Chapter 24. Differentiation Bias: The Misalignment of Explanation and Reality
I. From Default Extraction to Systematic Misalignment
Dangerous compression is not compression itself. It occurs when users forget that a compressed image is only a low-dimensional output, forget what they have downweighted, treat downweighted differences as nonexistent, and let labels decide in advance what later people should see. Once labels stabilize, default extraction readily occurs in explanation: on seeing “the bottom stratum,” it extracts wages, labor, debt, housing, and risk; on seeing “people from elsewhere,” it extracts accent, household registration, instability, and nonlocal relations; on seeing “objects of management,” it extracts records, risk, order, and costs. These extractions are not necessarily all wrong. The problem is that explanation no longer faces the concrete relational field anew.
Default extraction deserves to be identified because it turns a previously convenient entry point into the starting point of later judgments. That interpreters first see familiar labels and existing records does not mean that other differences have no effect; the question is whether subsequent inquiry can still bring those differences back into the concrete relational field. The visibility of materials is also conditional. What recorders preserve, what users consult first, and what processing positions require all affect what explanation extracts first. Unless these steps are made clear, a familiar image readily replaces the relational position currently occupied by the object.
Consider again the family of a worker from elsewhere. The husband works on a construction site, the wife does casual work at a restaurant, the child has reached school age, the elder's medical records are at a hospital back home, the rent is nearly due, wages are two months in arrears, the social-insurance record has been interrupted, and the family still owes money borrowed from a relative. If an explanation first sees this family as “workers in the bottom stratum,” it may have grasped real differences: unpaid wages, labor risks, housing instability, and debt pressure are all real differences, and they do carry great weight in the problem of unpaid wages. But if, whatever the later problem, explanation extracts differences only along the lines of the “workers in the bottom stratum” image, default extraction begins to recur. When the child seeks school admission, it still sees only wages and labor relations; when the family is refused a rental, it still sees only poverty and a weak market position; when relationships and marriage meet obstacles, it still sees only income levels; when the elder's medical records affect assistance, it still sees only labor circumstances; and when local reputation affects recognition among neighbors, it still sees only the class image.
If default extraction occurs only occasionally, it is still just a single error; if it occurs repeatedly, inertia in distinction extraction forms. Inertia in distinction extraction means that explanation repeatedly extracts familiar differences along the lines of an old image, rather than facing the concrete relational field anew. Once this inertia stabilizes, explanation no longer asks what other differences in this object are gaining weight at this moment. It asks only whether the object can be placed within an image I already know. Differentiation bias begins here: it is explanation repeatedly extracting the differences it knows while downweighting other differences gaining weight in reality. It is not a difference of opinion, but systematic misalignment in distinction extraction, weight assignment, fit to the relational field, nodes, relational chains, and historical indices.
Ordinary accounts understand bias as a matter of attitude: some sympathize with the poor, others dislike them; some trust the state, others dislike it; some value the nation, others are wary of it; some value the market, others criticize it. These certainly affect judgment, but the concern here is not prejudice at this level. Differentiation bias does not mean “I have my position, and you have yours,” nor “all viewpoints have limits, so they are much the same.” It goes deeper. It occurs in how explanation extracts differences, distributes importance, enters relational fields, identifies nodes, traces relational chains, and reads historical indices. Ordinary prejudice can take the form of likes and dislikes, whereas differentiation bias may occur in handling considered fair by those performing it. Even without ill intent, interpreters will bring familiar judgments into different problems if they cease examining the steps of distinction extraction, weight assignment, and fit to the relational field.
An explanation can be sincere in attitude and still exhibit differentiation bias. Out of concern, it may always extract only labor and poverty; out of a sense of order, only documents and risks; out of a wish to protect the community, only boundaries and identities; out of a critical spirit, only oppression and domination. Bias comes not only from ill intent, but also from a well-practiced operation of explanation. The more practiced explanation becomes, the more readily it skips differentiating anew; the more it has a ready-made set of images, the more readily it pulls reality back into familiar images. Differentiation bias is not a single mistaken perception, but an inertia of explanation formed through repeated distinction-extraction operations.
A temporary lack of information does not necessarily constitute bias. When materials are insufficient, those handling matters can acknowledge what is unknown, investigate records further, and leave room for revision. Bias arises when available or already present differences are consistently downweighted and no longer allowed to affect judgment. Different positions can still be compared in terms of their scope of distinction extraction, their reasons for assigning weights, and their treatment of counterexamples. The danger of differentiation bias is not that someone takes a position, but that the position is fixed as the sole entry point, preventing new materials and relational-field conditions from changing the existing conclusion. Prejudice, insufficient information, and differences of position can all impede judgment, but none can substitute for examining systematic misalignment. Those handling matters need to return the question to concrete operations: have the wrong materials been extracted, has the wrong weight been assigned, or has a judgment applicable in one place been carried into another? Identifying this allows criticism to move beyond attitudes and return to revisable steps of explanation.
Differentiation bias can therefore be examined concretely. Have those handling matters explained what they extracted and what they downweighted? Have they identified the materials and rules underlying the judgment? Have they allowed consequences that differ from the existing image the possibility of changing the conclusion? When these points are missing, even a sincere attitude cannot guarantee that explanation is free of misalignment. Whether judgment can be calibrated also depends on whether those handling matters explicitly leave uncertainties open. Insufficient materials can be investigated further, the application of rules can be checked, and existing images can be limited by counterexamples. These different responses free explanation from a false dichotomy between sincerity and bias. Inertia in distinction extraction has not yet closed explanation completely. New relational fields, supplementary materials, and counterexamples capable of changing consequences may still re-enter. Calibration begins precisely when interpreters acknowledge that entry points can reopen, rather than treating the scope of the first extraction as permanent.
II. Bias in Distinction Extraction: Familiar Differences Monopolize the Entry Point
The first step in differentiation bias is bias in distinction extraction. Bias in distinction extraction does not mean extracting a certain kind of difference, but always extracting only that kind. Interpreters using class theory may grasp real pressures by extracting wages, labor, debt, housing, risk, and capacity for domination. Those using a state image may also grasp real consequences by extracting household registration, documents, archives, eligibility, borders, and implementation. Those using a cultural image may likewise offer an account within a limited scope by extracting language, rites and customs, education, values, and habits of life. The error is not that an explanation sees part of reality, but that in every setting it extracts only familiar differences, making reality conform to an old image. Bias in distinction extraction turns one kind of difference into a preliminary filter for every explanation. Interpreters then first ask whether materials can prove a familiar image, rather than first asking which differences and consequences the current relational field actually involves.
Consider again the family of a worker from elsewhere. When wages go unpaid, extracting wages, labor, debt, and risk has force. But extracting only wages and labor relations when the child seeks school admission will overlook household registration, residence permits, years of social-insurance coverage, the student-registration system, school places, and local policies. Extracting only insufficient income when the wife is refused a rental will overlook local reputation, a nonlocal accent, deposit rules, the landlord's assessment of risk, and relations among neighbors.
Extracting only poverty when the elder needs assistance will overlook where medical records are preserved, how eligibility for assistance is certified, and how the places of household registration and residence provide mutual recognition. Extracting only class circumstances when family historical indices affect borrowing will overlook kinship relations, family memory, the division of the family household in earlier years, and who once cared for the elder. The error of bias in distinction extraction does not lie in seeing part of reality, but in refusing to see other realities anew. Far from ignoring reality entirely, explanation continues to believe it is explaining reality precisely because it has grasped part of it.
The same material can occupy different positions in different problems. Wage records may carry great weight in handling unpaid wages, yet may not take priority in school eligibility or certification for assistance. Whether household registration, medical records, reputation, or exchanges among relatives are extracted depends on current rules and processing positions. Extracting other differences anew does not deny the truth of the earlier materials. What needs adjustment is the order and conditions under which materials enter explanation, so that one kind of real pressure does not repeatedly occupy the entry point and other differences that have already gained weight in the present are not excluded.
Once the scope of distinction extraction is fixed, later materials readily become mere footnotes to the existing image. When the same kind of difference continually occupies the entry point, later materials may be treated as incidental details even when seen. To avoid this, interpreters should restore, one by one, the sources of materials, certification requirements, and relational positions involved in the current problem, rather than substituting familiar categories for fresh identification. Only then can the limits of a partial image be preserved, rather than the image being treated as all of reality. Setting out the distinction-extraction process step by step allows users to see that materials alone do not determine explanation. The steps of extraction, comparison, and weight assignment jointly determine which differences can enter judgments about consequences.
III. Weight Bias: Fixing Temporary Weight in Place
Weight is not a permanent rank carried by a difference itself, but the intensity with which it enters consequences in a concrete relational field. If interpreters omit this condition, they turn a once-effective ranking into a fixed hierarchy across relational fields. Real pressures often make weight bias harder to detect. Precisely because wages, debt, or housing risks do carry great weight, users more readily overlook whether they still occupy key positions in the immediate judgment about eligibility, procedures, or relations. Reassessing weights does not flatten differences; it brings current materials, rules, certification positions, and visible consequences back into judgment together. This preserves the force of explanation within a limited scope while preventing what is temporarily weighty from being written as the highest weight everywhere.
If bias in distinction extraction merely overlooks some differences, it may still be corrected. But once the differences extracted are fixed at the highest weight, the bias deepens. Explanation not only always extracts familiar differences; it also fixes them as the most important ones. Weight comes from concrete relational fields; it is not an eternal underlying reality. In labor disputes, wages, working hours, unpaid wages, and risks of work-related injury carry great weight. In a child's school admission, household registration, residence permits, student registration, years of social-insurance coverage, school places, and local policies may carry more. In relationships, marriage, and elder care, kinship responsibilities, medical history, local reputation, and family historical indices may carry more. Where the boundaries of national communities are concerned, language, shared memory, shared consequences, and paths of responsibility may carry more.
Weight bias occurs if explanation carries the high weight found in an unpaid-wages problem directly into every problem. The family of a worker from elsewhere does indeed face labor and income pressures, but a child's inability to enter school is not necessarily mainly due to low wages. Low wages will worsen the predicament, yet are not necessarily the key node in school eligibility. A rent increase carries great weight, but refusal of a rental may also concern nonlocal identity, deposit rules, local reputation, and the landlord's imaginings about risk. Nor is the effect of the elder's medical records on assistance merely an income problem: it also involves medical records, the places of household registration and residence, the assistance system, and certification nodes. Weight bias mistakes high weight within one relational field for the highest weight in every relational field. This bias is subtle because the differences fixed at the highest weight often do carry great weight. The problem is not that they lack weight, but that they are written as the weightiest everywhere.
Weight bias makes those handling matters mistake one kind of weighty consequence for the sole source of other consequences. A sounder judgment must distinguish which difference is worsening the predicament from which material or rule is determining immediate eligibility. The two may be connected, but cannot therefore be merged into a single permanent ranking. This also requires seeing which positions actually bear the consequences. Income pressures can deepen the predicament, while eligibility restrictions can directly block immediate options. The relationship between them must be explained through materials, rules, and certification procedures. A fixed name cannot be used to erase differences of intensity across relational fields.
Judgments of weight therefore need to explain their grounds as relational fields change. When new documentary requirements, certification results, or relational positions change actual consequences, those handling matters cannot merely restate the original pressures. They must also identify which differences now genuinely occupy important positions in affecting consequences.
IV. Relational-Field Bias: Using a Partial Judgment in the Wrong Field
Relational-field bias calls for clarity about the scope within which an explanation is effective. An interpreter's identification of important differences in a particular problem through an image does not authorize users to carry the same distinction extraction and ranking directly into other problems. Once weights are fixed, explanation carries judgments from one setting into others. Relational-field bias occurs precisely when explanation leaves the setting in which it was originally effective yet still demands that reality conform to it. Use in the wrong field usually occurs not because the original image is devoid of content, but because interpreters fail to check whether the materials, rules, certification nodes, and consequences in the new problem have changed. The original image thus oversteps the bounds of what it could explain. Returning to the relational field does not require starting from zero each time. It requires reconfirming entry points when conditions change: once household registration, student registration, local reputation, or kinship responsibilities have gained weight, explanation cannot simply retain the ranking from the unpaid-wages problem.
In an unpaid-wages problem, limited distinction extraction by class theory may grasp real differentiating units such as labor relations, unpaid wages, the bearing of risks, the width of available exits, and capacity for domination. These differentiating units provide the material basis; the processes of extraction, weight assignment, connection, and compression form partial explanatory power. The class image is a low-dimensional output and cannot directly explain every problem. In a child's school admission, images of institutionalization and state incorporation of differences are more effective: the key may be household registration, residence permits, social insurance, student registration, school places, local policies, and certification chains. In a problem of local recognition, an image of local networks is more effective: the key may be accent, years of residence, exchanges among neighbors, ties with people from the same place of origin, participation in weddings and funerals, and local reputation.
In relationships and marriage, family nodes and historical indices are more effective. The key may be medical history, kinship responsibilities, family historical indices, local reputation, who has owed something to whom, and who has cared for the elder. If explanation uses the image from the unpaid-wages problem to explain school admission, local recognition, relationships and marriage, and migration memory, it leaves the relational field in which it was originally effective. Use in the wrong field is an important way in which explanations effective within a limited scope move toward an overall misjudgment. Many erroneous explanations are not wholly ineffective: they are effective in their original settings, but overstep the bounds of their effectiveness. A relational field is not an abstract background. It is a concrete state of relations that has already caused certain differences to be extracted and to enter consequences. Only by checking whether the problem being handled concerns unpaid wages, school admission, renting, relationships and marriage, assistance, or local recognition can interpreters judge whether the original image remains in a position where it is effective.
An explanation effective within a limited scope can continue to be used only while its materials and consequences remain supported by the current relational field. Users wishing to retain the original image should explain which segment of relations it can still explain and which new materials require supplementation through other distinction extraction, rather than demanding that reality remain forever in the original setting. Correcting use in the wrong field does not replace one total image with another. It confines the original explanation to the relations it can still explain. The remainder must be identified anew through the materials and procedures of the new relational field, preventing limited force from being mistaken for universal authority.
V. Node Bias: Concrete Processing Positions Are Occluded
Node bias readily portrays differences as abstract forces that directly produce consequences. Once explanation is used in the wrong field, it becomes still easier to overlook the nodes that actually make differences take effect. Node bias means that explanation sees differences but not where they are preserved, who extracts them, who certifies them, or through which nodes they enter consequences. In reality, differences enter eligibility, evaluation, and action only through preservation, extraction, certification, passing-on, or implementation at concrete positions. Records, documents, medical records, judgments, and reputation are not subjects acting on their own. Consequences arise only when school personnel, medical personnel, employers, family elders, landlords, intermediaries, reviewers, or implementers use them within existing rules and procedures.
Nodes are not merely places where differences are preserved. They also determine who can extract differences, who can certify them, and who can make them produce consequences. Node bias also includes recognizing only materials that certain nodes can process while failing to see differences preserved in other relations. The problems of the family of a worker from elsewhere cannot be seen only as “low income.” If explanation says only “these are the circumstances of the bottom stratum,” it will see a grand image but not how differences are preserved, extracted, certified, and implemented through nodes.
The household-registration system can bring residential differences into consequences for school admission. When the child of a worker from elsewhere seeks school admission, the household-registration system, residence permits, social-insurance records, the student-registration system, and school places are all nodes. Yet under local admission rules, the school may review only some of the materials concerning household registration, residence, social insurance, or student registration. Ruptures in the family's migration history, care responsibilities, and actual residential difficulties may not be directly convertible into formal requirements.
Hospital medical records can bring bodily differences into consequences for employment, relationships and marriage, and assistance. When the elder in the family of a worker from elsewhere seeks medical care, hospital medical records, medical-insurance relations, assistance at the place of household registration, and proof of residence are all nodes. Hospital medical records preserve bodily differences; employers must still handle leave, capacity for work, or job-position issues according to their own rules. A corresponding general institutional node is a court judgment, which can bring past events into eligibility reviews and social evaluation.
Local reputation can bring migration and reputation into local recognition, while family elders can bring historical indices into consequences for relationships and marriage, household division, and elder care. When the family of a worker from elsewhere is refused a rental, landlords, neighborhood reputation, intermediaries' records, and deposit rules are all nodes. When relationships and marriage meet obstacles, family elders, local reputation, kinship historical indices, medical records, and exchanges of ceremonial cash gifts are all nodes. These family historical indices and local reputations belong to informal relations and do not automatically enter state archives.
Different nodes may preserve or recognize different materials, and those materials may conflict. Formal institutions do not necessarily preserve family responsibilities, nor does local recognition necessarily amount to formal eligibility. If explanation recognizes only one kind of node, it downweights reality at other nodes. To miss the nodes is to miss how reality takes effect. Making these branches clear can prevent a grand image from occluding concrete responsibilities and positions open to revision.
Examining nodes serves precisely to identify the distribution of concrete responsibility and the positions open to revision; node bias occludes that very distribution. Who registers materials, when they are retrieved, under which rules they are recognized, and who makes the determination: these differences determine which positions can be examined and corrected. If a result comes from documentary requirements, medical-record certification, student-registration records, or local reputation, those handling matters can distinguish whether missing materials, rule restrictions, certification methods, or acts of use brought differences into consequences. They need not compress every result into abstract circumstances. Speaking only of abstract circumstances omits the procedures and actors in actual handling that could be questioned.
VI. Relational-Chain Bias: Paths of Consequences Are Cut Off
Relational-chain bias does not merely mean failing to list a few conditions or consider a few factors. It cuts already connected paths of consequences into unrelated fragments, failing to see how differences connect with one another and intensify, trigger, and transmit consequences. If explanation stops at income, identity, or medical history alone, it cannot see how consequences are subsequently intensified or recast through other positions. Failing to see nodes leads further to failing to see how differences transmit consequences. Low income by itself does not amount to a fixed set of circumstances. Only when connected to rent, debt, medical history, care responsibilities, household registration, children's school admission, and the absence of fallback options does it form weighty circumstances. Nor does nonlocal identity by itself amount to a fixed set of circumstances. Only when connected to accent, household registration, renting, social insurance, school eligibility, local relations, and work risks does it form persistent exclusion.
Within the same family, when wages go unpaid, rent becomes harder to pay; when rent becomes harder to pay, moving may become more frequent. Frequent moves make proof of residence and school eligibility less stable, and the child's admission may consequently be delayed. Care pressures return to the mother, who reduces her working hours, and family income continues to fall. When the elder's medical history entails a need for money, kinship historical indices are raised anew, and borrowing becomes harder. The tighter the debt deadline, the more likely the husband is to accept dangerous work. This is not “many factors”; differences are transmitting consequences. Relational-chain bias does not simply miss one factor. Explanation sees a difference but fails to trace how it connects with other differences to form consequences, thus cutting off the paths of consequences between differences. When tracing relational chains, operations must be attributed to concrete positions: who fails to pay wages, who demands proof, who certifies according to rules, who adjusts working hours, and who undertakes care within the family? Only then can we explain how consequences are transmitted, rather than letting the relational chain move on its own.
Nor is a relational chain a single line of causation. The same material may be amplified, attenuated, or temporarily halted at different nodes; calibration requires looking back at actual paths and branches. When consequences travel along a relational chain, a change in an earlier link does not necessarily determine the final result directly, but may change the materials and options available for processing at later nodes. Each position may alter the direction and intensity of consequences. Once a piece of material has been supplemented, certification may succeed; applying a rule may also restrict a previously viable option. Explaining these relays and retaining intermediate positions does not add scattered details. It prevents a complex path from being misrepresented as a set of consequences caused entirely by a single difference acting on its own, and avoids falsely attributing consequences to a single image.
Examining relational chains also requires distinguishing consequences that have already occurred from conditions that may still change. Interpreters cannot portray a link as the sole cause of every subsequent result merely because it once worsened a predicament. They should see how materials are passed on, how rules are applied, and which positions can still change the path. Such tracing preserves the intermediate conditions in paths of consequences and prevents a single image from replacing concrete transmission.
VII. Historical-Index Bias: Treating the Past as Automatic or Excluding It
After relational chains are cut off, explanation also misreads how past differences enter the present. Historical-index bias is not an ordinary problem of historical memory. It involves two common errors. The first is historical determinism, which treats the past as a fixed cause, as though old harms, old boundaries, old victories, and old punishments automatically determined the present. The second is the thesis of historical irrelevance, which treats the past as irrelevant background, as though present interests, institutions, or attitudes alone could explain reality. Both views are wrong. Historical-index bias is precisely the failure to see which current events are triggering past differences. Once preserved, past differences do not automatically govern the present. They must be found anew, triggered anew, and assigned weight anew through indices in new relational fields.
Past differences do not return to reality on their own. They are extracted anew by triggering conditions in the current relational field. Excluding the past altogether also causes misalignment. Family members, members of local communities, archive custodians, or institutional processing positions may extract old records, past exchanges, and shared memories anew in current procedures, allowing them once again to change judgments and consequences in concrete events such as borrowing, school admission, relationships and marriage, elder care, eligibility applications, or local conflicts. National memory, local memory, and family history all need to re-enter through concrete relays, materials, and current conditions. Detached from these conditions of interaction, preservation, and triggering, historical communities are mistaken for essences of descent, fictional labels, or automatically continuing forces.
Family historical indices may be triggered by borrowing, relationships and marriage, elder care, or household division. Local memory may be triggered by grave relocation, land disputes, school catchment zoning, or local conflicts. National memory may be triggered by commemorative dates, external pressures, boundary conflicts, educational disputes, or shared protection. State records may be triggered by pre-employment screening, eligibility applications, loans, travel, or border checks.
The migration route of the family of a worker from elsewhere need not matter every day. But when the child's school admission, renting, relationships and marriage, borrowing, or local conflicts arise, it may suddenly gain weight. Where they came from, why they came, how long they have lived locally, whether they have participated in weddings and funerals, whether they have had past exchanges with local people, and whether they have been recognized through place names, accent, or ties with people from the same place of origin will all change present judgments. Nor do family historical indices always carry the same weight: they may be extracted anew in borrowing, household division, elder care, relationships and marriage, and caring for an elder. Local reputation is not an abstract background either. It may be brought back into the current relational field through a neighbor's remark, a landlord's judgment, a school's certificate, or a relative's reminder.
Historical indices need to be identified anew within current relational fields precisely because the same past occurrence may not carry the same weight in different events. The re-entry of historical indices also has boundaries. Even if an episode from the past has been preserved, we must still ask whether it is extracted in the current event, by whom, and on the basis of which materials it is recognized. This calibration prevents history from being written as automatic fate and present handling from being written as wholly without past conditions. Interpreters must neither treat old records as natural destiny nor refuse to consult them because they belong to the past. They should identify whether a particular person, procedure, or relation is bringing them back into judgment now. Those handling matters must explain what trigger makes old materials relevant again and examine whether they actually enter present consequences through visible nodes. This both reveals the real role of shared memory, responsibility, and local relations and prevents the past from being treated as an untestable abstract cause.
VIII. How Misalignment Forms and Persists
The six forms of bias are not six isolated errors alongside one another. They form a continuous process in which explanation moves step by step from repeated distinction extraction toward misalignment with reality. Repeated distinction extraction first narrows the entry point; fixed weights then entrench the existing ranking; use in the wrong field and the omission of nodes, relational chains, and historical indices gradually stabilize this misalignment. The process unfolds as follows. First comes bias in distinction extraction: explanation always extracts the same kind of difference. Then comes weight bias: that kind of difference is fixed at the highest weight. Next comes relational-field bias: explanation carries that weight into an unsuitable relational field. This is followed by node bias: explanation fails to see the nodes where differences are preserved, certified, and implemented. After that comes relational-chain bias: explanation fails to see how differences connect to form consequences. Finally comes historical-index bias: explanation misreads how past differences are found anew, triggered, and assigned weight in the present. Once this chain has run its full course, explanation moves ever further from concrete objects. At this point it is not entirely without materials; it recognizes only familiar materials. It is not unable to see reality; it sees only the reality it is already prepared to extract. This is the state of differentiation bias once stabilized.
Restoring the six forms of bias to a continuous process does not mean that every problem neatly passes through the same six steps in the same way, nor should the process be arranged as a set of mechanical steps. In different problems, this chain can display different speeds and combinations. Some problems first reveal the omission of nodes, while others first reveal use in the wrong field. An omission at an earlier link facilitates the next: after the entry point narrows, fixed rankings are more readily accepted; after rankings are fixed, use in the wrong field and omitted nodes become harder to see. This mutual reinforcement helps stabilize misalignment, but does not eliminate the concrete positions that must be examined for each form of bias. Whether causal paths have been cut off must be examined in relational chains; how the past enters the present, in historical indices; and who preserves, extracts, certifies, and implements differences, in nodes.
Each link may leave an entry point for calibration: materials can be restored, rankings adjusted, procedures examined, and consequences questioned. Continuity does not mean irreversibility. As long as explanation still allows new materials to change entry points, the chain of misalignment may be interrupted. Interpreters therefore need to ask concrete questions at every point of misalignment: why did the entry point narrow, why was the ranking fixed, in which relational field was explanation wrongly applied, which nodes and relations went untraced, and how was the past misread?
Differentiation bias persists not only because errors go undetected, but also because it can save processing costs in concrete use. See a family and say “bottom stratum”; see conflict and say “class”; see a community and say “backward culture”; see the state and say “machine of oppression”; see market consequences and say “free competition”—and interpreters need not face complex materials anew each time. Through familiar images, interpreters, disseminators, and mobilizers can more quickly supply names, causes, and positions, yet may thereby skip identifying the concrete relational field anew. Labels circulate and spread more readily than complete circumstances. Terms such as “workers in the bottom stratum,” “people from elsewhere,” and “objects of management” are very short, whereas the wages, household registration, medical history, local reputation, school eligibility, kinship historical indices, and migration memory of the family of a worker from elsewhere cannot easily be fully explained in a single sentence.
Mobilization likewise needs clear causes, objects, and courses of action. Compressed accounts can prompt people to take sides quickly, act, and find enemies or victims. During dissemination, short labels are readily extracted again by later users; the clear designation of objects in mobilization also readily leads to the downweighting of materials that are difficult to incorporate into the existing image. We should recognize that these are practical acts by concrete users, not active powers of choice and action belonging to labels or bias themselves.
Interpreters may also use this effort-saving mechanism to shield existing theories from pressure for revision. Overlooked nodes, relational chains, and historical indices are downgraded to secondary, superficial, or special circumstances, or blamed on insufficient materials, so that the original explanatory framework need not be reopened. Users trade on effectiveness within a limited scope to keep saving effort, and this inertia in distinction extraction is readily mistaken for explanatory power. Saving effort does not necessarily lead to bias. But when it becomes a reason to stop examining materials, the existing image grows more firmly established through dissemination and use. Once an effort-saving explanation is adopted repeatedly, it may leave clues in records, dissemination, and action that facilitate repeated application. If later users simply follow these clues and carry on using it, they will treat the existing image as a completed judgment.
Repeated adoption of an existing explanation may also bias subsequent material collection itself toward familiar clues. If recorders, disseminators, and users of theory preserve or recount only what fits the image, it becomes harder for other differences to enter the next judgment. Materials have not vanished into thin air on their own. Concrete positions of use repeatedly determine which materials can be seen, cited, and assigned weight. Persistent bias also leaves traces that can be examined: certain materials always receive priority in citation, certain problems are always called exceptions, and certain nodes are never questioned. Examining which materials are repeatedly omitted makes effort-saving inertia visible again. Asking exactly who cites, who adopts, and who actually implements brings abstract persistence back to repeated actions that can be examined. These traces allow habits of saving effort to be turned back into judgments open to comparison and revision.
IX. Counterexample Exclusion and the Turn Toward Illusory Explanatory Construction
Counterexample exclusion still retains an important boundary: counterexamples have not yet been rewritten as proof of the existing theory; they have merely been pushed outside explanation. As long as materials can re-enter, bias may still be open to calibration. If interpreters, users of theory, or those handling matters dismiss concrete institutional conditions, local memory, or community responsibilities wholesale as exceptions, they are in effect preventing counterexamples from changing existing distinction extraction and weights. It is these concrete users who exclude them, not the counterexamples themselves that have lost their effect.
Differentiation bias is not yet illusory explanatory construction; it may still be calibrated. As long as explanation is willing to return to concrete relational fields, reassess weights, examine nodes, trace relational chains, and reread historical indices, counterexamples can still change it. If it refuses to do so, however, it begins to exclude counterexamples. At the stage of differentiation bias, explanation first pushes counterexamples away on encountering them, calling them chance occurrences, exceptions, noise, special circumstances, or insufficient materials.
A family is plainly barred from school admission by household registration and the number of school places, yet explanation still says these are merely isolated institutional details and that the essence of the matter is still the family's labor circumstances. A local conflict is plainly triggered by family historical indices, place names, burial grounds, and local reputation, yet explanation still says these are merely old ideas and that the essence of the matter is still economic interests. A national community is plainly sustained through language, memory, shared protection, and paths of responsibility, yet explanation still says these are merely manufactured surface identities. Once counterexamples are pushed away, explanation need not revise itself. In counterexample exclusion, counterexamples have not yet been rewritten as materials supporting the explanation, so entry points for calibration may still reopen. In counterexample digestion, however, they are incorporated into the explanation's own structure, made part of the theory's self-verification, and presented as proof that the theory is correct at a deeper level, bringing explanation into a more closed state. If differentiation bias refuses to let counterexamples change it, it slides toward illusory explanatory construction.
Counterexamples do not serve to force explanation to abandon every judgment. They test whether existing entry points can acknowledge having missed differences that are gaining weight. When counterexamples are allowed to change the ranking of materials, the relevance of rules, or the understanding of relational fields, explanation can remain open. If they are merely kept outside, misalignment deepens. When excluding counterexamples, interpreters may still claim to be preserving only the main judgment. Yet if every nonconforming piece of material is called accidental or secondary, the main judgment ceases to be tested against relational fields and consequences. Only by retaining a place for counterexamples in calibration can effectiveness within a limited scope be kept from turning into a closed conclusion.
For a counterexample truly to enter explanation means that it can change the scope, ranking, or entry points of the original judgment. This does not abandon theory. On the contrary, it keeps explanation open to comparison and revision and maintains its tracing of real consequences. Calibrating differentiation bias does not require explanation to abandon judgment or to treat all differences equally. Without judgment there is no theory; without weights, explanation cannot act. The problem is not that explanation has extracted certain differences, but that it has forgotten it also downweighted others. Calibration therefore restores explanation's capacity to differentiate and to judge anew which differences are gaining weight in the current relational field.
Facing the relational field anew requires establishing which differences enter consequences through which materials, rules, and nodes. Interpreters or users must return to the materials themselves and ask, item by item: do they always extract the same kind of difference, do they fix that kind at the highest weight, and has this explanation been brought into the wrong relational field? Returning to the relational field and reassessing weights bring explanation back to concrete situations, examining the operative weight of different conditions in particular contexts.
Further examination must reach into the mechanisms of transmission. It must ask where differences are preserved, who extracts, certifies, and implements them, how they connect along relational chains and are transmitted into consequences, and which historical indices are currently being triggered. Examining nodes, tracing relational chains, and rereading historical indices serve precisely to establish how concrete differences connect through each link to form consequences, rooting explanation in materials again.
Explanation must also confront the question of whether counterexamples can change it rather than simply being pushed away. These questions are raised not to cancel explanation, but to let counterexamples truly enter it so that it can regain the capacity for calibration. If explanation can still be changed by these questions and counterexamples, it has not yet closed. If it refuses these questions, it will keep sliding along the inertia of distinction extraction. Only when it no longer merely excludes counterexamples but also incorporates them into its own structure does it enter the more closed state of illusory explanatory construction.
Calibration also requires distinguishing revision from self-negation. When interpreters adjust distinction extraction and weights, they are not declaring every previous judgment invalid. They are redefining the scope of applicability in light of new relational fields, materials, and consequences, so that theory can both make judgments and accept testing by counterexamples and concrete conditions. The practical meaning of calibration is to re-establish a return channel: materials can be supplied, weights reassessed, nodes and relational chains re-examined, and historical indices identified anew according to current conditions. These steps do not guarantee a complete explanation at the first attempt, but they can prevent existing images from closing off the very possibility of revision.
Calibration also requires returning the results of revision to testing against consequences. When new materials, rules, or counterexamples change the entry points of explanation, those handling matters should explain anew which judgments remain valid, which rankings need adjustment, and which questions still need pursuing. Such revision preserves responsibility for judgment while preventing existing images from escaping concrete relational fields merely because they are familiar.