English integrated edition · v003
Chapter 33. Platform Society: Differentiation Networks of Accounts, Records, Ranking, and Visibility
A platform is easily portrayed as a tool, as though people used it to find work, buy things, chat, rent housing, book medical appointments, submit homework, or publish content, and could leave once they were done. Another account treats it as a virtual space, as though a screen separated what happens there from real life. The distinction between online and offline cannot adequately explain platform society. Both understandings omit platforms’ positions within social relations. Personnel responsible for rulemaking, operations, review, customer service, and technical maintenance within companies use accounts, record systems, ranking rules, and review procedures to preserve materials, arrange opportunities, distribute visibility, and form evaluations. Labor relations, trust, and pathways of punishment also change as a result.
A platform node, as used here, refers to the relational positions of company personnel, rules, procedures, accounts, records, and users within an event. A platform has no unified personality and is neither an autonomous underlying reality nor the social totality. It can produce real consequences, but it does not think and act on its own. As people use platforms, account and record systems maintained by the relevant personnel also preserve records of behavior. Ranking rules, review procedures, and evaluators in the current event form account evaluations, arrange opportunities, limit visibility, and carry records into the future.
In concrete events, accounts, real-name information, records, scores, labels, ranking, recommendations, visibility, bans, complaints, orders, traffic, rules, permissions, social relations, transaction evaluations, and penalty pathways connect with one another to form differentiation networks. These items are only common entry points, not a complete list. Differentiating units within platforms remain potentially infinite.
People using the same platform service do not thereby enter identical relations. Workers are concerned with whether they can obtain orders and income; merchants, with whether their goods can be seen; parents, with whether notices and materials have arrived; and content publishers, with review and circulation. A user may simultaneously be a buyer, an evaluator, someone being evaluated, and an appellant. The materials encountered, permissions held, and consequences borne differ at each position. The term ‘platform user’ cannot flatten these differences.
Different positions also exist within platform companies. Those who make rules do not necessarily handle individual appeals. Customer-service personnel can receive explanations but may lack authority to amend records. Review personnel can make a determination about a piece of content or a transaction, while technical-maintenance personnel are responsible for whether the system operates according to established arrangements. An analysis of a concrete event must identify which personnel or procedures actually extract materials, form judgments, and put consequences into effect. It cannot assign every action to an abstract ‘platform.’
Let us continue with that family of workers from elsewhere. The family always lives within real relations. Platforms enter their lives through events involving job seeking, renting, education, medical care, kin, and debt; they do not create another world isolated from reality. The father works at construction sites and may find work through chat groups of people from the same place of origin, recruitment platforms, recruitment information in short videos, or construction-site groups. Information posted by recruiters is presented after ranking, affecting which site he goes to and whose claims he believes; the related chats also remain in accounts and record systems. The mother does casual work at a restaurant and may also take orders through delivery, domestic-service, or hourly-work platforms. When she takes orders, ratings, response speed, and completion records change her next opportunity. What actually needs tracing is how company personnel use accounts, records, ranking, and evaluation procedures to preserve real differences anew and then bring the processing results back into life.
Housing and education events likewise pass through different entry points. Landlords post rental listings through rental platforms and consult tenants’ credit, chat records, payment methods, and past evaluations. A child’s schooling involves home–school chat groups, application systems, homework platforms, notification software, and online materials. Whether parents can see group messages in time and successfully upload materials may enter teachers’ judgments of the family’s cooperation.
In medical care and debts to relatives, the elder’s treatment involves appointment platforms, medical-insurance settlement, medical-record uploads, and online-consultation records, and is affected by appointment timing, real-name information, medical-record uploads, and payment records. When relatives press for repayment, chat software, money-transfer records, and family chat groups also enter the event. The money-transfer records, chat content, and silence in the group that remain may later be extracted anew by family members.
The same family may also move materials among entry points on several platforms. Phone numbers, real-name information, payment methods, and addresses are used to establish different accounts, while work messages, school notices, appointment results, and money-transfer receipts are scattered across different procedures. When one entry point fails, the effect may not stop at a single service. Restricted payment may obstruct order-taking; a changed phone number may interrupt notifications; and a shared device may mix together records of different family members. These connections must be traced one by one, rather than attributed directly to the family’s ‘not knowing how to use platforms.’
Family members differ in their command of platforms. Parents may be familiar with taking orders or chatting but unfamiliar with submitting school materials. A child who can operate a device may not have authority to handle real-name, payment, or medical information. An elder relies on family help to book an appointment, yet the account’s operation record may show only one success or failure. Operating ability should not be treated as a fixed individual attribute. How care, devices, time, and permissions jointly change results must also enter the analysis.
An account is a basic entry point into platform society, with a role far beyond that of an ordinary name. It carries history, permissions, credit, relations, transactions, evaluations, labels, and penalty records. Taking orders, speaking, buying, selling, livestreaming, commenting, appealing, entering groups, or obtaining recommendations often depends on account status. The relevant personnel and procedures also identify users primarily through accounts. Real people are thus represented as real-name information, phone numbers, payment methods, past records, scores, devices, locations, group relations, and violation records. These data can present only part of a person’s circumstances.
Many platform difficulties arise at the point of translation: a person’s real circumstances cannot readily be turned into an account-based image of the person that company personnel and established procedures can process. Within this process, accounts, records, ranking, and visibility each perform a distinct role. Accounts carry entry eligibility and permissions; records preserve materials; ranking arranges the order of presentation within an event; and visibility indicates whether users can actually see or be seen. The four are related, but their distinctions should still be maintained.
The formation of an account-based image also involves selection. Which behaviors are recorded, which differences are placed under the same label, when activity is treated as continuous, and what anomalies require review all depend on classifications and processing arrangements established and maintained by company personnel. Facts of life that the system has not preserved may be difficult to bring into an appeal. Preserved fragments, meanwhile, may be detached from their circumstances at the time and reinterpreted in another event.
The relevant procedures encounter precisely these account statuses, records, scores, and labels, not a person’s whole life. This way of processing readily produces misalignment. A person reliable in real life may fail to gain the trust of trading partners or review procedures because the account is too new and has too few ratings and records. Someone has in fact encountered difficulties, yet the record retains only lateness, slow responses, or canceled orders. After processing, a malicious complaint may leave only the label ‘complaint upheld.’ A change of device, phone number, or city may also be marked as anomalous by established identification rules, and risk procedures may even judge normal behavior suspicious. All these examples show a distance between the real person and the account-based image.
The formation of records may itself be disputed. Behind cases in which the record ‘retains only lateness, slow responses, or canceled orders,’ an order may have been canceled voluntarily by the worker or in connection with the other party changing requirements, a device failure, or an emergency in real life. An unread message may not have been opened, or its alert may not have arrived; the account may have been switched, or the device shared by family members. If the final record retains only ‘canceled’ and ‘unread,’ the reasons for their formation are compressed. Later users retrieving the record must explain what it can prove and what it cannot.
One person may also occupy several account positions at once. Work accounts, shared family accounts, purchase accounts, and accounts for public speech carry different relations, while the same phone or payment method may connect them. Whether accounts are identified as belonging to the same person, whether their histories can be transferred, and whether restrictions are shared all change permissions and risks. Analysis must still distinguish the real person, identity materials that can be submitted, and the account-based image established by procedures. Identity information cannot replace the person.
Account and record systems preserve records of large amounts of past behavior in ways set by company personnel. A purchase, a negative review, a complaint, an instance of lateness, a statement, a violation, a ban, or a failed appeal may all be extracted anew in the future.
Such records can appear repeatedly across times, settings, and relational chains. Past behavior continues to participate in later events in forms such as labels, scores, recommendations, and restrictions. Chat records left by the father in a recruitment group may become evidence in a claim for unpaid wages, or be forwarded by others and become evaluation materials. A negative review received by the mother on an order-taking platform may affect the following week’s orders. A parent’s slower reply in a group may leave an impression on a teacher. A failed appointment booking for the elder may force the family to rearrange care and time.
Records become candidate materials for future judgments and therefore cannot be treated merely as neutral storage space. Historical indices do not produce power on their own. Only when operations, review, customer-service, or technical-maintenance personnel, or users, retrieve records in a current event, interpret them for concrete purposes, and assign them weight does the past enter the present.
Different records may also conflict. Transaction-completion records, chat content, proof of payment, location data, and the two parties’ evaluations may present different courses of events. Supplementary explanations from customer service and original labels may not be updated in step. Judgment cannot rely only on the most easily retrieved material. Current personnel should compare when records were formed, their sources, applicable scopes, and amendment statuses, while keeping entry points open for contrary materials to enter processing.
When two workers taking orders have accounts at different positions in the rankings, the opportunities they face differ. When one request for help is widely reposted and another is seen by almost no one, the two are also unlikely to produce the same consequences. Record systems preserve materials, while ranking rules arrange opportunities within the current event: recommendations and searches affect which accounts are readily seen; the order in which goods are presented, whether content is hidden, and which evaluations appear first also produce different distributions of opportunity. Visibility has real effects on income, reputation, relations, trust, and action space. It is anything but an empty name.
Content that is not visible still exists; it simply has difficulty entering chains of opportunity. Content has been posted, yet no one sees it; an account still exists, yet cannot be found in a search; a product has been listed, yet appears far down the list. Workers remain online but still receive no orders. If requests for help are pushed into positions where responding is difficult, they too are unlikely to produce consequences. In home–school groups, a parent's silence may also be interpreted as unwillingness to cooperate. As visibility distributes opportunities, it leaves many accounts, pieces of content, goods, forms of labor, and requests for help at the margins. These circumstances do not necessarily all result from penalties.
Power does indeed often operate through ranking, downweighting, restrictions on reach, hiding, bans, and the rejection of appeals, and it need not be accompanied by public commands. Once an account has been downweighted, opportunities have already diminished even if no one has explicitly stated what is prohibited. When content is hidden, that alone is enough to obstruct its effects even without public refutation. When an appeal is rejected, the reasons may remain unclear even though the penalty has already been entered into the record.
The effects of a ranking position must also be considered within the concrete event. Ranking higher does not necessarily bring the same kind of benefit: goods being seen may increase transactions; an appeal for help being seen may bring support; and personal content being widely seen may also increase misreading, attacks, or reports. Being farther down usually reduces opportunities for contact, yet not every consequence can be inferred from position alone. The weight of visibility depends on who can see, when they see, what action they can take afterward, and whether subsequent evaluations are in turn preserved.
Actions among users also enter ranking chains. Clicks, time spent viewing, evaluations, reposts, blocking, and reports can become candidate materials, from which ranking arrangements set by company personnel then extract some differences. These behaviors provide only limited signals and cannot fully determine the value of content; nor do ranking rules discover what is true on their own. The formation of signals always involves concrete users, record systems, and rule arrangements.
Differences in accounts and records do not turn into consequences on their own. Personnel or procedures must extract, evaluate, rank, or restrict them. Personnel and procedures that can extract differences and turn them into scores, labels, recommendations, restrictions, or penalties can also affect users’ positions within platform relations. Who determines whether an account is reliable? Each of the following has actual handlers: whether a delivery rider or domestic worker can be recommended, where a merchant ranks, whether content is visible, whether a complaint is upheld, whether an appeal succeeds, and whether past records affect the next opportunity. These processes involve evaluation authority and ranking authority within platform networks, far more than mere opinions.
Rulemakers, customer-service staff, and review personnel within platform companies, together with merchants, evaluators, group administrators, and complainants, may all hold differing degrees of authority to extract differences and evaluation authority. Recommendation, payment, and credit procedures also carry established processing authority. Platform power appears not only in account ownership, but also in record preservation, rule setting, ranking for visibility, credit evaluation, hiding content, banning accounts, and handling appeals.
Different evaluators have different materials available. Trading partners may see only a page, price, and past ratings. Customer-service personnel can see appeal materials and processing records. Review personnel judge content or behavior under current rules, while group administrators may impose restrictions based on interactions within the group. If a judgment formed at one position is extended directly to other positions, limited materials may be expanded into an overall evaluation of the entire account.
Authority to extract differences also determines which materials remain outside judgment. When procedures accept only established formats, circumstances that cannot be turned into images, text, timestamps, or specified options may fail to enter processing. If personnel examine only recent records, long-term reliable conduct is also easily overlooked. Analysis need not add all materials of life without limit, but it should explain how the current window was formed and examine whether excluded materials could change the conclusion.
Platform rules change weights and cannot be treated as a stable background. Recommendation methods, review standards, appeal entry points, the scale of penalties, and merchant rules may all be adjusted, changing accounts’ weights within the relevant events. After a rule update, previously normal speech, order-taking, transactions, and interactions may face new ranking, evaluation, or restrictions. Previously effective practices may cease to work, and long-silent records may receive new interpretations. A once-unimportant label may even begin to affect recommendations, searches, and order-taking. The same account, the same set of records, and the same behavior may thus produce consequences different from those in the past.
For those who rely on platforms for orders, a rule update directly affects income; it is no distant technical adjustment. Families relying on home–school platforms for notices are also affected by changes in entry points, alerts, and submission methods. After rulemaking and operations personnel complete an update, review procedures and ranking rules may give old records weight anew or interrupt actions that could otherwise have continued.
Rule changes also produce gaps in understanding. Company personnel may already have changed entry points, standards, or classifications while users still act on past experience. Customer-service explanations, page prompts, and actual processing may also fall out of step. Records of mistaken selections or delays left in this way cannot immediately be understood as users’ refusal to comply. Analysis must trace when rules changed, how notices were delivered, who had time to adjust, and whether supplementation and correction are allowed in dealing with penalties.
Rules also have scopes of application. A restriction may concern only a certain kind of transaction, content, or account status and should not automatically be extended into a judgment on all actions. An update may change recommendations without necessarily changing payment or appeals. Treating platform rules as a single whole obscures differences among departments, personnel, and procedures, and makes responsibility difficult to locate.
Platforms are not closed online systems. The key to platform society lies in how real differences gain new representations through accounts, records, and ranking procedures maintained by company personnel, and are then preserved, ranked, and written back. Online life has not replaced offline life; the two have long been connected through concrete events.
Care and the body offer an initial view of how life circumstances are compressed. In everyday life, care responsibilities reduce the time available online, and bodily condition may be reflected in completion rates, lateness rates, and the risk of complaints. Once they enter a platform, care responsibilities may leave only a record of time offline, and physical fatigue only a record of completion speed. When picking up her child prevents the mother from staying online for long periods, the record system may retain only slow responses, few orders taken, and low activity.
Spatial and material conditions are also recast: geographical location may become delivery distance, nearby recommendations, and local rankings; device condition affects response speed and content quality; and network stability is reflected in online status. Once movement between places enters a platform, it may leave only changes in location. After the father moves to another city to find work, his location, device, and account behavior all change, and the relevant rules may generate markers on this basis. Mandarin, dialects, and modes of expression bring another set of life differences into procedures: they may enter customer-service evaluations, livestream viewing, and content circulation; once language differences enter a platform, they may leave only the form of replies. In another situation, a parent working the night shift misses messages in a home–school group, while the teacher or school review personnel may see only a failure to respond in time.
Real-name information also bears on account certification and payment permissions; market credit bears on eligibility to transact; and local acquaintance groups carry information on jobs and rental housing, along with reputations within the groups. The differences that procedures can handle are fewer and more readily compared, but a compressed marker cannot replace the full circumstances that produced it.
Real differences may also be recombined on platforms. Location, device, payment method, and time of operation may each look ordinary in isolation, yet form a risk marker when jointly extracted under the same rule. When scores, completion rates, and response speed enter ranking together, they may also change a person's position in taking orders. Such a recombination does not create a person's essence. It only means that procedures set by company personnel represent and assign weight to differences in a particular way within a specific event; personnel and procedures in the current event then carry the results back into reality.
Different platforms turn different business processes into calculable time. Time in everyday life does not enter systems automatically. Only after rulemaking and operations personnel set time limits is time in everyday life within the corresponding activity converted into response periods, completion periods, popularity periods, appeal periods, penalty periods, or record periods. Depending on the specific activity and published rules, the relevant times may include some of those associated with registration, responses, fulfillment, refunds, appeals, handling, or record retention. Recommendation cycles, popularity windows, read-status timestamps, money-transfer times, and deletion times do not necessarily all coexist on the same platform.
Once established, these times are readily recorded and compared precisely, and may even become grounds for penalties. For example, lateness on an ordinary order may have different consequences from lateness on an order during peak hours. Whether a message is answered immediately or several hours later may also change trust. Once the appeal deadline for a complaint has passed, the relevant position may also become temporarily fixed through procedural handling.
Record systems and ranking rules turn people’s busyness, fatigue, care responsibilities, network delays, and real difficulties into processable markers such as slow response, poor completion, missed deadlines, or untrustworthiness. A late reply, a missed appeal period, or lateness at a peak time may all be transformed within the current event into differences in credit, scores, recommendations, and penalties.
Platform time also distributes waiting unevenly. Some people can wait for a review, refund, or appeal; others depend on that day’s income, and every day of account restrictions adds another layer of pressure on their lives. Some can submit supplementary materials repeatedly, while others can deal with the matter only once in a break from work. The same processing time need not produce the same weight. Judging this difference also requires returning income, care, device use, and qualification deadlines together to the current relational field.
Under rules that certain platforms have published and that apply to the corresponding activities, measures taken against accounts may lower visibility, restrict some functions, suspend services, or terminate eligibility. Platforms do not necessarily all employ these measures together. Downweighting, restrictions on reach, point deductions, and hiding, as well as restrictions on withdrawals, taking orders, speaking, or group activities, each have their corresponding scope of application. The specific measures, targets, periods, and means of appeal still vary with the platform, activity, and version of the rules. A complaint may trigger point deductions, restrictions on reach, an account ban, a refund, compensation, or cancellation of eligibility, but this does not establish that every complaint passes through the same set of measures.
When review and operations personnel impose a penalty through procedures and then preserve the result of handling it as a difference that can be extracted again later, this constitutes a form of write-back. Whether penalty records can actually be retrieved in subsequent applications, appeals, recommendations, and transactions must still be judged according to the rules, authority, and subsequent handling of the corresponding activity. A successful appeal can restore some opportunities but does not necessarily guarantee the deletion of all historical records; failure may leave the penalty in the record.
Complaint materials and appeal materials often occupy different positions. Complainants can submit screenshots, evaluations, or accounts of events, while those complained about need to know why they are restricted, which materials can answer the complaint, and who will review them anew. If a penalty occurs before its reasons are delivered, appellants must also prepare materials after opportunities have already diminished. Procedural sequence thus becomes part of real consequences.
Appeals both serve to correct errors and incur costs. Whether people can find an entry point, understand the rules, possess evidence, and wait for a reply all change the actual weight of a penalty. If customer service offers only stock replies, appellants must spend more time finding effective explanations. When an account is banned, people who depend on that day’s income find it especially hard to wait. After a failed appeal, the penalty may remain fixed in the record.
For a casual worker such as the mother, every stretch of appeal time after an account restriction may also be a stretch without income. If the father is removed from a recruitment group, the work information he can access diminishes. If parents fail to submit materials and cannot find an appeal entry point, a child’s school-admission procedure may also be delayed. An appeal requires an entry point, time, evidence, and the ability to understand rules; those concerned must also be able to afford to wait for the result. In life, it has never been merely an abstract right.
After an appeal succeeds, relational chains may not all recover at once. A penalty can be revoked, yet previous visibility, trust in transactions, group position, or family arrangements do not immediately return to their former state. If others have already retrieved the records, the associated evaluations may persist. Examining appeals therefore cannot stop at the result field. Which permissions have been restored, which records corrected, which consequences have already spread, and whether supplementation and revision can continue must each be checked.
Time records themselves may also require calibration. System time can prove when an operation was preserved, but may not fully explain when a user saw a notice, when the user began dealing with it, or why the process was interrupted. Read status, online status, and response duration are all limited materials and cannot automatically prove a person’s attitude. If new materials can explain a delayed notice, device failure, or real event, the weight of the original time marker may need adjustment.
The consequences of platform events do not stop inside platforms. Accounts, records, scores, ranking, visibility, and penalties are transmitted through nodes such as labor, family, locality, state, market, and school, connecting events of labor, education, medical care, kinship, and markets. For example, when the father finds work through a recruitment platform or coworkers’ group, whether information is true, coworkers’ evaluations are reliable, and past chat records have been retained all affect which job he enters. Scores also enter income chains: when the mother takes hourly work or casual food-service work through a platform, scores affect her next opportunity; a negative review may lower her recommendation position, and lateness may also affect her eligibility to take orders.
Home–school platforms enter education chains. A child’s homework, notices, applications, and material uploads may all have to be completed through platforms. Whether parents can see messages in time, successfully upload materials, and reply on time also enters teachers’ evaluations and school procedures. Appointment platforms enter medical chains. The elder’s treatment involves appointments, medical records, medical insurance, and online materials. Unfamiliarity with operations, missed appointment slots, or failed medical-record uploads all change access to medical care. Chat records enter kinship chains: relatives’ demands for repayment, family disputes, money-transfer records, voice messages, and group silence may all be preserved, making the past within kinship relations easier to extract anew.
The same platform record may also enter several relational chains. An account restriction can first change income from orders, in turn leading the family to rearrange spending and care. Such consequences in life do not require family members to have consulted platform records beforehand. An argument in a group chat can affect local reputation and then be carried into renting, job seeking, or kinship relations. When records or evaluations pass between nodes, personnel or procedures must actually encounter, adopt, or forward the relevant materials. Transmission cannot be portrayed as records spreading on their own. At the same time, relational chains also branch, break, and weaken. A negative review may affect one ranking without necessarily entering long-term penalties. Chat records can serve as materials for claiming unpaid wages, yet may fail to change the processing result for lack of other proof. A failed submission of a home–school notice may also be repaired through a teacher's supplementary confirmation. Analysts must see both continuity and breaks; a platform's leaving traces cannot be equated with consequences necessarily lasting forever.
If specific platform rules and lawful data-processing arrangements preserve records of transactions, chats, scores, complaints, appeals, or payments, some of their content may be extracted anew in subsequent matters. Whether traces of likes, reposts, browsing, follows, blocking, group chats, location, and deletion are preserved, for how long, and by whom they can be retrieved must still be judged separately by platform, business activity, authority, and version. Account and record systems do not thereby become subjects acting on their own apart from personnel and rules. Records silent in ordinary times may gain weight anew when people take orders, seek jobs, rent housing, apply for services, join groups, or handle disputes.
The relevant personnel retrieve, evaluate, and adopt records in the current event, forming new judgments. When the results of handling the event are preserved as differences that can be extracted again in the future, this constitutes write-back. When later personnel or procedures use these materials in new events, they may change how the person concerned is recommended, trusted, or employed, how housing is rented to them, or how others pay attention to, help, or exclude them. They may also carry evaluations back into family, local, school, market, medical, and kinship relations.
This chain of subsequent effects must be distinguished from immediate consequences. A negative review may reduce opportunities to take orders; a complaint changes account status; and a ban may even cut off income. A failed appointment booking forces the family to adjust care. Not all of these consequences have to wait until the results of handling the event are preserved before they occur. A group-chat dispute may enter local relations, money-transfer records may be used by relatives in debt disputes, and rental chat records may affect a landlord's choice. A parent's absence from a home–school platform may also lead teachers to judge the family's cooperation on that basis. Whether these materials are used and when they take effect must be traced along concrete relations. Once the results left by company personnel and users handling events at hand are preserved, they can become materials extracted again in the future. Preservation, however, does not guarantee that they will take effect later. New materials, contrary records, counterexamples raised in appeals, and subsequent events may all prompt the relevant personnel to adjust weights, revoke judgments, or correct the original image. Write-back is not the final word.
Historical indices also require distinctions among versions. An original evaluation, a customer-service supplement, an appeal result, and later handling may appear in succession. Users who see only the earliest label may overlook corrections already made; retaining only the final result, meanwhile, may conceal losses previously caused by a penalty. Only by explaining when a record was formed, who amended it, why it was amended, and the event to which it applies can different versions be kept from being pieced together into a seamless history. Moreover, a record’s deletion or temporary silence still does not warrant a direct judgment that the relevant history has disappeared. A user’s inability to see a record does not establish whether other authorized personnel or procedures can retrieve it. Nor does a trace remaining in a system mean that it necessarily acquires high weight in the next event. Whether history can enter the present depends on who retrieves it and why, how it is interpreted, and whether contrary materials can enter judgment at the same time.
Platforms are often compressed into several fixed images. The tool image sees only life becoming more convenient. Although a tool explanation can account for this convenience, it leaves record preservation, opportunity ranking, identity evaluation, visibility distribution, and penalties out of view. The virtual-space image places platforms outside reality. A virtual-space explanation grasps the distinctive forms of online interaction, yet easily overlooks the fact that platforms have long entered nodes of labor, family, locality, school, medical care, market, and state.
Analyzing systems, rules, recommendations, and review helps explain how technology works. The purely technological image, however, assigns the entire problem to these mechanisms, concealing the social differences, evaluation authority, appeal costs, and real consequences behind them. Another account depicts technology as an all-powerful subject, as though every consequence were directly determined by a single algorithmic will. Actual events often involve rulemaking, receipt of data, user actions, human review, customer-service handling, merchant choices, and external qualification procedures simultaneously. Technical procedures certainly matter, but not every decision necessarily comes from the same mechanism, and there is no invisible unified will. The responsibilities that concrete personnel should bear must be made clear, while the continuing effects of ranking and record procedures on visibility and opportunities must also be retained.
The public-opinion image focuses on voices, quarrels, and expression. The public-opinion explanation sees expression and circulation; accounts, records, ranking, taking orders, payment, penalties, and transaction opportunities thus recede from view. Portraying all users as passive recipients likewise produces misjudgment. Users are subject to ranking, penalties, and entry restrictions, but they also preserve materials, lodge appeals, adjust actions, establish groups, evaluate one another, or move relations elsewhere. These responses may not be effective, and must certainly not be romanticized. They nonetheless form new records and nodes, thereby changing subsequent judgments.
Interpreters who adopt class theory may also subsume platforms under a total class image. Within a limited scope, they can extract income differences and relations of control in platform labor, yet cannot explain another set of questions. People whose incomes are similarly low still experience entirely different consequences because of differences in account status, scores, complaint records, response speed, ranking position, group relations, appeal outcomes, device condition, rule updates, and the risk of bans. The fact that people with similarly low incomes experience different consequences is admittedly insufficient on its own to prove that classes do not exist. The key is that class classification alone cannot explain the concrete mechanisms of accounts, scores, ranking, visibility, appeals, and write-back. Class theory also reduces accounts, records, scores, ranking, visibility, complaints, appeals, bans, group relations, and write-back to subsidiary materials of the total image. Class entities do not exist, but labor, wages, property, debt, risk, domination, and heavy exit costs are real; real communities and social differentiation networks continue to exist. The capacity for distinction extraction within a limited scope comes from real differentiating units and the processes of extraction, weight assignment, connection, and compression applied to them. It cannot be attributed to the class image. The image itself is an output of low-dimensional compression, projection, and naming of these differences. Class theory belongs neither to Model A nor to Model B; it belongs from the outset to the total-image-subsumption direction of Model C. Model C’s direction of erasure of distinctions between explanations, meanwhile, uses the multiplicity of differences to cancel judgments of weight. Both mechanisms destroy the conditions for healthy comparison.
Platform society is indeed complex. If we therefore declare all explanations much the same, or mix labor, consumption, social interaction, public opinion, and rules together, analysis cannot continue. Concrete events provide different entry points, and the materials to be checked must be identified according to the question.
Investigating a dispute over taking orders requires examining how account status, scores, response time, ranking, historical completion rates, complaint records, and penalties operate. In a dispute over home–school communication, one may first need to distinguish whether messages can be seen, when parents reply, and whether materials have been uploaded successfully. How teachers record matters and whether the system provides confirmation also affect the understanding of this course of events. Even where accounts and records are involved in both cases, their effects cannot be judged apart from the matter at hand.
In a housing rental, property listings, chat records, credit, payment methods, evaluations, and identity certification enter rental relations. Real-name information, available appointment slots, time windows, medical-record uploads, and payment in medical appointment booking must instead be placed back within the concrete process of seeking medical care. In content circulation, visibility, reposting relations, group position, reports, restrictions on reach, and account history in turn acquire greater weight. Only by distinguishing these events can we establish how each item of material connects to the corresponding consequences, rather than substituting a string of platform terms for explanation.
Acknowledging complexity in a healthy way means selecting a warranted window for the question and making the explanation's limits explicit. This window may be narrowed or expanded on the basis of materials. In handling a dispute over taking orders, we can first compare order requirements, completion records, both parties' materials, evaluations, and appeal handling, without having to explain the whole of platform society at once. If we find that account history, care responsibilities, or rule changes are altering the result, we can then expand the window. Limited distinction extraction remains constrained by events and materials; it cannot arbitrarily select differences favorable to an existing position.
Although platforms are complex, judgments remain possible, as do comparisons among models. Calibration of explanations must compare which key differences various models actually extract and which nodes of family, locality, state, market, and school they obscure. An explanation that identifies concrete personnel, procedures, materials, and chains of consequences, and explains how counterexamples compel revision of judgments, has greater explanatory power than one that sums up everything as “technical control” or “market domination.” Testing a model's scope of application must be done within the scope it claims and against evidence concerning the same question. Another explanation's better performance in a different event does not in itself directly invalidate the first. If, within the original model's claimed scope, materials or counterexamples concerning the same question require revisions to weights, paths, or scope of application, the model should be adjusted accordingly. Total-image subsumption that refuses to acknowledge counterexamples and destroys the conditions for comparison cannot be transformed into a healthy model merely by narrowing its scope of application after the fact.
Understanding platform society is best begun with an event in progress. Taking orders, scoring, account bans, complaints, renting, appointment booking, home–school communication, group-chat disputes, payment disputes, content circulation, account appeals, and information searches each offer a different entry point. Analysis has no fixed starting point. An account-ban event can be investigated from the grounds for the penalty and the appeal entry point; an order-taking event often first reveals ranking and scores; home–school communication may first encounter delivery problems involving notices, accounts, and materials. Analysts should trace the differences that first change eligibility, income, trust, or visibility and adjust the relational-field window in light of new materials. Steps that are always the same merely turn platform explanation into a procedural checklist.
As an event is traced, who extracts differences in life and which personnel or procedures preserve, rank, evaluate, restrict, recommend, or penalize gradually become visible as concrete relations open to examination. Account-based images may obscure whole persons; rule updates change weights; visibility distributes opportunities; and the costs of penalties and appeals change paths of action. Each of these effects has its own bearers, and records also return along relational chains to family, labor, school, medical, and kinship relations. Calibration must ask when old records regain weight, when current judgments are preserved as differences that can be extracted again later, and who retrieves, interprets, or assigns weight to them in subsequent handling. When counterexamples appear, the original explanation should keep an entry point open for revision. Any explanation that compresses these relations into a single image must undergo calibration by counterexamples.
Calibration must also retain positions from which those concerned can give their accounts. Company personnel, workers, merchants, consumers, parents, teachers, and group members may hold different records, and their accounts may conflict. Formal records present only some facts; the existence of disagreement does not mean all accounts are equally valid. Interpreters must compare the sources and times of formation of materials, the authority held by actors, and their connection to the current event, then mark which conclusions can stand and which parts still require revision.
Accounts, records, ranking, visibility, and penalties do indeed change people's positions and opportunities. Their effects must pass through concrete personnel, rules, materials, and procedures and cannot be attributed to an abstract platform. Differences in labor, family, locality, state, market, school, medical care, and community enter platforms within limited scopes, and processing results also return to these real relations. Tracing this path prevents platforms from being elevated into what society fundamentally is and complexity from becoming an excuse to abandon comparison. The total image of class theory likewise cannot subsume real and diverse platform mechanisms.