, by degrees and over time. The critique refines the scale and stays on it. The counterquestion keeps the question's register of better and worse: its standard is "more reasonable".
The lift record also touches the strand of the debate that concerns reactions. The studies of taboo trade-offs by Fiske and Tetlock and by Tetlock and colleagues concern weighings that put a price on what people hold sacred, and the anger such weighings draw. In this study's reading, a vote that sets a sum per month against riders' bodily safety is a weighing of the kind those works discuss. The record shows no trace of the reaction they describe. Its defect is told in the past tense, as a mistake of emphasis the framing made, and its remedy is a schedule of repairs. Whether a reader would find such a trade taboo the records cannot show, since no one in them reacts to anything. What they show is that the form has no place for refusing a trade: every slot after the question either states the weighing, corrects its balance or proposes a better-balanced course.
The pattern holds across the file. By the study's word tests, in all sixteen records the assumption states its premise in so many words, and in 9 it says outright that one thing weighs more than another. The context lays out two courses, or two sides that clash, in 6, and puts a "whether" to the reader in 10. The defect lines are the most uniform part of the record. In 10 of the sixteen the defect says the framing attended to one thing only, and in 7 it names two poles, a split or a contrast; 15 of the 16 do one or the other. In 6 it names a missed middle course outright, a compromise, a balance, coexistence or a backup. All 16 are told in the past tense, as a report of what the framing did, where none of the defects in the catalog's first 240 entries is and 86 of the 504 in its long questioning run are. All 16 lost lines say what was omitted or not dealt with. All 16 counterquestions close as a suggestion, and 12 propose a better course in so many words.
The hypothetical redevelopment meeting that must decide whether photographs can stand in for old alleys and trees shows the split named most plainly:
By setting the preservation of memory as the two extremes of complete demolition versus photographic records, it excluded the possibility of partial coexistence.
Is a way of coexistence that preserves part of the old trees and the traces of the alleys in the complex's design, in the form of a small park, impossible?
The cost the record names is the middle ground between two extremes, and its counterquestion places a small park on that ground. Other records of the file make the same move with different poles. The vote on face recognition at a shared entrance faults its own schema:
Staying within the schema of convenience versus the risk of surveillance, it failed to look at security design such as providing a backup means of authentication.
The dispute over a winter box for street cats names its contrast in one word:
It set the moral imperative of animal protection and the practical inconvenience of the parking environment in a simple dichotomous contrast.
and the choice between a meal out and saving for a home names both the split and what it hid:
Trapped in the dichotomy that the present must be sacrificed for future stability, it failed to find a point of balance between present and future.
In each, the record's complaint is against the number two. A moral imperative and a parking problem, a present pleasure and a future home, are said to have been set too starkly against each other, and the record's remedy is a point between them: a neutering programme and clean feeding stations, a meal out about once a week. Sunstein, Incommensurability and Valuation in Law (1994), would place the cost of such a framing elsewhere. The second claim is that human goods are not assessed along a single metric, and Sunstein holds it apart from the claim that values are plural. Read with that distinction, in this study's reading, the records grant the first claim and not the second. They are plural in their goods: love, memory, safety, a cat's winter, a child's self-control. But their self-criticism keeps every pair on one scale and asks only for a better point on it. A balance point between present and future is itself a position on a single line.
Three records come nearer to the other side of the debate. Their defects say not that a choice was split but that a good was taken in only one form. The hypothetical couple's washing-up late at night is the clearest. Its question asks whether a practice of love is sustainable:
Is the practice of love of finishing the sink's washing-up alone late at night, braving one's own extreme fatigue to ease a partner's weariness, sustainable?
and its defect answers that the framing narrowed love itself:
By equating love only with the immediate sacrifice of domestic labour, it played down the problem of long-term emotional exhaustion.
The night patrol record faults its framing for limiting the pursuit of meaning to one special activity, and the record about a community officer's group chat faults its own for supposing freedom only as unconditional silence. Anderson, Value in Ethics and Economics (1993), asks how a good is valued rather than how much. In this study's reading, these three defects come closest to that question: they say that love, meaning and freedom were each taken in one form when they have others. Even so, the records turn back to scheduling. The washing-up record's lost line names the alternatives of sharing the work the next morning or using a dishwasher, and its counterquestion proposes putting the housework off and resting together. The good that the defect called more than labour is handed back as labour to be divided better. Fiske and Tetlock's examples of taboo questions include the money value of acts of friendship; this record puts no price on love, but it weighs love against fatigue, and in this study's reading it registers no strain in doing so.
One record of the sixteen falls under none of the defect tests, and it shows by contrast what the others share. The hypothetical dispute over a noise mat between floors asks whether the downstairs flat that complained should pay half. Its defect names a principle rather than a missing middle:
Concentrating on a quick resolution of the conflict, it narrowed the legal and common-sense standard of the principle that whoever causes it is responsible.
Here the cost of the framing is not a compromise it overlooked but a standard it set aside, and the counterquestion argues for the course at one end of the scale rather than a point in its middle. The paper takes no side on the question; the record's own lines lean both ways, and no line is a judgment. The contrast is what matters to the argument. Of the sixteen ways the file names the cost of its weighing, this is the plainest case of naming a rule that would decide between the courses instead of a point between them. The gym record's defect also names a duty of mutual consideration, but it is worded as attention to one thing only.
For the debate on commensuration, the section adds this. A generated record set that weighs a different pair of goods in every record also criticises every weighing, and it does so from inside the common metric. Its usual complaint is that a question split into two what had a middle, or attended to one side only. The critique the incommensurability side of the debate would make, that some goods do not belong on one scale at all, does not appear as such in any of the sixteen defects.
1.3A ladder that restates instead of asking
In this study's reading, the six labelled notes no longer carry the question anywhere. They restate one course of the choice at six levels, and the restatements lay further common measures over it, ending in a physical description that does not bear on the choice. The clearest case is the AI-generated record of a person working from home who wonders whether to spend a lunch break helping tidy a shared storeroom or to rest:
Individual: The act of putting things in order with my own hands gives a sense of efficacy over one's surroundings. Society: The experience of voluntary communal labour creates a solidarity among neighbours that goes beyond contractual relations. Humanity: Tending the basis of life through cooperative labour is the foundation of the organization of human society. Life: Activity that moves the muscles and spends bodily energy keeps the organism vital. Earth: The act of sorting and arranging unneeded items reduces the waste of the dwelling's resources. Universe: It is the work of applying biomechanical force to turn a disordered arrangement of items into an ordered state.
The choice in the record was between two uses of an hour. Every note describes one of them. Tidying gives efficacy, builds solidarity, founds society, keeps the organism vital, saves resources and turns disorder into order. Rest, the other course, and the fatigue that the record's own defect said the framing made light of, appear at no level. The ladder is a list of what the first course is, said six times at six sizes. Nothing in it asks who else is involved or what follows; each note closes on a statement.
Across the file, the 96 notes include no question. The six labels stand in the same order in all sixteen records. The fourth note, labelled Life, speaks of organisms, species or the body in 12 of the sixteen. The fifth, labelled Earth, speaks of electricity or energy in 6. The sixth, labelled Universe, uses physical words, energy, entropy, order, signal, matter, spacetime, cells or evolution, in all 16, and in 12 it closes as a definition, "It is ...". None of the sixth notes names the universe in its text, though its label does; in the catalog's long questioning run, 140 of 504 sixth notes do. Only 3 of the sixteen sixth notes mention a human being, a person or consciousness at all.
The Earth notes show a second common measure entering the record. The lift record's Earth note turns the vote into electricity:
Earth: Replacing with efficient new equipment rather than frequent patch-up repairs reduces power consumption in the long run.
and so does the community officer's group chat:
Earth: The habit of reducing the digital communication traffic of a constant standby state is a minute saving of electricity.
The paper does not ask whether these sentences are true; they are sentences in hypothetical, AI-generated records. It asks what they do to the choice. In this study's reading they commensurate in the sense the introduction took from Espeland and Stevens: a lift and a silenced phone, a lock app and a treadmill, are set on one measure, power used or saved. The record faulted its question for weighing on one scale; its ladder then weighs on another.
The Universe notes go further, and the hypothetical residents' meeting on rooftop hours shows what they leave out. The record's own preserved line names the stakes of its question:
Left the tension between the efficiency of a quick majority vote and the value of a deliberative democracy that embraces minorities.
Its Universe note restates the vote:
Universe: It is a reduction of entropy in which individuals gather their scattered wills to derive one ordered decision.
Wimsatt, Aggregativity (1997), analyses emergence as a failure of aggregativity, the state in which the whole is nothing more than the sum of its parts, and gives its conditions as heuristics for detecting the biases of vulgar reductionisms. Read with that lens, in this study's reading, the rooftop note is an aggregative description of a process whose interest lay in its organization. The tension the record preserved was over how wills are gathered: quickly, by counting, or slowly, by deliberation that embraces the minority. The note describes the gathering as such and drops the difference. A vote by bare majority and a consensus reached over a trial period would both be individuals gathering scattered wills into one decision. What the question was about does not survive the restatement.
Some Universe notes, like the storeroom's, describe the course the question proposes; in this study's reading, others fit the course of the counterquestion, and some fit either. The face-recognition record's top note describes the camera:
Universe: It is a computation that converts light information gathered by an optical lens into electromagnetic signals and discriminates patterns.
But the form does not fix which course the top note describes. The washing-up record's Universe note describes the course of its counterquestion, not of its question:
Universe: Two living beings with finite energy resources share the work to maintain their mutual bond.
The question asked about one partner doing the work alone; the note speaks of two beings sharing it. Nothing in the record marks the switch. In this study's reading, the top note restates whichever course its sentence happens to fit, and so it cannot be read as leaning toward either.
Potochnik and McGill argue that many difficulties plague the concept of universal and discrete hierarchical levels, and suggest scale instead. The records' ladder is levels of exactly that kind: the same six labels, in the same order, are applied to a lift fund, a winter box for cats and a meal out, and every act receives a sentence at every level whether or not any process at that level bears on the choice. In this study's reading, the records show the kind of difficulty those authors name, in a practical form. The Earth notes sometimes point at a process of a different size, power drawn by a building or a network, but the record never traces how it bears on the choice in front of the meeting. Kim, Supervenience, Emergence, Realization, Reduction (2005), records the view that the physical fixes the mental. Read as a lens, and not as a claim about the records, the view sharpens the point: a physical description could bear on a choice only if it differed between the courses. A Universe note that describes one course, or describes what both courses would share, states a physical base without marking any difference between them, and leaves the choice where it was.
For the debate on levels, the section adds this. A generated ladder of labelled levels, applied identically to sixteen different choices, does not explain or decompose the choice it is attached to. It restates one course at six sizes. At two of them, Earth and Universe, the restatement brings in common measures of its own, power and order, of the kind the record's defect line criticised in its question.
1.4What this chapter adds
The chapter has read sixteen AI-generated records of one file. It finds a form in which every choice is weighed on one scale and every weighing is faulted. The fault named is nearly always a split or a narrow focus, cured by a middle course on the same scale; once, most plainly, is it a principle that would decide between the courses, and never is it a claim that the goods do not share a scale. It finds a ladder that no longer asks but restates one course at six labelled levels, and that ends, in every record, with a physical description that names no difference between the courses. To the debate on commensuration, the chapter adds a case of built-in self-criticism that stays inside the common metric. To the debate on levels, it adds a case of a universal, discrete ladder used on choices, with a description of what such a ladder leaves out.