Chapter 23. Dangerous Compression
I. How Necessary Compression Oversteps Its Bounds
Compression makes judgment possible in the first place. Without compression, people could not judge, institutions could not record, history could not be transmitted, and theory could not be expressed. The problem lies not in compression itself, but in its being so useful. An entirely ineffective image can hardly sustain people's belief over time; what is truly dangerous is often an image effective only within a limited scope. It captures part of reality, so people are willing to keep using it. The more it can explain part of reality, the more readily it is mistaken for an explanation of all reality. This is the entry point into dangerous compression.
Validity within a limited scope needs particular clarification because it is both the reason an image can be used and a condition under which it may overstep its bounds. In a case of unpaid wages involving the family of a worker from elsewhere, interpreters can form an image along lines of labor, wages, working hours, debt, housing, and risk that captures the heavy consequences now occurring. In a child's school admission, those handling the matter can form an image along lines of documents, archives, eligibility, and mutual recognition that likewise shows how institutional nodes restrict passage through procedures. Weddings and funerals, accents, exchanges, and historical indices within local communities can reveal boundaries of recognition and responsibility. The danger is not that these extractions were once effective, but that users forget what the image downweighted and come to treat a partial image as the whole of reality. When users repeatedly see wages and labor risks in unpaid-wage cases, they may carry this set of effective extractions into different relational fields, such as school admission, local recognition, or kinship responsibilities. If the arrangement of weights originally suited to one question is no longer re-examined, a limited image may be treated as a fixed entry point.
What must therefore be asked is not whether an image was once effective, but who uses it in which relational field, from what materials it was formed, and whether it allows other differences to re-enter. Users of theory, institutional personnel handling matters, and disseminators may all carry forward the same image, but what they carry forward is a way of undertaking distinction extraction, not reality itself. Only when the boundaries of distinction extraction within a limited scope are forgotten does the image's clarity turn into occlusion of the remaining differences.
When the same image is used at different positions, sameness of materials must also be distinguished from sameness of weights. Wages, documents, and local reputation can all be present at once, but the questions faced by those handling unpaid wages, by school-admission procedures, and by community members differ, and the same ordering cannot be retained. Only if users preserve one effective judgment as an answer valid in every relational field does the image begin to detach from the conditions that formed it. Keeping those conditions visible allows limited distinction extraction to remain open to comparison; only when the conditions are erased can a partial image acquire unrestricted explanatory standing. Comparison among images must retain this condition as well. If two explanations extract different materials, the fact that one was effective in a particular question cannot authorize it to govern the other in advance. Comparison should return to the current relational field, establish which nodes, relational chains, and remaining differences each explanation sees, and use whether each explanation can accept counterexamples and revision as an important boundary for judging and calibrating limited explanations.
II. Compression Loss and the Loss of Entry Points
Every compression makes some differences recede; receding is not disappearing. Calling a family a “low-income family” may capture its income difficulties, but does not mean that income is its only problem. It may simultaneously face elder care, medical history, migration, debt, children's school admission, historical indices of kinship, local relations, household-registration eligibility, and marital pressures. If explanation sees only “low income,” the other differences recede, but they do not disappear: the sick still need medical care, elders still need care, debts still fall due, children still need to enter school, historical indices of kinship are still extracted in borrowing money, household division, and relationships and marriage, and local relations still play a role in finding work, renting, school admission, and wedding and funeral exchanges.
Likewise, calling a community a “backward culture” does not mean that its circumstances are in fact determined by culture alone. Institutional pressures, historical harms, state borders, economic circumstances, educational resources, and external hostility may all have been downweighted. Calling a state a “machine of oppression” does not mean that oppression is all it does. Passage enabled by documents, public order, assistance, education, border protection, allocation and coordination in disasters, and resource allocation may also be downweighted by this designation. Writing a historical community as purely a victim does not mean that it contains no disagreements, mixed settlement, protection, intermarriage, boundaries of responsibility, or actual interests. Downweighted differences do not disappear because an explanation has not extracted them; they have only temporarily remained outside the current image for judgment. The danger begins here: explanation treats differences that have not entered the image as though they had never existed.
The distinction between downweighting and disappearance must also be identified in the materials. Medical records, debt deadlines, exchanges among relatives, local reputation, and institutional records may remain preserved at different nodes even if an explanation has not extracted them. When events change, these materials can become visible again along new relational chains. If interpreters acknowledge the possibility of extracting them anew, the earlier omission remains merely a selection within a limited judgment.
The real risk appears when users take unextracted differences to have no actual efficacy. At this point, even people whose differences have been downweighted may be told that the materials they present about medical history, care, migration, or historical indices are irrelevant to the question. What is excluded is not abstractly “more detail,” but concrete differentiating units that could have changed eligibility, responsibility, risk, and consequences for action. Nor does a difference regaining weight mean that the earlier judgment must have been wrong. Income can be the dominant weight in one instance of handling unpaid wages; only when the question turns to care, school admission, or local recognition do other materials acquire a greater role. Healthy explanation permits this shift in weights and can explain why judgment changes. Closed explanation fixes an earlier extraction as a permanent feature of the object, so that later materials can only be downgraded to exceptions. The boundary between the two lies in whether users acknowledge that a change of relational field requires renewed distinction extraction.
If downweighted differences have merely receded temporarily, the problem is not yet at its most serious. As long as explanation still allows the image to be reopened, these differences may return within new relational fields. What is truly dangerous is compression loss: certain differences lose their entry points back into explanation. Healthy compression merely downweights differences temporarily while retaining an entry point for reopening. Heavy compression not only downweights differences but makes it increasingly difficult for them to re-enter explanation. Closed compression treats the image as reality itself, refusing to let counterexamples, remaining differences, and other nodes re-enter. Dangerous compression downweights differences and cuts off their entry points.
The obstruction of entry points for naming takes the form of differences being difficult to articulate if they have no names. A family bears the pressures of low income, medical history, elder care, migration, historical indices of kinship, and school-admission eligibility at the same time. If it is named only a “low-income family,” the other differences become difficult to voice. Elder care is classified as “an internal family matter”; medical history becomes “a personal health matter”; migration is described as “their own choice to move”; historical indices of kinship are absorbed into “family affairs”; and the child's school-admission eligibility is reduced to “incomplete paperwork.”
The loss of entry points through indices means that differences are difficult to extract anew if they lack place names, archives, stories, records, and evidence. If a local conflict leaves only a reputation for “households from elsewhere causing trouble,” with no record of land allocation, the course of migration, or old contracts, later people can hardly reopen the antecedents. They can only continue judging along the lines of the reputation that remains: who causes trouble, who will not stay in line, and who is not local.
The loss of entry points into explanation lies in differences being difficult to bring into judgment if theory does not recognize them. If a theory recognizes only labor, income, property, debt, and capacity for domination, nation, locality, family, kinship responsibilities, and historical indices are readily downgraded to subordinate differences. Even when people see these differences, they will say they are merely surface phenomena, secondary matters, forms of occlusion, or displacements.
The loss of entry points into action means that differences can hardly change how matters are handled if they cannot enter institutions, public opinion, or community action. Once a person is compressed into an “object of management,” their care responsibilities, historical circumstances, local relations, and community recognition may be difficult to bring into actual handling. When institutions deal with the person, they see only records, documents, risk categories, and management requirements. When public opinion discusses the person, it sees only trouble, order, and cost. When community members avoid the person, they too see only the label and no longer the relations. The deepest loss in dangerous compression is that certain differences no longer have names, indices, positions in explanation, or entry points into action. These differences have not disappeared; they are trapped outside the image.
These four kinds of entry point do not form a list of mutually isolated entries. Differences without names are difficult to record; differences without records and indices are difficult for users of theory to extract anew; and differences that cannot enter explanation are in turn difficult to bring to bear on how institutions, public opinion, or communities handle matters. Compression loss may therefore accumulate along nodes and relational chains, but these are not inevitable stages. Preserving any one entry point may allow materials to reappear in a new relational field.
Entry points are sustained at concrete positions of use, and different people may control entry points for naming, indexing, explanation, and action. Family members can articulate care burdens; archives preserve medical histories or qualifications; local stories preserve migration and contracts; and record keepers decide whether these materials are preserved. Interpreters recognize national, local, and kinship relations; users of theory decide whether materials can acquire explanatory standing; institutional personnel bring materials into judgment and decide whether they can change actual handling. Together, these positions constitute the conditions under which differences regain a position. Only by distinguishing these actors can we judge precisely where a difference lost its entry point and through which node it can still return.
Losing entry points also changes who has the capacity to present differences. When records retain only management categories, even if the people being processed know their own care responsibilities, migration, and local relations, they may still be unable to bring these materials into institutional positions or positions within public opinion. Later people, in turn, can read only the existing classifications. Downweighting thus ceases to be merely an omission in one judgment and changes the possibility of materials passing between nodes. The conditional character of this mechanism must be retained: entry points may be weakened or cut off, but not every compression necessarily loses all four kinds in sequence. Conversely, if positions of naming, recording, theory, and action accept only existing labels, the remaining differences will be trapped outside the image.
III. The Hardening of Images and Fixed Labels
An image for judgment initially helps people grasp a set of differences within complex reality, but use and dissemination may gradually harden it. Statements such as “They are the bottom stratum” circulate more readily than an account unfolding wages, debt, household registration, medical history, kinship responsibilities, and local relations. “This is a backward culture” dispenses with institutional pressure, historical harm, state borders, and economic circumstances. “The state is a machine of oppression” is also more striking than tracing documents, archives, borders, assistance, public order, and chains of implementation. Management favors categories, mobilization relies on slogans, argument seeks opponents, and dissemination needs short sentences. Through these practices, an image originally used provisionally acquires the appearance of reality itself. Once users forget that an image is only a provisional low-dimensional tool, they stop using it to return to complex reality and may instead use it to prevent other differences from entering explanation.
The hardening of images is brought about by concrete carriers operating at different nodes; not all hardening can be attributed to a single abstract force. Disseminators choose short sentences that are easy to repeat, relying on repeatability; management procedures require stable categories, relying on classifiability; mobilizers favor clear boundaries; and participants in arguments need both to identify opponents quickly and to classify quickly. Each of these positions selects some differences, assigns relatively high weight to the same name through repeated use, and continues using the same compression.
The format of materials also affects this process. Short sentences can be transmitted quickly, categories readily enter forms and records, and labels readily connect to existing reputations. Medical history, kinship responsibilities, the course of migration, and historical indices, by comparison, often require longer explanations. When processing positions accept only formats convenient for transmission, differences that resist compression into short sentences are more likely to recede. Different positions harden images in different ways; these ways may connect with one another or occur only within a limited scope. What matters is tracing precisely which differences are retained as materials pass from one position to another, and which drop out because of format requirements.
An image for judgment can still serve as a provisional tool, provided users know when it needs reopening. If new materials can change existing weights as they emerge, counterexamples can compel interpreters to revise, and other nodes can fill gaps in severed relational chains, the image has not closed. Hardening occurs when users stop examining these conditions and treat whether the object fits an established category as the only question to be answered.
Dangerous compression cuts continuous differences into labels. Income admits no single clean division: earning slightly less or more in a month, carrying debt or not, renting or not, having elders and sick people at home or not, and having a fallback or not all change circumstances. Yet dangerous compression cuts this set of continuous differences into “the bottom stratum,” “the middle class,” “the poor,” or “the vulnerable.” People with similar incomes may bear different risks because their debt deadlines, housing stability, care burdens, and width of available exits differ. It is precisely these variable weights that the labels omit.
Local recognition admits no single clean division either. How long a person has lived somewhere, whether they attend weddings and funerals, whether they understand the dialect, whether they have ties with people from the same place of origin, whether their children attend school locally, and whether their relatives live locally all change their position. Yet dangerous compression divides these continuous relations into “locals” and “people from elsewhere.” People all described as being from elsewhere may inhabit different relations because their length of residence, wedding and funeral exchanges, and local recognition differ. Historical processes likewise admit no single clean division. Migration, mixed settlement, conflict, intermarriage, protection, betrayal, distribution, and mediation all form later relations. Yet dangerous compression cuts complex processes into “victim,” “perpetrator,” “from elsewhere,” “unruly,” or “backward.” Labels originally help judgment; the danger is that they turn back and swallow up continuous differences. Differences swallowed by labels often do not disappear immediately. They become difficult to articulate, find, explain, or act on.
Those who name and use names perform this division. When disseminators, institutional personnel, or community members receive an object under one name, they often simultaneously determine which materials deserve further questioning. Intermediate states falling outside the label's boundaries are not readily recorded, leaving later people able to understand only through the classifications left behind. A label can be a preliminary entry point, but cannot be the endpoint of judgment. As long as users still ask how income, debt, local relations, migration, and historical materials combine, continuous differences can be unfolded again. If a name directly replaces these questions, the label does more than compress information: it restricts which differences can acquire a position in explanation.
Continuity also reminds interpreters that materials near the boundaries of labels are especially important. A slight change in income, a longer period of residence, resumed exchanges among relatives, or new historical materials may all change an object's concrete position in reality. If an existing classification system permits only a choice between two presupposed extremes, these subtle dynamic changes cannot alter actual handling. Re-examining these intermediate states does not abolish classification; it prevents classification from occluding changes in relations now taking place.
IV. Default Extraction and a Single Position
Once a label stabilizes, it decides in advance for later people which differences should be extracted. Once dangerous compression stabilizes, later people stop differentiating anew and automatically extract familiar differences when they see a label: on seeing “the bottom stratum,” they default to extracting poverty, being oppressed, being without property, and being dominated; on seeing “people from elsewhere,” they default to extracting accent, household registration, instability, and nonlocal relations; on seeing “backward culture,” they default to extracting ignorance, conservatism, and tradition, while downweighting institutional pressures, historical harms, state borders, and economic circumstances; and on seeing “machine of oppression,” they default to extracting control, violence, and deprivation, while downweighting passage enabled by documents, public order, assistance, and protection.
Default extraction is not simple prejudice, but inertia in distinction extraction formed after a compressed image hardens. Prejudice can be a personal attitude; default extraction goes deeper. Explanation no longer returns to the concrete relational field each time or asks anew which differences are gaining weight in this object, but automatically extracts familiar differences along the lines of the old image. When default extraction occurs repeatedly, explanation ceases to face new relational fields and repeats the same set of distinction-extraction operations on different objects. This is how differentiation bias forms.
Default extraction relies on later people's trust in existing names. Someone newly entering a processing position may never have seen the original materials, yet can receive a completed judgment through archives, reputation, classifications, and short sentences. Here, the label is not choosing differences on its own; it offers users a path that dispenses with differentiating anew. Once this path is repeatedly adopted, users assign relatively high weight to familiar differences: on seeing the bottom stratum, they first ask about wages and domination; on seeing people from elsewhere, they first ask about household registration and accent; on seeing an object of management, they first look at records and risk categories. These extractions may be effective in certain relational fields, but if the current question is no longer examined, the original fit becomes inertia.
Inertia in distinction extraction also affects how questions are asked. If later people accept a label first, their questions often merely seek to complete the existing image—for example, asking only about wages, household registration, or risk categories, without asking about medical history, local relations, and kinship responsibilities. Who frames the questions themselves and which answers are permitted determine whether new materials can acquire weight. Once default extraction forms, even new materials that are mentioned may be placed back into the old image's existing classifications. If local relations, care responsibilities, and institutional eligibility can be described only as surface, secondary, or subordinate differences, they have not actually acquired a role in changing weights. It is concrete interpreters and processing positions that handle matters in this way; the label does not act on its own. Only by identifying the position in explanation into which materials are placed can we judge whether they have actually entered the image and changed the original distinction extraction. Differentiating anew must therefore also include re-examining the questions.
Correcting default extraction does not mean forbidding names. It requires later people to face the materials anew. They need to establish which nodes this event involves, which differences are gaining weight, what existing labels have downweighted, and whether new counterexamples can change judgment. Retain these questions, and the label remains an entry point; cancel them, and the label seals off other entry points in advance.
The family of a worker from elsewhere is not a single position. Interpreters using class theory can extract its labor, wages, housing, debt, and risks. Interpreters using a state image can extract household registration, residence permits, social insurance, student registration, and administrative eligibility. Those using an image of local networks can extract accent, ties with people from the same place of origin, exchanges with neighbors, and local recognition. Those using a historical image can extract migration routes, family historical indices, and local memory. The same family displays different differences in different relational fields. Yet after default extraction occurs repeatedly, the family is compressed into a single position: the bottom stratum, people from elsewhere, a mobile population, a backward culture, the vulnerable in the market, or objects of state management.
Each of these designations may capture part of reality, but once any one masquerades as the whole, it occludes other differences. Say only “the bottom stratum,” and household registration, local recognition, national memory, children's school admission, and kinship responsibilities are downweighted. Say only “people from elsewhere,” and labor risks, debt, housing, and capacity for domination are downweighted. Say only “objects of management,” and care responsibilities, the course of migration, community recognition, and historical circumstances are downweighted. Dangerous compression compresses a complex differentiation position into a single position, allowing a person to be understood through only one label.
A differentiation position forms through the joint action of multiple differentiating units within a concrete relational field; it is not a simple sum of identity names. Different materials, nodes, and relational chains change the position of a family of a worker from elsewhere in questions such as unpaid wages, school admission, renting, weddings and funerals, and assistance. Interpreters using any image can extract only part of this and must not rewrite the rest as subordinate. To reopen a differentiation position, interpreters must return to the event now unfolding. Who is handling it, which materials have been submitted, which qualifications are being compared, and which relational chains change the consequences must all be established anew. Doing so does not arrange all differences on an equal footing; it allows differences currently gaining weight to revise the weights fixed by a single label.
V. Historical Processes Must Not Be Compressed into Fixed Labels
Historicization must compress the past; without compression, the past cannot be transmitted. Yet dangerous compression reduces complex historical processes to fixed labels: migration to “from elsewhere,” mixed settlement to “unclear boundaries,” shared harm to “eternal victimhood,” state punishment to “a person who will always be a problem,” and family historical indices to “that branch of the family owes something.” These labels sometimes preserve certain past differences, but they also cut off processes. If a branch of a family once contributed less money during household division, this may have been because someone was ill, working away from home, caring for elders, or bearing other debts at the time. If history is compressed into nothing but “that branch of the family owes something,” later judgments about relationships and marriage, borrowing money, ancestral rites, and land distribution will continue to follow this label. When a local conflict is compressed into “households from elsewhere causing trouble,” migration routes, land allocation, old contracts, marital relations, and local mediation may lose their entry points through indices. Once historical labels sever processes, they turn complex past differences into rigid future judgments. The past does not automatically bear down on the present; a compressed image of the past is repeatedly extracted in later relational fields.
Historical materials need to be transmitted through names, stories, archives, and shared memory; otherwise later people cannot approach the past. The problem is that preserving certain indices may also delete the conditions at the time. Once reasons for migration, land allocation, care burdens, debts, and relations of mediation are no longer recorded, later people receive only conclusions such as being from elsewhere, owing something, or being unruly.
When fixed labels are carried into new relational fields, users treat past images as presupposed entry points for present judgment. Relationships and marriage, borrowing money, ancestral rites, and land distribution each involve different materials and relational chains. Yet the same historical name may prearrange who is trustworthy, who bears responsibility, and who should be excluded. These consequences come from later people repeatedly extracting compressed records, not from the past itself acting automatically.
Preserving historical indices must be distinguished from fixing historical conclusions. Place names, contracts, migration routes, divisions of work within families, and local mediation can allow the past to be re-examined. As long as they can still be extracted anew, concrete materials may correct fixed labels. Family stories can likewise allow the past to be re-examined. If names such as being from elsewhere, owing something, and being unruly no longer connect to their processes of formation, only the outcomes of judgment remain. Community members, record keepers, and later interpreters may all participate in this transmission. Theory needs to trace which materials they use, rather than let historical images replace actual positions of preservation and extraction. Only if records preserve conclusions while severing their antecedents will historical compression constrain new judgments over the long term.
VI. Action Write-Back and the Self-Reinforcement of Images
Dangerous compression does more than misread reality; it reorganizes action according to compressed images. When a person is called part of “the bottom stratum,” their ethnicity, locality, family, medical history, debt, and kinship responsibilities may be downweighted, and they may also be required to understand themselves through a class image. They speak of local exclusion, and others say it is merely the surface of class contradictions; they speak of kinship responsibilities, and others say these are merely old family relations; they speak of national memory, and others say it is merely an occlusion of consciousness. These differences have not been directly eliminated from the person, but have lost their entry points into explanation and action.
When a community is called a “backward culture,” its institutional pressures, historical harms, state borders, and economic circumstances may be downweighted. External action turns into cultural transformation, and internal defenses are likewise forced to revolve around culture, pushing institutional questions, questions of boundaries, resources, and history to the rear.
When a state is called a “machine of oppression,” its protective functions, production of passability, allocation and coordination of public resources, assistance, and border protection may be downweighted. Later people discussing the state then default to extracting control, violence, and deprivation. Dangerous compression not only occludes reality; it also prescribes how later people continue extracting differences.
At a deeper level, interpreters, institutional personnel, or procedures using dangerous compression do more than explain reality. They produce new records, new actions, and new evidence according to their images, then use these consequences to prove the original images. When a person is compressed into an “object of management,” institutional personnel treat them as an object of management. After they have been restricted, scrutinized, and excluded, archives and public opinion will in turn show that this person does indeed frequently come into contact with the management system, so the original label appears to have been proved. In fact, users first organize action through a compressed image, then treat the consequences of that action as evidence for the original image.
This reveals the distinction between an image as an explanatory output and an image brought into action. Once users bring a compressed image into action, they produce concrete consequences in reality and form a loop: users undertake distinction extraction according to the image; distinction extraction guides action; action leaves records; and those records in turn reinforce the image.
Write-back occurs first in how matters are handled. Institutional personnel determine actual handling according to management categories; positions within public opinion select materials for dissemination according to existing labels; community members decide whether to approach or avoid someone according to reputation. The image itself has no will. It is these users and the procedures in which they operate that actually bring compression into consequences. Handling at different positions may connect, or it may leave records only within a limited scope.
Action in turn leaves new materials. Someone repeatedly scrutinized accumulates more management records in the archives; someone avoided by a community has less evidence of exchanges; someone forced to defend themselves in terms of culture leaves more cultural expressions. If later people do not ask how these records arose, they may treat results produced by action as proof that the object fitted the label all along. This loop does not require a centralized controller. Different nodes continuing to use the same image may separately perform extraction, handling, recording, and reinterpretation.
Consequences produced by write-back further change how the object can be seen. Restrictions may reduce their exchanges, scrutiny may increase their management records, and cultural transformation may force their defense to focus on cultural names. These new materials have all actually been left behind, but their formation has already been influenced by earlier images. If later interpreters treat them as original evidence independent of the handling process, they occlude how images participated in the consequences. Tracing the conditions under which records formed is necessary work to prevent the loop from closing in on itself.
Action write-back may also change the allocation of responsibility. When someone is handled solely as an object of management, institutional personnel are apt to exclude care, migration, and local recognition from judgments of responsibility. When a community is understood solely as a backward culture, external actors push institutional and resource questions into the background. Specific people using compressed images thus do more than arrange what is seen; they also affect who is required to explain, bear responsibility, or change.
The conditional character of the loop must be retained. Once users bring images into action, concrete handling may leave records that reinforce the original images, but new nodes, counterexamples, and materials may still interrupt the connection. To interrupt the loop, new materials and counterexamples must be allowed into one of its positions, enabling subsequent users to see that records are evidence but may also be consequences left by earlier ways of handling matters. If institutional personnel re-examine the process that formed a record, or community members add omitted exchanges, later explanation need not close along the lines of the original image. Write-back has real force, but it is not an irreversible automatic determination.
VII. Dangerous Compression in the Class Image
The class image is one of the most typical forms of dangerous compression, but this must be stated clearly: class is neither an underlying entity of society nor a discovered underlying reality. Class entities do not exist. Real differentiating units such as labor, income, property, debt, housing, risk, fallbacks, and capacity for domination supply the material basis. The processes of extracting, assigning weight to, connecting, and compressing these units form partial explanatory power; the class image is the low-dimensional output of this process. These differences really exist and powerfully change people's lives. Someone lives by selling their labor, earns low wages, carries debt, has unstable housing, works in danger, has no fallback, and is dominated by others. These differences are of course weighty. Opposition to the ontologization of class must not be understood as diminishing the weight of real pressures. Unpaid wages, debt, dangerous work, unstable housing, and the absence of fallbacks still produce heavy consequences. What must be rejected is the use of a low-dimensional projection to monopolize the entire explanation of these consequences.
The class image is so readily seductive precisely because distinction extraction within a limited scope from these real, weighty differences is not fabrication out of nothing. Such limited distinction extraction is not entirely wrong. Yet partial explanatory power comes from real differentiating units and the processes of extracting, assigning weight to, connecting, and compressing them. The class image itself is only a brutal projection of high-weight differentiating units onto axes of economic, labor, property, and domination relations; it is not what society fundamentally is. Precisely because it presents part of the reality extracted, it is more readily mistaken for society itself. Yet the claim that concrete differentiating units and their processing form partial explanatory power differs from the claim that class exists as an underlying reality. We cannot infer from the former that a social entity named class partially exists. Class theory still belongs from the outset to the total-image subsumption direction of Model C; the class image itself does not thereby acquire the standing of an ordinary limited tool. Here this chapter addresses how projection oversteps its bounds, without anticipating the full account of model structures in other chapters.
Once interpreters forget that the class image is only a brutal projection, they mistake a low-dimensional image for society's underlying reality. They use projection onto economic, labor, property, and domination axes as the explanatory premise for other differences, downweighting household registration, locality, nation, sex, age, medical history, kinship responsibilities, institutional eligibility, historical indices, and community networks. Household registration, local recognition, national memory, sex, age, medical history, kinship responsibilities, and institutional eligibility are thus readily reduced to surface or subordinate matters. Once interpreters using class theory pass this image off as underlying reality, they describe differences that class theory has not extracted as subordinate or secondary matters, forms of occlusion, falsehoods, or displacements. This is what is most dangerous about class theory: it is dangerous not because it is wholly wrong, but because it extracts real pressures and more readily leads people to believe it has explained everything.
The projective relation of the class image also requires distinguishing extraction from naming. Real differentiating units such as wages, labor, property, debt, housing, risk, fallbacks, and capacity for domination can change people's eligibility, choices, and consequences within concrete relational fields. They therefore deserve to be extracted and assigned corresponding weights, and each can also be demonstrated by materials and acquire high weight in different relational fields. Yet projecting them and naming them class does not discover a common entity hidden behind the materials. Once this distinction is erased, interpreters using class theory may require differences of household registration, nation, locality, family, and history all to be reinterpreted along the same axis.
In the concrete arrangement of weights, the danger lies not in continuing to trace wages and debt, but in refusing to let other differences now gaining weight change the image. Wages, debt, and capacity for domination can indeed change circumstances more directly than other differences in certain relational fields. Yet high weight must not be illicitly turned into sole weight. Whenever medical history, household registration, local recognition, or kinship responsibilities regain weight in an event, interpreters must allow them to revise the original projection.
VIII. The Compression Boundaries of Other Images
The class image is not the only image that can overstep its bounds. State, market, and cultural images can also undergo dangerous compression. Concrete materials, nodes, and processing involved in state certification and implementation can form partial explanatory power: a document can change passage through institutions, an archive can change eligibility, a border can change status, and a punishment record can change the future. Interpreters using a state image extract materials and nodes such as documents, archives, borders, eligibility, mutual recognition, and implementation. Tracing these differences can explain how passage, education, assistance, or sanctions occur. Yet once interpreters let the state image masquerade as all of society, they downweight family, markets, localities, communities, national networks, and historical memory. All consequences are described as the will of the state, so that family responsibilities, market conditions, local networks, and historical memory can no longer change judgment. This boundary limits the image's scope of application; it does not deny the actual action of state nodes.
A market image develops around prices, wages, rents, debt, and purchasing power. Concrete differences and their processing can form partial explanatory power and explain some pressures on resources: a family may be evicted because of rent, may not dare to fall ill because wages are unstable, and may be forced to accept dangerous work because of debt deadlines. Yet such limited explanation cannot automatically explain care, institutional eligibility, local responsibilities, and boundaries of national communities. Once interpreters using a market image let it masquerade as all of society, they downweight family care, institutional eligibility, local responsibilities, boundaries of national communities, and historical indices.
Concrete differences such as language, rites and customs, education, values, and orders of meaning, together with their processing, can likewise form partial explanatory power. What language a person speaks, what rites they observe, whom they recognize as kin, and how they regard marriage, death, education, and responsibility may all change their relations. Yet once interpreters let a cultural image masquerade as all of society, they downweight institutions, the economy, the state, risks, violence, history, and actual interests. Interpreters using a cultural image may extract language, rites and customs, education, and orders of meaning, but must not relegate violence, risks, state implementation, economic conditions, and historical materials to irrelevance.
These three images form different connections among materials. A state image depends on nodes of certification and implementation; documents and archives enter consequences through certification and implementation. A market image depends on relations of prices, wages, rents, and debts; prices and debts change choices through exchange deadlines and the ability to pay. A cultural image depends on language, rites and customs, education, and orders of meaning; language, rites, and customs affect action through relations of recognition, education, and responsibility. The differences they downweight, and the relational chains through which they form consequences, differ. In comparing these images, we should not write them as having the same content. Nor does this comparison include class theory, which belongs to Model C from the outset. Only when users elevate their limited scope of fit into the whole of society do they share the structure of dangerous compression. Their shared risk is not that users have extracted wrong materials, but that they expand the effective scope to all of society. Calibration requires specifying, separately for each image, which nodes its users rely on, which relational fields it fits, and which differences are downweighted. This preserves the partial explanatory power formed by the relevant differentiating units and their processing while preventing users from giving any single image a standing that excludes other explanations. Clarifying these relations limits each image's boundaries of application; it does not establish a list of interchangeable factors.
Explanations concerning nations require particular care. A nation is not a brutal projection in the way class is, nor a subjective imagining, a fictional label, or an appendage of the class image. National communities arise from partially overlapping networks, long-term interaction, shared memory, language, rites and customs, boundaries, and paths of shared responsibility.
Within national networks, members, families, those who transmit things across generations, record keepers, educators, organizers, or disseminators also undertake compression, reducing complex interactions to names, designations, boundaries, responsibilities, and shared memory. This compression is useful. Names connect relations dispersed across language, rites and customs, boundaries, and long-term interaction, and can preserve paths of memory, responsibility, and protection across generations. Community members and later people can use these materials to establish why they are connected to one another. Without names, many responsibilities cannot be transmitted across generations. Without names and boundaries, much shared memory, shared protection, shared responsibility, and shared harm become difficult to transmit. Without shared memory, later people can hardly know why they are connected to certain people far away.
Yet this compression also has dangers. A national community consists of paths of recognition, memory, and responsibility transmitted over time through partially overlapping networks. Such compression may turn real networks into an essence, a matter of bloodline, and something eternal. Dangerous compression describes them as that way by birth, that way in the blood, and that way forever, thereby downweighting migration, intermarriage, shared life, institutional recognition, changing boundaries, and historical contestation, and depriving materials about the process of a position from which to prompt revision. As long as users can still extract migration, intermarriage, shared life, institutional recognition, and changing boundaries anew, this compression retains historical process. The danger of class is brutal projection masquerading as underlying reality; the danger of the nation is the essentialization, reduction to bloodline, and eternalization of historical community networks. These two dangers differ and must not be conflated.
The fact that national names preserve responsibility and memory does not mean that names can replace community networks. A community's actual efficacy still comes from relations formed through long-term interaction, recognition, protection, harm, and intergenerational transmission. Conversely, treating a name as pure fiction also deletes these actual paths. When shared memory is transmitted, users need to preserve the interactions and responsibilities connected to it as well. If only the name remains, without paths of shared life, protection, harm, and recognition, the name is readily essentialized. If all shared memory is deleted out of concern about essentialization, however, a real community loses its historical indices.
Calibrating compression concerning nations must therefore protect two boundaries at once: it must neither write historically formed networks as innately unchanging essences nor, in opposing essentialization, erase communities into fictional labels or deny their basis in real relations. Users of theory need both to explain which paths of recognition, memory, and responsibility actually exist and to allow new interactions and experiences of boundaries to change weights anew. A safe exposition can explain how compression is useful and how it oversteps its bounds only between these two boundaries.
IX. How Return Channels Are Cut Off
Dangerous compression ultimately cuts off return channels and no longer allows people to reopen images for judgment. Once an image hardens, people stop asking which differences were downweighted, which differences are regaining weight in the current relational field, which nodes have gone unseen, which relational chains were severed, which historical indices were misread, and which counterexamples were treated as noise. They ask only whether the object fits the image: Does it belong to the bottom stratum? Is it from elsewhere? Is it a backward culture? Is it a machine of oppression? Is it an object of management? Once it cannot be reopened, explanation increasingly resembles a closed image. A closed image is more than stubbornness. It turns counterexamples into exceptions, remaining differences into noise, other nodes into irrelevancies, and new relational fields into sites for repeating the old image. At this point, explanation no longer learns from reality. It merely searches reality for differences it has already prepared.
A return channel consists of a set of distinction-extraction operations that can be performed anew, allowing interpreters to question downweighted differentiating units, find materials that never entered the image, supply omitted nodes and relational chains, reread historical indices, and let counterexamples change the original weights. It is not an additionally posited legal procedure, but the capacity of limited explanation to keep learning from reality. The cutoff occurs when concrete positions of use refuse these operations: users of theory call counterexamples accidental, management procedures accept only existing categories, disseminators treat remaining differences as noise, and later people look only for materials that fit the label. At these stages, the image's boundaries are treated as reality's boundaries, and new relational fields can only be interpreted as repetitions of the old image.
Whether a channel remains cannot be judged solely by whether interpreters verbally acknowledge complexity. We must see whether materials can actually enter. If medical history, local relations, kinship responsibilities, historical indices, or new nodes still cannot change handling and judgment after appearing, the supposed openness is merely formal. Only when counterexamples can trigger revision does return have an actual effect. Return channels also depend on whether different nodes can correct one another. Archival records, family members' statements, local reputation, and historical materials may conflict. Open explanation allows comparison among these materials; a closed image accepts only the parts that confirm existing classifications. Rejecting counterexamples is not an act theory performs on its own. It is exclusion carried out by users as they select materials, set formats, and handle objections.
After a channel is cut off, supplementary materials may still be seen, but can acquire only a predetermined standing: medical history is called a personal problem, local relations are called secondary background, and historical indices are called accidental exceptions. Materials have formally entered discussion, but have not actually changed weights. Checking a return channel therefore requires looking at whether materials can change explanation, not merely whether they are mentioned.
X. Reopening and the Next Misalignment
Calibrating dangerous compression does not mean abandoning compression: without compression, there is no judgment. It means reopening the compressed image, beginning with the selections and omissions the current image has already made. First ask what users extracted and downweighted through this image, and list the extracted differences and the weights assigned. Then examine which differences and materials never entered the image, and which entry points for naming, indexing, explanation, and action were obstructed by the ways naming, indexing, theory, or action operated. This sequence of questions is not a uniform set of stages, but a set of questions that can correct one another. If an explanation can still be questioned in this way, it retains the possibility of reopening; if it refuses these questions, it is turning into closed compression.
The question must then be returned to a concrete relational field. Examine which downweighted differences are regaining weight, which nodes and relational chains have still not entered the image or been seen, which relational chain was interrupted by existing labels, and which historical indices were misread. Whether new medical histories, debt deadlines, eligibility materials, local relations, or historical indices are gaining weight at this point may change the original judgment. Adding these contents does not mean paying equal attention to every difference. It allows actual changes to regain a position in explanation.
The quantity of materials cannot tell us whether calibration is complete. If the same label still prearranges their standing, adding many details leaves the return channel unopened. The key test is whether users allow counterexamples actually to enter and produce consequences. If counterexamples can only be renamed noise, chance occurrences, or exceptions, the image remains closed. If they can adjust weights, expand the range of materials, or compel an explanation to rewrite its boundaries, compression returns to a revisable limited image. Effective calibration requires new materials to be capable of actually changing weights, judgments about nodes, or the understanding of relational chains. It also requires interpreters to specify which counterexamples would compel them to revise. This preserves a limited image's capacity to accept correction by reality, rather than preserving unbounded complexity.
Reopening also requires explaining the scope of revision. A counterexample may require only an adjustment of weights within the current relational field, or it may expose a node or relational chain long omitted. One revision cannot justify declaring an image wholly ineffective; nor can the fact that an image was once effective justify refusing every adjustment. A limited explanation's capacity for calibration shows precisely in its ability to determine from the materials how far revision must go. This chapter stops at the conditions for reopening and at how a long-standing refusal to reopen supplies a prior mechanism for differentiation bias.
If a compressed image is left unopened for a long time, explanation becomes accustomed to repeatedly extracting the same kind of difference while downweighting other differences now gaining weight. After default extraction occurs repeatedly, explanation no longer faces new relational fields, but repeats the same set of distinction-extraction operations on different objects. Differentiation bias thereby stabilizes.