Category Capture

The rankings change substantially between participation and recognition.

Natalie’s Magnificent Obsession ranked fifteenth in displayed comments but first in captured Reader Pick recognition, accumulating 31,748 recommendations. Yes, Woke 1.0 Was Crazy. Now Where’s the Accountability? moved from eighteenth in participation to second in recognition. The $20 burrito discussion moved from twenty-first to third.

The important question is not which discussion “won.” It is what readers rewarded once the discussion formed.

The second-ranked recognition discussion provides an unusually clean example of the behavior MRI-002 is trying to observe.

“Woke” appears in the headline as a single category. The recognition layer shows how little agreement exists about what the category actually contains.

The leading Reader Pick, with 3,591 recommendations, attacks the author’s own record on climate change and Iraq. The next, with 3,444, argues that cancel culture is at least as characteristic of the political right. Another predicts a future article about the excesses of the Trump era. Then, with 2,220 recommendations:

“I am at a public University in North Carolina. Now we have to deal with right wing wokeness. It is worse.”

Farther down the same recognition layer, one reader defines woke as compassion, fairness, equality and civil rights; another describes it as awareness of discrimination.

The word remains fixed while its contents move underneath it. Woke functions as accusation, counter-accusation, moral defense and category reversal inside the same highly rewarded discussion.

The object of study is therefore not the people supposedly described by the category. It is the classification behavior itself: who defines it, what is placed inside it, what is excluded from it, and which version receives recognition.

This is category capture.

A category compresses variation into something recognizable. Once it becomes familiar, the choices that produced that compression become harder to see. A reader encounters woke, rich, middle class or working class and supplies a meaning almost automatically, even though another reader may be supplying a materially different one.

The underlying circumstances can be real while the boundary remains contingent. What becomes consequential is which definition acquires enough authority to stand for the variation underneath it. Institutions can allocate through categories, researchers can compare through them, and audiences can judge through them. A representation can begin looking less like one way of organizing reality and more like a property of reality itself.

Authority over a subject can therefore become authority to determine what the subject means.

Class presents a different classification problem.

Woke is a label whose contents can be reassigned. Class begins by dividing a much more continuous field into named populations. Before “class warfare” can mean anything, participants have already had to decide who the classes are: rich and poor, working and middle class, wealthy and disadvantaged, oligarchs and everyone else.

The headline Is This a New Era of Class Warfare? therefore contains more structure than it first appears to. “Class” establishes the sides. “Warfare” establishes the relationship between them.

The comments immediately begin renegotiating both.

The article drew 1,708 displayed comments. Its leading Reader Pick argued that “we’ve been in a class war since Reaganomics” and received 2,862 recommendations. Other highly recognized comments placed the aggression among wealthy political donors and oligarchs, while another asked why class warfare seems to describe the behavior of disadvantaged people but not the behavior directed toward them.

Responsibility remains contested. The representational axis does not.

Readers disagree about who began the war, who belongs on either side, and what behavior qualifies as warfare; in doing so, they compete to determine what class warfare itself will mean.

This is the fight for the mean: not an arithmetic average, but the struggle over which version of the category becomes representative.

The same problem becomes even clearer when the boundary itself is unstable. An essay criticizing colleges for paying “rich kids” to enroll produced readers who immediately wanted to know who, exactly, belonged inside rich.

Some described themselves as neither wealthy nor poor. Others described themselves as middle class despite expecting financial-aid formulas to classify them as wealthy. Another Reader Pick simply asked the author to define “rich.”

The category was already operational. Its boundary was not.

A category does not need to be precise to become consequential.

The snapshot also provides an important negative case.

Race is prominent in the Harry and Meghan article, yet the leading Reader Pick received 3,347 recommendations while explicitly accepting that racism existed and rejecting race as the principal explanation. Monarchy, duty, agency and personal conduct were offered instead. Other highly recognized comments made related moves.

Here the proposed representation does not simply accumulate recognition. Its authority is contested and partially displaced.

That distinction is essential. Category capture cannot mean that a category was mentioned. A useful research object must be capable of gaining authority, losing it, having its boundary altered or being replaced.

The $20 burrito discussion is useful for a different reason.

The article begins with affordability. Its leading Reader Pick moves through housing, education, medical costs, wages and the concentration of economic rewards before ending with three words:

“Eat the rich.”

3,431 recommendations.

Another highly recognized reader describes a household with advanced degrees and director-level jobs that is nevertheless “allegedly comfortably middle class” while struggling with housing and living costs.

Unlike the class-warfare article, the category was not supplied in the headline.

The discussion moved there.

Class-related language also travels beyond explicitly class-framed stories. The $20 burrito discussion is one example. Whether that portability persists is something later observations can test; MRI-002 does not need to decide it here.

A contemporary snapshot shows the category being fought over at one moment. The historical archive lets us ask whether that instability is new.

It is not.

Across the 2017–2018 historical corpus, an exact word-boundary audit found 1,437 candidate comments and 1,494 matched tokens. After ordinary wake/sleep verb uses were excluded, 391 comments remained, containing 418 categorical matched tokens. The uses were mixed from the beginning: positive self-description, political awareness, criticism, group labeling and arguments about the word itself.

In January 2017, a Women’s March comment declaring “We will stay woke” received 707 recommendations. The same historical period also contains readers talking about “woke folk,” criticizing a “woke professor,” disputing what woke properly meant, and later describing a “woke intellectual elite.”

By 2018, the term appears as praise, ridicule, political identity, cultural shorthand and something readers explicitly argue should be abandoned.

The historical record therefore does not give us a clean story in which woke begins with one meaning and gradually becomes another.

It gives us something more useful.

The word persists while people repeatedly renegotiate what it contains, whom it describes and who gets to use it.

The 2026 phrase “right wing wokeness” is another move in that continuing struggle: the category is transferred onto a group that would ordinarily reject its ownership.

The question for MRI is not which definition of woke is correct. It is how a representation becomes portable enough to survive competing definitions at all.

These examples reveal the mechanism, but examples can always be selected.

MRI-002 therefore moves from the contemporary snapshot to the full historical corpus: more than 2.1 million normalized New York Times comments across thousands of discussions.

The purpose of scale is not to find more quotations that fit the theory. It is to ask whether the apparent structure survives outside the examples that first made it visible.

The historical corpus gives us repeated discussions, early comments, later recognition and known outcomes. That makes it possible to move from illustration toward a test.

The archive also makes clear that class language itself is not new. A 2018 inequality discussion moves immediately through working class, middle class, wealthy capture, oligarchy, race and the “woke intellectual elite.” That is enough to prevent the 2026 examples from being mistaken for evidence that class has suddenly appeared.

The archive also lets us hold the object roughly constant while watching the representations around it change.

Donald Trump provides the obvious case. He remains a major object of attention in the 2026 snapshot and is already pervasive in the 2017–2018 material. That persistence is not itself surprising. It is useful.

Hold the person roughly constant and the surrounding explanations can be watched moving around him.

On February 15, 2017, Trump Aides Had Contact With Russian Intelligence begins with a comparatively concrete reported event.

The abstraction happens quickly.

By comment 13, a report about contacts with Russian intelligence had become an impeachment question.

No political judgment is required. The observable movement is enough: a comparatively concrete event enters the discussion and increasingly consequential abstractions begin organizing what the event is taken to mean.

The historical material contains something deeper as well.

In April 2018, Public Servants Losing Foothold in Middle Class produced an early comment arguing that public service had been downgraded in the national consciousness. Government was represented as ineffective; that representation justified defunding; reduced resources made government less effective; the resulting deterioration then appeared to confirm the original belief that government does not work.

The comment received 1,266 recommendations.

That sequence is closer to the full object of Magnetic Regression.

The representation is no longer merely organizing an argument. It may be participating in the production of the conditions later offered as evidence for it.

The same discussion immediately begins fighting over working class and middle class. One reader asks when working class became different from and inferior to middle class. Another asks directly: “What is the middle class?”

Later highly recognized comments return to Reagan, taxation, unions, privatization and wealth concentration.

These cases expose different forms of the same observable behavior.

With woke, the label remains while its ownership and contents move.

With class, a continuous field is divided into named populations before the conflict can even be explained.

With the $20 burrito, a story that does not begin as a class story is recruited into class language.

With Harry and Meghan, one proposed explanation loses authority to another.

With Trump and Russia, a comparatively concrete report is rapidly recruited into increasingly consequential abstractions.

With public servants, a representation may even participate in producing the conditions later used to confirm it.

MRI is not trying to decide which of these categories is correct. It is trying to locate the movement by which a representation begins acquiring enough recognition to organize what comes next.

The examples alone still leave an obvious problem:

they were chosen because they are interesting.

So MRI-002 stopped looking for examples.

A simple early-versus-later procedure was developed on 120 historical discussions and then frozen. It was applied unchanged to 813 additional discussions that had played no role in building the test.

In 699 of those 813 unseen discussions, later high-recognition comments were, on average, more aligned with the representational structure visible in the first 25 comments than matched ordinary comments appearing at essentially the same stage.

Across the individual winner-versus-control comparisons, the future recognition winner was more aligned with the early field 61.4 percent of the time.

The relationship remained after accounting for comment length, which was itself a substantially stronger predictor of recognition.

The test does not prove category capture, and semantic alignment is not a direct category-capture score.

It establishes something narrower:

the early representational field contains measurable information about what recognition subsequently selects.