1. The Poverty of Attention

In 1971 a scholar who studied organizations wrote a sentence that would be quoted for decades afterward: that once information becomes abundant, it manufactures a poverty of attention. In a world rich with information, he went on, what information truly consumes is the attention of whoever receives it, and so most of the real cost of information is paid not by the party producing it but by the party forced to take it in.

At the time this was meant only as a diagnosis. What deserves more attention is what he did immediately afterward: he began asking whether this newly scarce resource could be given a unit of measure, and whether, once measured, it could be efficiently allocated. From the moment attention was named as scarce, it became something waiting to be counted.

This starting point breaks from the pattern set by the nineteen essays before this one. In each of those, the thing the construct needed to render commensurable already sat in the world, plain and physical: a field of land, a hold full of cargo, a day's labor. All that was missing was a shared unit. Attention has no such prior existence. It occupies no space, cannot be moved, cannot be banked for tomorrow, and disappears in the very instant it is spent. To build a scale for a thing like that, one first has to manufacture its footprints.

Manufacturing footprints is the thread running through every section that follows, and it grows finer each time. First: was it seen at all. Then: was it clicked. Then: how long did the person linger, where did they scroll to, where did they arrive from, where did they go next, whom are they connected to, and what can all of this, added together, be made to predict. Each added variable sharpens the scale by one more degree, and every sharpening pulls in a fresh population of things that had never been countable before.

Twenty-three years later, in October of 1994, a website ran what is generally credited as the first banner advertisement on the web. It is remembered today chiefly for one statistic: its click-through rate is said to have reached forty-four percent.

People who worked in the industry at the time, looking back on it, did not dwell on the fact that advertising had finally arrived online. What they dwelled on was something else entirely: for the first time, an advertiser could know how many people had actually seen an advertisement, and how many had done something in response.

That distinction set the terms for the next thirty years. Newspapers and television also sold advertising, but the only figures they could ever report were circulation and ratings — estimated totals, guesses dressed up as numbers. A television commercial airs, and no one can say who watched it, who got up to refill a glass of water halfway through, who watched with total indifference. Half of what an advertiser paid was, in effect, payment for a number it could never actually verify.

This was the dividing line.

From this point forward, the audience stopped being a formless mass of people. It broke apart into discrete, countable events: an impression, a click, an interaction.

And once a thing can be counted, it can be settled; once it can be settled, it can be bid on; and once it can be bid on, someone will always find a way to make the number larger. This road has already been walked several times over the course of this series, except that this time, what waits at the far end of it is neither a plot of land nor a shipment of goods.

In the nineteen essays before this one, the remainder was always what spilled out of the ledger: the old debt no one could ever quite settle, the column measured with precision but never actually examined, the proof that could not be produced, the connections with no column to sit in, the person whose reading on systemic risk came back as zero.

The construct had tried nine different ways of handling it, and every one of the nine was defensive: loosening the requirement, taking a new name, measuring precisely and declining to ask further, converting it into a figure that could be folded into cost, drawing the boundary a notch tighter, giving it an orbit that circled the ledger without ever touching it, demanding a voucher it could not produce, reclassifying it, or simply keeping it off the books.

What happens across the next thirty years runs in the opposite direction.

The construct stops waiting for the remainder to spill over. It goes out and takes it.

To be precise, going out and taking it is not the same as seizing it by force. What actually happens is more fundamental: shape gets manufactured for things that never had any. A mind wandering today, three extra minutes spent somewhere yesterday, a sudden urge to buy something last month — these things had always been happening; they had simply never become events that could be written down. The construct's first move is to turn them into events.

2. The Warning They Filed Away

In 1998, two graduate students published a paper on a large-scale hypertextual web search engine. That paper marked the technical origin of the company that would grow out of it.

In the appendix of the paper's full version, buried in a section on advertising, the two of them expressed open doubt about search funded by advertisers. The sentence that matters reads: advertising-funded search engines will be inherently biased toward the advertisers and away from the needs of the consumers.

This is not a paradox that critics invented for them after the fact. It is a judgment the two of them wrote down themselves, at the very start of their venture.

The place this sentence occupies is the same kind of place occupied by the half-deleted Smith of Essay 9, or the clause in Essay 17 that almost never gets invoked. The construct left a warning against itself inside its own archive, and that warning survives intact, available to anyone willing to look. It was never destroyed. It was simply never followed.

Precisely because the warning survives, the turn that came later deserves closer scrutiny. It was not technological necessity working itself out. It was a choice, made under commercial pressure and within reach of a scalable business model, to walk the very road its founders had once flagged as dangerous.

This matters as an argument against fatalism. The whole subsequent machinery tends to get told as a story of technology growing on its own: give it a network and it will grow advertising; give it advertising and it will grow tracking; give it tracking and it will grow profiles. But what these two wrote down proves that someone, at the time, could see another road, and could see in advance what the chosen road would cost. The road was chosen. No one was simply swept along by it.

The shape of that road was worked out, during these very same years, by an entirely different company.

Between 1998 and 2001, a search company turned pay-per-click search advertising into an organized market. In 2001 it announced publicly that it was the creator of the market for pay-for-performance search. In the second quarter of that year alone, it brokered three hundred and twenty-three million paid, targeted referrals, with roughly forty-five thousand advertisers bidding against one another for position in the search results.

It charged nothing unless someone clicked.

The weight of that one design choice is easy to pass over, so it is worth spelling out.

Before this, advertising had always been priced by space and by time: so much for a full page, so much for thirty seconds in prime time, priced off a circulation figure, off a ratings estimate. What was being bought was position — a chance at exposure, the possibility of being seen by some estimated number of people over some stretch of time.

Pay-per-click changed what was actually being priced. The advertiser was no longer buying a position. It was buying an action.

On its face this looked like nothing more than a fairer way to charge: if nobody paid attention to your ad, you owed nothing. It sounds, if anything, like a concession to the advertiser. But the real consequence lies elsewhere. Once only the action counts, the entire machine grows in whatever direction produces more actions, not in whatever direction makes the content itself better. Essay 13 already made this point: once the construct settles on a unit of measure, all subsequent effort bends toward that unit.

And that action is performed by one specific person, at one specific moment.

For the first time, then, a verifiable line ran between the advertiser and that person. In the old, murky account, an advertiser paid for an estimated total; now every cent it spent corresponded to an actual person actually lifting a finger. Inside the advertising industry this was treated as a thorough and unambiguous advance. It was.

In the fourth quarter of 2000, the same search company launched self-serve text ads, at first charged by impression; by the first quarter of 2002 it had converted its entire advertising system to pay-per-click.

From this point, a machine had its basic unit. That unit was not an inch of page, not a second of airtime, not a thousand impressions.

It was a person lifting a finger to click.

This unit had a property no scale before it had ever possessed: it could only be completed by the very person being measured. An acre of land does not have to consent to being surveyed. A hold of cargo does not have to cooperate with being weighed. This unit does. If no one clicks, the number simply does not exist. The raw material for the entire machine could only be manufactured, by hand, by billions of people every day — people who, in manufacturing it, did not believe they were producing anything at all.

3. The Auction No One Attends

In 2004, that company prepared to go public, and the filing it submitted described its machine with unusual clarity.

Advertisers paid only when someone clicked, and every single search query triggered an automatic auction.

An ad's position depended on more than its bid. It also depended on measures like click-through rate, used as a stand-in for interest: an ad that was more relevant, more precisely targeted, would rise automatically simply by being clicked more often.

The filing offered a worked example of the pricing. Bidder A bid one dollar but paid only sixty-one cents; bidder B bid sixty cents but paid fifty-one; bidder C bid fifty cents and was charged the minimum, five cents.

This example is worth pausing over, because it reveals something quite refined: the highest bidder does not necessarily pay its own bid. What it pays depends on how much the bidder ranked below it offered, and on how likely the platform judges its own ad to be clicked.

Between what a person bids and what that person actually pays lies a layer of prediction.

That layer of prediction is the single most important part of the whole machine. It means what an advertiser pays depends not only on what it is willing to offer, but on how likely the platform thinks its ad is to be clicked — which is to say the platform is simultaneously running the auction and appraising the goods, and the number it appraises determines directly how much the platform itself collects. No earlier market had ever occupied this position. An auction house does not set the price of the lot it sells. An exchange does not wager on which stock is about to rise.

This is precisely where this machine parts ways with every market that came before it.

In earlier markets, price took shape between two people. A buyer names a figure, a seller counters, or an auctioneer calls a number until no one raises it further. Price is the outcome of a negotiation or a bidding contest, and the parties to it are the buyer and the seller.

Here, a third party takes part in setting the price — and that third party is not present, and does not know it took part at all.

It is the person doing the searching.

His search terms entered the auction. His past clicking habits entered the auction. His estimated interests entered the auction. He decided who would rank first, and he decided how much the advertiser ranked first would owe.

And he himself never knew the auction had taken place. The entire sequence, from the moment he pressed enter to the moment the page appeared, took under a second.

That fraction of a second is not a technical footnote. It is the condition that makes the whole arrangement possible at all. Any step requiring a human reaction would slow the system down past the point where billions of repetitions could run; and anything too fast for a person to notice is, by the same stroke, too fast to be questioned. The construct once relied on concealment to keep questions from being asked — Essay 17 already described concealment's function as a way of saving effort. Here concealment is not even necessary, because the whole transaction is finished before anyone could notice it happening.

In the first quarter of 2003, the same company extended this logic beyond the search page itself. Websites that joined its network could carry these ads on their own pages, with advertisers still paying strictly by the click.

It was no longer only the moment of searching. Many of the moments spent simply browsing were now wired into the same settlement system.

Web content, ad space, an advertiser's spending, and a viewer's actions had all been tied together into one apparatus.

In 2007 a federal agency approved that company's acquisition of an advertising-technology firm for three point one billion dollars, stating explicitly that it would not block the deal on privacy grounds, since privacy fell outside its antitrust authority — while at the same time acknowledging that the entire online advertising market had privacy problems of its own.

Looking back, the significance of that acquisition is this: the auction logic built for search advertising, the delivery mechanics of display advertising, and the infrastructure of tracking and third-party ad-serving were stitched together at a far deeper level than before.

Once stitched together, a single person's path across the internet stopped being a scatter of unrelated fragments. What he looked at on one site began shaping what he was shown on another; the words he searched yesterday followed him into today. Essay 18 described a machine that eventually began keeping other ledgers besides its own; here the direction runs differently — what is being recorded is still the outside world, except that world has now been strung into a single continuous line.

4. Counted Without a Click

Turning behavior into a trajectory requires one specific capability: the ability to recognize that this is the same person as before.

Around 1999 and 2000, an identifying marker stored inside the browser, working together with advertising networks, advanced what a person had looked at on one site into a record of what he had done across a great many sites.

In 2000, that same federal agency submitted a report on online profiling to Congress.

The report noted that the information gathered by ad networks was usually linked to the network's own identifying number rather than tied directly to any real name.

Then it wrote down a sentence that mattered more than the rest: this kind of information was collected, in its own words, even if the consumer never clicks on any advertisement.

That sentence describes a situation that had never existed before.

In the previous nineteen essays, for a person to enter the ledger at all, he first had to do something. He had to borrow money, sign a contract, pay rent, walk into a factory, buy an item, or at the very least be entered once onto some roster somewhere. The ledger recorded transactions that had actually taken place.

Here, no transaction is required. No click is required. No consent is required. No awareness is required.

Merely passing by is enough.

Those three words mark a boundary. In the nineteen essays before this one, a person and the construct always shared at least one point of contact: he signed something, he paid something, he was entered on a list, he drew a wage, he stood before a judgment once. Contact leaves behind a point in time — something that can be pointed to, argued over, revisited after the fact. Passing by leaves no such point behind.

This also reveals something about the remainder's nature. In earlier essays, the remainder usually needed some specific moment to show itself: the day of settlement, the day of judgment, the weeks when resources ran short. Here the recording never stops, and no such moment exists. A thing recorded continuously, with no day ever set aside for reconciling the account, is a thing that never gets the chance to be laid out on the table for anyone to examine.

A few figures from that report are worth setting down exactly as given. In the sample surveyed at the time, fifty-seven percent of the busiest websites allowed third parties to place identifying markers, and of those, only twenty-two percent mentioned third-party markers or data collection anywhere in their privacy policy. Among the hundred largest websites, seventy-eight percent allowed third-party markers; only fifty-one percent disclosed it.

One more figure says even more about the situation: at the time, only forty percent of computer users had ever heard of this kind of identifying marker at all.

The report also recorded a worry current at the time: that this sort of profiling might be used to set the price and terms of goods and services, even life insurance. Commentators gave the practice a name, borrowed from an older banking habit of drawing lines on a map to exclude certain neighborhoods from lending — they called it drawing lines online.

This distinction is worth making precisely, because it moves the whole matter from the mild territory of getting to know you better to somewhere else entirely.

Being profiled does not only mean seeing advertisements that fit you more closely. It can mean the very same item is quoted to you at a different price. It can mean some insurance policy carries different terms for you. It can mean certain opportunities simply never appear in front of you at all — and you would never know they had existed.

Essay 19 observed that the scale built for rescue could illuminate institutions but not individuals. Here the direction runs the other way: this scale resolves down to a level so fine that even the person being measured does not notice it happening.

What the two cases share is this: in both, the party being illuminated is not there.

Being absent is one more form of a question the previous essays have kept circling back to. In Essay 13, the people who could not produce documentation were not present at their own judgment. In Essay 16, the people left outside the register were not present when the register was compiled. In Essay 19, the people who lost their homes were not present when it was decided who would be rescued. This time it goes further still: he is not only absent from the room — he does not even know he has been counted at all.

5. A Friend's Endorsement, For Sale

In 2012, another company preparing to go public disclosed, in its own filing, a different sort of thing entirely.

When a person clicks like, or leaves a comment on a post, that action gets pushed out to his friends through the news feed — and a brand, the filing explained, could purchase a product that would significantly increase the reach, the frequency, and the prominence of that social action.

Translated into plain speech: the fact that you think something is worthwhile had just become purchasable inventory.

A word of fairness belongs here. A person telling a friend that something works well is one of the oldest transactions in any market, and it was happening long before any of this — the economy of remembered favors in Essay 1 ran on exactly this fuel. The platform did not invent word of mouth. What it did was attach a price list to it.

This step reaches further than pay-per-click ever did.

Pay-per-click bought your action. This bought the weight your action carried with someone else.

And that weight was valuable precisely because it was not advertising.

An advertisement on a page puts people on guard: they know the position was bought, they know whoever is speaking got paid for it. A like from a friend carries no such warning. It appears in the same feed, looks exactly like everything around it, and has already been paid for. A person doubts an advertisement; that same person does not doubt the fact that a friend clicked like. Personal trust was not routed around here. It was taken as raw material.

This step succeeded because of one small technical difference: a social platform knows who is friends with whom. The advertising machines that came before it knew what a person had looked at; this machine also knows who he knows, who will see what he does, whose word he is more likely to take. A relationship — a thing that had only ever existed inside the memories of the people involved, with no roster, no identifying number, no measure of its own strength — acquired, the moment it was recorded, all three.

Earlier essays have said, more than once, that anonymous pricing and personal, face-to-face trust have never simply replaced one another.

In Essay 6, paper money needed people willing to vouch for it before it could stand on its own. In Essay 17, the common scale ultimately depended on a group of people willing to keep honoring the account. In Essay 18, beneath every standard contract and every rating, some irreducible requirement to trust the other party remained buried at each joint.

In each of those cases, the personal was propping the anonymous machine up from underneath.

This time is different.

This time the anonymous machine reaches out on its own and takes hold of the personal, converting it into a distribution capacity with a price attached.

Trust did not prop up the machine. Trust became the machine's feedstock.

The difference between the two can be put this way: whatever props a machine up is remembered only once the machine breaks down; whatever serves as feedstock gets used up every single day the machine runs normally. The first reveals itself in a crisis. The second thins out, day by day, in ordinary time, without there ever being a single day when anyone notices it has grown thinner.

And feedstock gets consumed. If a person discovers that a friend's recommendation was mixed in with paid distribution, the next time he sees that friend click like he will think about it one degree more than before. That extra degree of thought will not show up immediately in any number anywhere; it accumulates slowly, and once it has accumulated far enough, the thing being treated as raw material is simply worth less than it used to be. Here the construct runs into an object it finds very hard to manage: the harder the vein is mined, the thinner it runs.

6. The Price of a Possibility

This is the point where the single most important part of the whole story comes into view.

On the display-advertising side, real-time bidding emerged around 2009. One industry retrospective notes that on a particular ad exchange, the share of inventory sold through real-time bidding jumped from eight percent in January of 2010 to sixty-eight percent by May of 2011.

In a little over a year, it went from under one-tenth to nearly seven-tenths.

This curve's shape has appeared before in this series. Essay 18's notional principal, Essay 14's reserves and liabilities — whenever they climbed, they climbed this same way: no single year produced a dramatic break; each year simply did a little more than the year before it; and only in hindsight had the world already changed into something else.

By this stage, an advertisement was no longer something sold once, in a fixed corner of a page. Every single impression could now trigger an automatic, instantaneous, cross-platform auction.

And inside the formula these auctions used to rank their results, one component deserves to be pulled out and examined on its own.

According to that social platform's own explanation to advertisers, the auction ranked bids by total value, built out of three components: the bid itself, an estimated action rate, and ad quality.

An estimated action rate.

Those three words deserve to be read slowly.

What they mean is this: the platform estimates, for this one specific person, right now, how likely he is, having seen this particular advertisement, to take the action in question. Will he click. Will he install the app. Will he fill out the form. Will he buy.

And that estimate feeds directly into the price.

To be clear, this estimate does not need to be accurate. It only needs to be slightly more accurate than a competitor's estimate, and only on average, across enormous numbers. Looking at any single person, the platform can be entirely wrong; averaged across hundreds of millions of impressions, that small edge is money. This is one more way this machine differs from every scale that came before it: it never sets out to measure any one specific person correctly. It only needs to measure a very large population slightly more correctly than everyone else does.

In the nineteen essays before this one, whatever the construct priced was always something that already existed, or had already happened. A piece of land has an area. A day of labor has a duration. A debt has a principal. A contract has a notional value. A house has a sale price. Even the most abstract instruments — options, swaps, synthetic products — price some already-existing underlying asset's future performance.

Here, what gets priced is something that has not happened yet, and quite possibly never will.

You did not click. You may never click. But whether you will click has already, tens of milliseconds earlier, been reduced to a number, and that number has already decided which advertiser won this particular impression, and how much that advertiser owes for it.

A person's possibility has been settled — and that possibility neither needs to come true, nor does.

It is worth glancing back at this point. Essay 11 explained that labor power is a peculiar commodity because what gets sold is not work already finished but a capacity to work over some future stretch of time — already one step ahead of what is being traded, since the thing traded had not happened yet. Here the step goes further still. Labor power still eventually has to be used: an employer who pays still has to call the worker in. Here, that final step is not even required. Your possibility gets sold, and you are free to do nothing at all.

The common measures of the industrial age were hours worked, units produced, freight charged, prices, interest rates — all of them pointing at something already completed.

This era's newly added common measures are click-through rate, cost per click, relevance score, quality ranking, estimated action rate, average price per ad, a person's lifetime value.

All of them point at something that has not happened yet, and at how easily a person can be anticipated.

Strictly speaking, this is not a seizure of anyone's inner life. Nobody has taken hold of your thoughts.

What gets possessed and traded is the outward trace an inner life leaves behind, and the predictability that can be calculated from those traces.

This distinction has to be held onto, not overstated. Nobody reads your thoughts. Nobody takes your will away from you. Predictability is not an inner life; it is the regularity an inner life leaves on the outside. A person remains free, at any moment, to break the pattern — and that fact turns out to matter a great deal for what comes next.

And here comes the single most important turn in this entire history.

Once predictability itself becomes a commodity, increasing predictability becomes a business.

A scholar who studies this mechanism carried the thought all the way to its conclusion: the surest way to predict behavior, she wrote, is to intervene at its source and shape it.

Essay 18 observed that a model does not stand beside the market describing it; it enters the market and changes it, so that the thing being measured no longer holds still — it moves according to whatever reading the scale gives it.

Here it goes one level further.

The scale is no longer merely waiting for the thing it measures to move on its own. It has begun reaching in and reshaping the thing it measures into something easier to measure.

A person who is easier to predict reads higher on this particular ledger. And so every practice that makes people more predictable now carries an economic rationale of its own: a fixed prompt, a recommendation that surfaces exactly when you are about to leave, a day sliced into many identically sized segments.

Not because anyone involved is malicious. Because on this ledger, being predictable simply is the value.

This is the form the construct finds hardest of all to handle. The nine methods described earlier were all ways of handling a remainder after it had already appeared, and here a tenth emerges — the only one that pushes outward rather than inward. It no longer processes whatever has already overflowed. It goes and remakes the very source the overflow comes from, so that it overflows a little less. A thing that was never going to run past the edge of the scale in the first place is a thing the construct never has to trouble itself with cleaning up.

7. Too New, or Not New Enough

Everything laid out above traces back to a thesis proposed by one particular scholar — a thesis that has proven highly controversial, and both sides of the controversy deserve to be laid out.

Her central account runs as follows. That search company discovered, in the internet's early years, that the traces left behind by human behavior were not merely a byproduct of improving its service — they could be extracted, refined, and sold. She gave the extracted material a name: behavioral surplus. Behavioral surplus, in her account, gets manufactured into prediction products and sold into a market that trades in people's future behavior; and in a later phase, platforms stop merely predicting behavior and start, through an entire architecture of design, tuning and inducing it.

She also supplied a figure: that company's revenue grew from eighty-six million dollars in 2001 to three point two billion dollars in 2004, a rise of more than thirty-five times over. It was precisely this rate of growth, she argues, that exposed an invention that had until then stayed hidden.

Her supporters value, above all, her capacity for naming. Free services, personalization, the small red notification dot, the recommendation feed, the smart-home device, the black-box auction — scattered experiences like these get gathered into a single continuous historical thread. One reviewer called the resulting book both a descriptive tour de force and a theoretical triumph, agreeing that it had genuinely captured a new business model, a new mode of extraction, and a new mode of control.

The criticism, meanwhile, comes in four distinct varieties, and they do not resemble one another.

Laying out all four exactly as they stand matters, because they are not the same objection wearing different clothes. Some say she made the new thing sound newer than it is. Some say she accounted for too little of the old. Some say her language is pitched too strongly. Some say all this was said before, and said closer to the root of the matter. Every one of the four can hold up while still granting every fact laid out in the six sections above.

The first kind says she made it sound too new. The same reviewer quoted above points out that earlier research traditions were never adequately reckoned with, and that describing all this as the occupation of a lawless frontier will not do, because its relationship to law is far more complicated, and far more generative, than that image allows. The critics do not deny that something new exists. They simply think it stands atop a much longer history of law, infrastructure, and governance.

The second kind says: too much surveillance, too little capitalism. One critic put it in a single sharp line: surveillance capitalism, unsurprisingly, turns out to be more surveillance than capitalism. His point is not to let the platforms off the hook. It is that fixating only on behavioral surplus risks treating the deeper logic of capital accumulation as background noise — and that is exactly where long-run profitability, competitive pressure, the externalizing of costs, state support, and global inequality all actually live.

The third kind comes from analysts working in law and policy. One long review concedes that she has identified a genuine danger, but argues the case is repeatedly buried under overstatement; these critics prefer the term surveillance economy to surveillance capitalism, and explicitly want to challenge the claim of a radical break from what came before. The disagreement here is not over whether a problem exists, but over whether these problems earn the right to be named a new stage of capitalism.

The fourth kind comes out of an older tradition of labor scholarship. Some researchers describe social media as a site of what they call playbor, stressing how the line between leisure time and labor time has worn away; others describe platformization as the practice of parceling out a vast quantity of severely underpaid or entirely unpaid tasks to users, coordinated through standardization, fragmentation, and the breaking of work into micro-tasks. For this camp, the point is not some wholly new capitalist logic but that digital capitalism has reorganized labor, consumption, play, and sociality into a single mechanism of value accumulation.

A further strand emphasizes global inequality, framing the data relationship as a form of data colonialism — not a claim that today simply repeats old colonialism, but that human life itself is now continuously appropriated as raw material for accumulation.

On this point, a scholar who also studies digital labor offered a warning worth recording exactly as he gave it. Reaching straight for analogies between today's data arrangements and slavery, imperialism, or colonialism, he wrote, risks missing what is historically distinctive about this new arrangement.

That warning matters a great deal for how this piece of history ought to be written.

Essay 8 described an episode in which an entire person was pressed flat into a price plane — one that involved legal ownership and sale, involved violence, involved a person's name being entered into an insurance company's records. None of that is present here.

The distinction is not only about the presence or absence of violence. The more crucial distinction concerns the object of the transaction. In that earlier episode, what got pressed into the price plane was a person's entire body, and the law said so explicitly: it could be mortgaged, it could be inherited, it could be entered into an insurance policy. Here, no document anywhere states who belongs to whom, and none could, because the thing being traded is not an entity that could belong to anyone in the first place.

What happens here is something else entirely: no one owns you under the law, while your attention and your behavior are strenuously kept in a state of being priceable, calculable, and sellable at every moment.

Collapsing the two into a single story does an injustice to history, and lets slip past whatever is actually distinctive about the one happening now.

This discipline has already been applied once in this series. Essay 16 read three famines side by side, and it too required keeping three different technologies of exclusion carefully distinct rather than folding them into one undifferentiated story. Distinguishing them was never meant to lessen their weight — quite the opposite: only once a thing's exact shape has been stated precisely does it become possible to say what it actually did.

8. The Thing That Got Away

The final set of debates concerns what ought to be done about all this, and it does not resolve either.

One camp argues that data should be treated as labor. Several researchers proposed, in 2018, that data should be regarded, at least to some degree, as labor — rather than as capital that firms simply generate by observing what people naturally do. Their argument: platforms function as the dominant buyer of data, approaching something close to monopsony power; treating data as free underestimates what users actually contribute, deepens distributional inequality, and lets dominant platforms go on profiting indefinitely from a price of zero.

This camp's strength lies in how thoroughly it dismantles the word free.

The word free has already surfaced earlier in this series. In Essay 12, housework registered as zero on the books partly because no one had ever been required to put a price on it. The situation here is similar but not identical: there, no one was doing the counting at all; here, the counting is extraordinarily precise, except the proceeds of that counting land on the other side of the table. The people using these platforms are not contributing nothing in exchange for a free service — a platform's value genuinely depends, in part, on the signals, preferences, text, images, clicks, tags, and feedback its users go on producing.

The objections to this camp come in several forms.

One is conceptual. A scholar asked directly, in 2024: is data contribution really labor? His answer was that, measured against everyday intuitions about what work is, data contribution usually fails to meet people's instinctive standards for significance, effort, and interpersonal recognition. His point was not that nothing should be regulated, only that calling every act of data contribution labor may be stretching the concept past where it can hold.

A second is political-economic. Some researchers argue that the more accurate phrase is not data as labor but data as capital; what should matter, on this view, is not issuing users back pay but how data feeds into an accumulation process with no natural endpoint.

A third runs deeper still. Some researchers resist putting the emphasis on compensation at all, arguing that framing the problem as one of paying wages for data risks conceding, too quickly, the very common measure the platform itself has already set — as though everything could be resolved through compensation. What they want to point out instead is that the platform is not simply underpaying you by some fixed amount; it is remaking the boundaries of the lived world itself, turning more and more things that should never have been subject to continuous extraction into inputs available for harvesting at any moment.

Compressed to two sentences, the two camps read like this.

One asks: since the people using the platform are the ones creating the value, why shouldn't they be paid for it.

The other asks: why accept, in the first place, the premise that a life can be sliced into settleable inputs at all.

The first camp wants to pull these people back into contract and distribution. The second suspects the language of contract has already arrived one step too late, because the boundary in question had already been redrawn before contract ever got there.

Both sides, in fact, agree on the same underlying fact, and differ only on which direction to take it from there. What they agree on is this: something belonging to a person is being continuously taken, and the taking of it has never gone through anything resembling real negotiation. Where they part is on whether the language of negotiation is still the right tool for getting it back.

No verdict will be offered here.

But whichever side one takes, one thing is clear from the record itself: this scale has never once measured everything there was to measure.

The 2000 report already noted that only forty percent of people had heard of that kind of identifying marker at the time, and that the profile itself remains invisible to the person it describes. Identifying markers can be deleted, browser settings can be changed, and the simple fact of not knowing is itself enough to destabilize a profile.

What has happened since is more direct still. Research found that after one region introduced privacy rules, the effectiveness of display advertising at shifting purchase intent fell by roughly sixty-five percent.

That figure deserves to be read two ways at once.

On one side: targeted advertising's value is not an invention of the imagination. It is real, and it is considerable.

On the other side: the machine that sets internal prices for all of this depends heavily on its capacity to track people, and that capacity is not irreversible. It can be cut away by law, cut away by browsers, cut away by operating systems.

This point is worth holding onto, because it appears only rarely in the nineteen essays before this one. The construct's measuring range has, all along, expanded outward: from clay tablets to coins, from coins to paper, from paper to equations, from equations to contracts worth hundreds of trillions in notional value. This time, for once, the range was deliberately pushed back a notch — and what pushed it back was not a crisis. It was rules, and it was devices. For the first time, the construct's boundary was pulled inward from the outside, and the pulling actually worked.

In its 2024 annual report, that same social platform still had to concede to its investors that regulatory conditions, along with changes to third-party mobile operating systems and browsers, had limited its ability to gather signals, target advertising, and measure results, and had hurt its advertising revenue as a consequence.

A company that had made predictability its core business, writing that sentence into its own official filing, amounts to an admission: the range of that scale was never set by the world it was measuring. It was set by how far it was permitted to reach.

Now it is possible to return to that widely repeated line: you are not the customer, you are the product being sold.

That line is crude, and it is not quite right.

A platform rarely puts a single price tag on an entire person. What it produces instead is a string of far more fragmented, constantly shifting shadow prices: what your next click is worth, how much three extra seconds of your attention raises an ad's relevance score, how much leverage one like from your friend hands a brand, how much more an advertiser will pay because you have been judged high-intent, whether your profile gets you sorted into a high-risk category in some market and charged accordingly.

What gets turned into an object in this market is not the person in any legal sense.

It is that person's predictable likeness.

This might sound like a softening of the claim. It is not. Something being traded as a stand-in for you, and you yourself being traded, are entirely different matters under the law; but in the texture of daily life, the distance between the two is not nearly so wide. You apply for insurance, and what the other side examines is that likeness. You open a page, and the price you are shown is calculated off that likeness. You go looking for a job, and that likeness is what gets read first. You yourself are always there in the room, and the you that is there takes no part in any of these calculations.

The part of a person hardest to predict is exactly the part most essentially their own: changing your mind, ignoring the prompt, turning the phone face down, not wanting to buy today, suddenly wanting to buy tomorrow, unable to say why.

On this particular ledger, none of that counts as value. It counts as noise.

In the nineteen essays before this one, the remainder was whatever the ledger let slip past its edges. Here the remainder is whatever the ledger classifies as noise: it was not missed. It was measured, and then it was subtracted, because it lowered the accuracy of the prediction.

And the very fact that it lowers the accuracy of the prediction is exactly the proof that it is still there.

The ledger has not yet balanced. It is still being kept.