0.1The question and the claim
A number that stands for a person does two jobs at once. It reports something, a rate or a score or a share, and it licenses something: a cut, a rejection, a reassignment, a tax. Arguments about such numbers tend to fasten on the first job, asking whether the figure is accurate, or on the second, asking whether it is fair to act on it. Less often do they ask what has to be true of the relation between the number and the person before it may do either.
ASI 2030, one of the RISCON family's studios, writes scenarios in which machines move toward what it names artificial superintelligence by 2030, and it watches them with a device taken from accounting. After each judgment about a case, a human auditor gives an unqualified, qualified or adverse opinion, or declines to give one. The studio's story begins as a single sequence of quarters, from late 2026 to the end of 2028, which this paper calls the shared past; then it forks into two futures that run side by side. The scenarios are the studio's inventions. Nothing in them reports what has happened, and nothing in them is a forecast made by this paper.
The ten records studied here are every record of the shared past that the studio files under two of its eight axes, Measuring Human Worth and Work and Income. They are chosen because the two axes are where the studio's scenarios put numbers about people. An HR team asks how many positions an automated-response rate lets it cut. A working group designs an audit of the contribution scores that companies use in hiring. A task force asks what share of a productivity gain to put down to AI, so that it can set a robot-tax rate. The series to which this paper belongs asks, in every domain, what must be decided now and what would make the decision change. Here the decision is whether a number may decide something about a person, and the question is what the record set makes that permission depend on.
This paper argues that, in these scenarios, the auditor's permission depends on two tests met together, one about the number and one about the person. The first is a test of stability. The auditor asks for the figure again from a different three-month window, from the platforms that withheld their logs, from half of a score's inputs, or from another body using another method. The second is a test of reach. The auditor asks whether the raw data are open to the people concerned, whether there is somewhere for them to object, and whether objections have ever changed a case. Where only one test is met the opinion is qualified; where neither is, the auditor gives an adverse opinion or declines to give any. The one unqualified opinion in the ten falls on the one record where the number was reproduced across three quarters and the people it sorted were told in advance of the criteria and of a way to appeal.
A second finding concerns the record's own form. Each judgment in these scenarios rests on a short basis whose entries the studio marks as evidence, claim or assumption. In the records read here, the step that lets a figure stand for a person, that a resolution rate is complex-case ability, that a gain splits cleanly into labour, capital and AI, that one dispute is the same type as another, is not once marked as evidence. It is written down as a claim or an assumption. The form makes visible the act that the sociology of quantification calls commensuration, and it lets the auditor ask of that act what it asks of any unverified premise.
0.2The debate on numbers and people
The paper speaks to the sociology of quantification, a body of work on how numbers are made and what they do once they circulate. Each work is read here through the account of it that this paper could read, and where the paper goes beyond that account it says so.
The starting point is the appeal of numbers in public and commercial life. Theodore M. Porter, Trust in Numbers (1995), traces the appeal of quantification through the changing cultural meanings of objectivity over two centuries, and turns the usual explanation around: rather than numbers being prized in social and economic affairs because they succeeded in the study of nature, the attractions of quantification in business, government and social research explain something about its place in the sciences themselves. This paper reads the records as full of that appeal: in many of them a firm, a task force, a court or a platform reaches for a single figure to settle a matter.
Wendy Nelson Espeland and Mitchell L. Stevens, Commensuration as a Social Process (1998), give the core concept. Commensuration is the comparison of different entities according to a common metric. They treat it as a vehicle of rationalization and a mode of power, and they draw attention to the cognitive and political stakes of calling something incommensurable. A decade later, in A Sociology of Quantification (2008), the same authors treat the production and communication of numbers as a general social phenomenon, set out five sociological dimensions of it, and close with a call for an ethics of numbers. That call is the hinge of this paper. The records offer, in an invented setting, one shape that such an ethics might take when it is written into an audit procedure.
The second strand concerns what numbers do to those they measure. Wendy Nelson Espeland and Michael Sauder, Rankings and Reactivity: How Public Measures Recreate Social Worlds (2007), argue from the case of law-school rankings that people change their behaviour in reaction to being measured, and they name two mechanisms of this reactivity, self-fulfilling prophecy and commensuration. Marion Fourcade and Kieran Healy, Seeing like a market (2016), ask what markets see when they look at people sorted by data into categories of taste, riskiness or worth. They argue that individuals accrue a form of capital from their positions on digital scores and rankings; that the facticity of these scores makes them organizational devices that can stratify; and that they create classification situations, shared life chances, in which outcomes come to be experienced as deserved. Barbara Kiviat, The Moral Limits of Predictive Practices: The Case of Credit-Based Insurance Scores (2019), examines when people refuse a predictive number. Contention arises, she finds, when a prediction rests on distinctions that do not match wider understandings of good and bad behaviour, and when the theory of why it works holds people accountable for what is not their fault; policymakers then de-commensurate, pulling the market back toward their own notions of what is deserved.
The third strand places numbers at work. Katherine C. Kellogg, Melissa A. Valentine and Angèle Christin, Algorithms at Work: The New Contested Terrain of Control (2020), describe algorithmic control as operating through six mechanisms: directing workers by restricting and recommending, evaluating them by recording and rating, and disciplining them by replacing and rewarding. They argue that workers resist, individually and collectively, through tactics they call algoactivism. Four of the ten records are in the studio's labour domain, and this paper reads them as placing the scenario on that contested terrain.
Finally, Andrea Mennicken and Wendy Nelson Espeland, What's New with Numbers? Sociological Approaches to the Study of Quantification (2019), review the field across administration, democratic rule, economics and personal life. They caution against any unifying account, stress the essential ambiguity of quantification, and ask for work that tracks numbers across sites and studies how different quantification regimes interact. This paper reads the records as well suited to that request, because the studio threads them into lines in which one number's logic seems to pass from one setting to the next.
0.3What this paper adds and how it proceeds
The debate has said much about how numbers acquire power over people and about when people push back. It has said less about a procedure that stands between the two: a test, applied before a number is allowed to decide, that asks both whether the number is stable and whether the person it counts can reach it. The contribution of this paper is to describe such a test as the studio's auditor applies it, to show that it has two sides that must hold together, and to show that the record form itself separates the figure from the equivalence that lets the figure stand for a person. These are findings about how a synthetic record set frames a decision. They are not claims about how any real audit works.
The paper also adds a point of method to the request that quantification be tracked across sites. The studio's shared past threads its records into two lines, one for each axis, and each line ends where the story forks. Reading along the lines shows a number's logic moving from a firm's own ledger toward a single person who asks why, and shows the auditor's opinions changing as it moves.
Chapter 1 sets out the records, the selection rule and what is counted, and then reads the ten cases in three sections: the two lines from a firm's ledger to a person; the equivalences the records write down; and the two tests, holding twice and being seen. The conclusion returns to the debate and states the contributions, the limits and the further research the reading suggests. An appendix gives the ten records word for word.