How Users Enter Evaluation
If the distribution of gains and losses is to remain open to questioning, users need the actual conditions for understanding and opposing evaluation. Describing arrangements from the analyst's position makes it easy to miss how hard it is for the person concerned to obtain the material. In real arrangements, the person most implicated in every entry of the ledger is precisely not in that position: she is the person the ledger records. The data are generated by her behavior, the profile takes her as its object, and the decisions fall on her opportunities. Under what conditions she can move from being the one recorded to being the one who questions is the first problem that participation in evaluation must solve. The Linchuan skills program and Ning remain fictional. Chapter 16 described its loop: data collection, personalization, more data. That chapter erected the institutional side of the right to observe — authorization, proportionality, prohibited boundaries; this chapter supplies the other side: the exit of the loop at the user's end. What conditions the evaluated need in order to participate in evaluation, and under what circumstances this participation gets wired back into the loop itself, are this chapter's two questions. This setting cannot prove anything about the participatory effects of real platforms; evaluating a concrete platform requires checking the permissions of the channels, the records of decisions, and what users actually encounter.
The relation between the person recorded and the ledger is asymmetrical: every entry of the ledger comes from her, every conclusion acts upon her, while the notation, the criteria, and the destinations lie by default outside her field of vision. As long as the person recorded can put questions to the record, the ledger is to her more than weather; if she cannot, then even the mildest ledger is only climate.
First, remove a confusion. Participation in evaluation is not answering "Are you satisfied with this service?" Satisfaction leaves the user on the measured side: her answer is one more row of data, entering at the loop's inlet. Participation is the measured one's intervention in the measurement itself: knowledge and control over collection, viewing and objection regarding the profile, appeal of decisions. The former takes her opinion as raw material; the latter places her in the evaluator's position — whether a system has a user's end at all is decided here.
Chapter 16 has already argued that the right to observe requires authorization, that consent cannot be proven by form alone, and that deletion, portability, and starting over each have their boundaries; this chapter carries all of that over without re-deriving it. What this chapter adds is their combination on the user's side: under what conditions scattered rights assemble into effective participation, and how the assembling process itself gets reused — that chapter asked what institutions may observe; this one asks whether the observed can intervene in that very matter.
The Three Layers of Participation in Evaluation
Participation in evaluation can target collection, inference, and decision respectively. The three layers are interrelated, but which permissions they require depends on the actual process: where no individual profile is formed, there is no need to fabricate a profile query; and where necessary records cannot allow individuals to alter the scope of collection, the basis must still be stated and submitted to proportionality review.
The first layer faces the inlet of the loop. Being informed means reaching the layer of use: what is collected, for what purpose, shared with whom, affecting which decisions; having control means that among these answers there are some she can genuinely change. Chapter 16 established that technical observability does not imply that observation is warranted, and that the permission to observe cannot migrate without limit internally; the user-side counterpart is: being informed is not receiving a notice, and having control is not possessing a switch. A notice delivered does not make the use knowable; a switch existing does not make the flow alterable.
The second layer faces the middle of the loop: inference. The collected data are folded into a profile — a risk score, a preference model, a competence label. Viewing means the profile is readable to the subject: what she reads is not a heap of raw data but "who the system currently takes me to be, and into which decisions this judgment has entered"; objection means the subject can say no to this judgment, and what that no triggers is review, not silence. Chapter 16 distinguished correcting facts, challenging inferences, and objecting to uses as three distinct rights; this chapter assigns these three across the three layers of participation.
The third layer faces the output of the loop: the profile becomes a decision — recommendation, ranking, eligibility, restriction. Appeal means the decision can be reopened: the reasons can be asked, someone is answerable for the review, and errors can be revoked. Its difference from objection lies in the object: objection targets the profile's judgment, appeal targets the consequences of the judgment. A correct profile put to improper use, and an erroneous profile whose consequences have already issued, are two different failures; run through a single channel, both are handled badly.
Within the setting of the Linchuan program, the profile feeds eligibility decisions, so breaks between the layers have concrete consequences. When Ning cannot see the basis of the profile, she may find it hard to point to the error, but she should still be able to appeal the decision and its reasons; review cannot be refused on the ground that she cannot name an internal error. Even if the scope of collection cannot for now be adjusted, viewing and correction retain independent value; if correction outside appeal can halt an improper use, it has already changed part of the loop. Evaluating participation cannot be settled by "an opinion channel exists"; one must look at whether the permissions each actual link requires are available, and at the gaps that remain unresolved.
Informed Consent, Profiles, and Appeals
The first layer of participation is the easiest to satisfy formally, and the easiest to lose its content while being formally satisfied. The conditions need to be written out item by item.
Blanket consent strips each concrete use of its own consent; Chapter 16 also said that splitting must not pile up into dozens of incomprehensible switches — both ends are failures, the blanket too coarse, the fragmented switches pushing the cost of control back onto the user. The default direction is equally decisive: collect by default with explicit opt-out, and no collection by default with explicit opt-in, yield two entirely different kinds of participation. Granularity and defaults have no uniquely correct values; what can be checked is their consequences: how many people made informed choices on how many uses — not how many people clicked consent on a single master form.
Chapter 16's conclusion applies here verbatim: consent under which refusing tracking forfeits the basic entrance bears survival pressure, not the expression of preference. The user-side supplement is this: the truth of control lies not in the switch's existence but in what actually happens after the switch is turned to refusal. Degradation can be honest — the version that provides no data really does have fewer functions, and Chapter 16 permits that difference; punishment cannot be — the quality of the basic service deliberately turned down to herd refusers back. The two can look identical on the interface; distinguishing them requires version records and change histories, not wording.
A promise that collection is controllable must be checkable in its fulfillment: after the switch is flipped, does the data flow actually change, or does only one column of the display disappear while the backend feeds as before? This is the same method Chapter 16 used in requiring that the costs of free services be visible: promise and performance side by side. The asymmetry of the material for this test — the user cannot see the backend — determines that the truth of control cannot be verified by individuals case by case; the external right of audit and collective inspection are therefore extensions of first-layer participation, not substitutes.
The honest boundary: being informed cannot be exhaustive. Explaining every layer of the model's inference to users amounts to publishing the system's entire capability, and to producing a new, enormous text no one finishes reading — Chapter 19's warning about recovery reports applies here: the design of checking must not in turn devour the time of the one checked. The goal of being informed is not omniscience but reaching the layers of use and consequence: which decisions are affected, which uses are expressly forbidden. Chapter 16's prohibited boundaries connect here — knowing how one is not being observed and knowing how one is being observed are the two halves of one and the same knowledge.
Viewing and Objecting to Profiles
The second layer is the least built of the three: its object is not data but judgment — and judgment is the part the system has the most motive to protect.
Profile visibility is not data export. Export pushes the labor of organizing back onto the subject: she receives a pile of fragments of her own behavior and still does not know what judgment the system has made on that basis. What the right to view demands is the readability of judgment: which category the system has placed her in, on what types of basis, and into which decisions it has entered. Chapter 16's portability right targets departure — what she can take away; the right to view here targets staying — how she is being treated.
Factual objection demands correction: the record says she interrupted a course, and she did not. Inferential objection demands review: the record is true, but the risk score derived from it is disproportionate. Use objection targets legitimacy: even if the judgment is true, is it valid to use it here — a genuine record of default does not entail entry into every recommendation. Chapter 16 established this tripartite division; this chapter supplies the receiving side: the three kinds of objection have different standards of validity, and merged into one channel, the strongest is usually diluted by the weakest — a flood of "I don't like this outcome" submissions supplies precisely the statistical camouflage for rejecting well-grounded inferential objections.
The mechanism's risk: objection records flowing back into the profile. The person who objects is flagged as difficult, her risk score raised, her opportunities contracted — this is no new invention; Chapter 16 already saw risk classification alter resources first; here only the trigger is swapped for the objection itself. The minimal condition is isolation: the existence and content of an objection enter no data stream other than the review of that item. Where this fails, the right of objection becomes structurally self-incriminating material — every exercise hands the other side a knife.
Without a temporal side to objection against the profile, every step the subject changes must first be negotiated against the whole of her own history. Ning need not deny that she interrupted a course three years ago; her objection points to which decisions that record is still weighting today. Chapter 16 said that past data freezes the subject's change, and gave the directions — time limits, decay, renewed proof; this chapter folds that into the framework of participation: the object of objection is often not the fact but the present efficacy of an old fact.
Appealing Algorithmic Decisions
The third layer touches decisions directly, and is therefore the one most often given form while substance is grudged.
Review is not running the model over the decision once more. The same model on the same input yielding the same output is not review; it is replay. Substantive review means the decision can be overturned by non-model reasons: supplementary material is read, exceptions are weighed, the conclusion can be written as "the model does not apply in this case." If the human link is reduced to a stamp, it lacks even the model's virtues — the model is at least cheap. What determines whether review is substantive is not whether someone signs, but whether the one who signs can state a reason the model did not state.
The first step of appeal is obtaining the reasons, not submitting evidence. Chapter 16 saw the reverse order: to correct a risk score, one must first submit health, family, and complete study records — appeal becomes the purchase of basic fairness with deeper exposure. The rightful order runs the other way: the receiving side first states what the decision rested on, the appeal proceeds against that basis, and only the material needed to resolve the dispute is collected. The burden of proof lies with the side that made the decision. For the platform, this means it cannot first compress a person into a risk score and then demand that she hand over more of her life to prove she is not that score.
Appeal consumes time, material, and emotion — these are precisely the margin in Chapter 19's ledger. A process that is long, harsh in its deadlines, and demanding of repeated self-proof collects the scarcest resource a second time: those who most need to appeal have the least margin for it — isomorphic with the distribution of overdraft in Chapter 19. The design of participatory channels is therefore itself a question of distribution: every threshold sorts, by margin, who can question. To evaluate an appeal system, do not look at how earnest the wording is; look at how much time that could have gone to recovery and to living is consumed in walking it through.
How Participation Gets Used Again
The sternest inference lies here, and it should be stated frontally and submitted to test: the channels of participatory evaluation are themselves wired into the profiling loop, becoming a new seam of ore.
Every flip of a switch, every deferred consent, every terms page half-closed is behavioral data. The system can treat consent not as a boundary but as a measurement: who cares about privacy, who clicks consent in a second, which wording lifts the consent rate by a few points. Consent thereby completes a change of identity — from the exercise of a right into an optimized behavior. Chapter 16 said that formal consent cannot prove fairness; this step goes further: formal consent itself becomes the object of optimization, and the direction of the optimization is the consent rate, not the comprehension rate.
Every "less of this content" on a recommendation, every objection to a profile, tells the system what she does not want — and the expression of dissatisfaction is often denser in information than the satisfied click. Objection thus acquires a double identity: for the user, the exercise of a right; for the system, scarce preference data. Chapter 16 said that feedback alters the one measured; this is its extreme form — the protest of the measured itself enters the measurement.
Appeal texts, stated reasons, verbatim dissatisfaction — language with intrinsic motive, unpolished. Chapter 21 said that unpaid creation is precious precisely because it is unpaid; the same logic is more piercing on the feedback side: the more genuine the complaint, the closer it is to a true expression of preference, and the closer to high-grade training material. The inference stops there — whether real platforms so use feedback requires verification of data flows; this chapter does not write it as an accomplished fact.
Taken together, this is a two-layer structure. The visible layer handles the content of participation: appeals are answered, feedback is thanked. The hidden layer records the behavior of participation: who participates, for how long, in what mood, whether they come back. The visible layer can be entirely honest while the hidden layer runs as usual; identifying this double layer requires proving no one's intent, only looking at a single data flow — whether participation records are used solely to process that participation. This item is also the core material of the evidentiary threshold that follows.
Channels Without the Capacity to Change
The second co-optation is more outward: the channels exist, operate, even bustle — they only cannot change anything. It turns participation from a right into a legitimacy device.
Platform participation needs to distinguish reception, substantive review, upholding the decision, and implementing correction: a response that does not address the problem posed proves no substantive participation; a decision upheld after full review does not automatically become ritual for want of reversal. Reversal rates can help disclose long-term trends, but they must be read together with appeal content, review reasons, and implementation; they cannot by themselves judge a channel effective or void.
Transparency reports, ethics pages, annual trust texts can be genuine disclosure or liability-exempting performance. The identifying marks are isomorphic with that volume's identification of decorative display: the samples drift toward stories attributable to individuals and untouched by rules — a bizarre one-off error, an earnest apology, a consummately handled redress; systematic problems have no corresponding statistics in the report, only curated cases. Whether transparency is performance is not read off the text's posture but off whether anything changes outside the disclosure: after the report, is there any follow-up to the problems reported.
What counts as handled, what counts as answered, what counts as closed — if all are registered unilaterally by the receiving side, the definition of participation is handed back to the side being questioned. The user takes her appeal for a questioning; the ledger records a closed ticket. The minimal correction is the same as in the institutional context: handling records can be checked by someone other than the receiving side — time received, the decision changed or unchanged, its reasons, the position from which it was made. The only novelty on the platform side is scale: the ticket system lets "closed" be produced massively, automatically.
The abundance of channels can itself become concealment: suggestion boxes, questionnaires, communities, councils — each really exists, each works in isolation, yet together they do not constitute participation — because none of them touches the rules. Identifying this structure cannot rest on counting channels; one must see where each channel's outlet leads: those that lead anywhere lead to the product — appearance, features, wording — or to the rules — the uses of data, the manner of decisions, the criteria of evaluation. Participation that leads only to the product changes the interface.
Who Benefits from Participation
Co-opted participation yields returns on both sides of the ledger, and presses its costs onto particular positions.
The data side: consent behavior, objection texts, feedback language — each a new seam of ore for the loop. The legitimacy side: response rates, participation statistics, transparency reports — each a credential toward regulators and public opinion. And there is a third, often overlooked: participation occupies time and feeling; deeply engaged users leave less often, and participation itself becomes a retention mechanism. Chapter 20 said that retention data cannot distinguish love from having nowhere else to go; participation rates likewise cannot distinguish the exercise of a right from engineered staying — what they may measure is precisely the intensity with which the channel holds people.
The thresholds of participation sort by margin. Those with time, with language, with institutional experience can write an appeal into effective material; the exhausted, those unfamiliar with the rules, those who articulate poorly, withdraw before the first form. Participation is nominally open to all; those who actually use it are the better-bargaining, while those deepest in the loop, most in need of questioning, participate least. After Chapter 19's overdraft gradient and Chapter 20's exit gradient, here is the participation gradient — a third line sinking in the same direction.
Those excluded from participation are not excluded from the ledger: their data enter the profile as usual, the decisions they bear are made as usual; only the questioning of the decisions is in fact frozen. The failure of a participatory system is therefore not a uniform loss — procedural burdens may make it harder for those with less time, information, or support to object. Whether the impact concentrates on these people must be checked against usage records and material from those who never entered the procedure; beyond formal participation, other channels of correction may still exist. Returns accrue to the two ends of the loop, costs to the deepest point; this is the shape, on the side of participation, of the distribution question this book has pursued throughout.
Evidence, Value, and the Conditions of Ledger Correction
The mechanism has been reasoned this far; before entering the empirical layer, one must stop and set out the threshold.
To judge whether participatory evaluation has lost its force, opinion submissions, handling records, and rule changes should be checked in connection. A high consent rate does not prove that opinions were adopted; the absence of changes, likewise, does not by itself prove the channel false: the institution must also state its reasons for refusing modification, and those reasons submit to review.
The test requires checking input, reasons, and consequences together: what users submitted, how those with authority handled it, why the decision stood or changed, whether confirmed errors were actually corrected. Statistics should separate established errors, upheld decisions, and open items, and examine the reachability of different groups. Unavailable material limits external judgment and also constitutes grounds for demanding explanation; by itself it does not prove that the platform has co-opted participation.
The reverse misreading must equally be blocked: high participation rates, eager feedback, active communities — these can be read neither directly as engineered staying nor directly as genuine empowerment. Chapter 10 said that duration of stay does not equal user benefit, and the participation rate is stay's close cousin. Judgment always returns to the structural triad: records of change, isolation of objections, and the cost of not participating.
To judge whether participation has departed from its mandate, one can check data uses, agenda authority, and records of replies and fulfillment, without first proving individual motive. But this material still needs its full context: non-disclosure may involve privacy or other concrete limits, or may conceal problems; each should be checked separately. A low rate of change and hard-to-obtain material are leads for investigation, not substitutes for judging the actual decisions.
The Irreplaceable Value of Genuine Participation
To judge all participation co-optation commits the error symmetric with judging all commercialization co-optation — Chapter 21 blocked the latter; here the former is blocked with equal force.
Designers know how the system is supposed to work; users know how it actually falls on a life: which recommendation is harmful late at night, which default is cruel during illness, which rule is impossible to keep in a concrete situation. This knowledge is distributed across countless local situations; no internal test can substitute for it. User participation is an important source of material for testing a system's effects. Independent review, field observation, and outcome tracking can also supply material, but they cannot automatically substitute for the parties' own expression of their needs and encounters.
An arrangement without participatory channels is not more honest, only more deaf — errors do not diminish, they only lose their address. The platform side carries one further risk: its errors are mass-copied, and one ranking defect falls at once on countless people's opportunities. The participatory channel is therefore not a luxury of individual relief; it is the system's device of error correction at scale.
The condition Chapter 20 established is honored here: affordable exit is the backbone of participation. The questioning of one who knows she can leave is questioning; the questioning of one who cannot slides toward pleading, and the far end of the channel can tell the two apart. In the other direction, effective participation also lowers the necessity of exit — many departures could have been won back by one questioning that was heard. Participation and exit are therefore not two topics but the two ends of one optionality: one end governs the weight of speaking, the other the price of silence.
The boundary must be argued as forcefully as the inference. Where user feedback has actually changed a rule, where an erroneous decision has actually been revoked, where a data use has actually been contracted — where these exist, participation is what it claims to be: the evaluated's intervention in the evaluation. The criterion is not the purity of participation but the record of change. A channel occasionally ritual and mostly real, and a channel occasionally real and mostly ritual, need the same material, not the same cap; to void both with a stroke of "co-optation" shields the latter and erases the former's contribution.
Ledger-Correction Conditions and Boundaries of Participation in Evaluation
Draw the conditions together. These conditions aim to point participation toward review of the evaluation and to preserve the authority to correct; they require neither that everyone participate, nor do they guarantee that every review be right. Their concrete effect must still be checked through handling records, the bases of decisions, and consequences.
The outlet of participation must be wired onto verifiable change: the handling records of appeals and feedback — time received, decision, reasons — checkable by someone other than the receiving side; the statistics of change published at intervals, with the definition of "closed" not registered unilaterally by the receiving side. Without this item, all remaining clauses are the decoration of an interface; with it, the response rate becomes a meaningful number.
Participation records are isolated from the profiling backend: the behavior of consent, the texts of objection, the material of appeal by default enter no data stream beyond the handling of that item, and exceptions require new authorization. Chapter 21 demanded a low-profiling path for activities that are ends in themselves; here the same principle is advanced onto participation itself: participation in evaluation is the behavior least of all to be wired into the thing evaluated; proportionality at this layer is not a balance but a clear tilt toward not recording.
Participation must not become a threshold. The decision quality of those who do not appeal, the completeness of service for those who do not give feedback, the basic entrance of those who refuse profiling — all must remain intact; Chapter 16 established equivalent paths, Chapter 21 established low-profiling participation, and this chapter merges them into one sentence: participation is a right, not a qualification. Silence must not be interpreted as satisfaction, nor converted into risk; Chapter 16 said that missing data cannot automatically become risk; silence here is the same.
There are three boundaries. First, participation cannot replace collective governance and external regulation: Chapter 16 said that no individual can negotiate an entire model ecology; user participation handles cases and experience, while the boundaries of use and the distribution of power require representative mechanisms and public rules; rewriting institutional responsibility as "users can choose" is the handiest trespass of the concept of participation. Second, participation cannot be erected as a duty: requiring everyone to evaluate how they are evaluated would turn participation into one more unpaid assignment — the shadow maintenance of Chapter 18 reappears here, and it lands on those already exhausting their margin on appeals. Third, participation in evaluation does not restore margin; it presupposes margin available: what the deeply overdrawn first need is not an appeal channel but Chapter 19's conditions of recovery and Chapter 20's exits. What participation in evaluation needs to keep checking is this: when a user raises an objection, is there someone who reviews the evaluation itself, and has this questioning been improperly turned into new profiling material. One page of the ledger has now passed into the hands of the person recorded; whether she can open it, and whether, once open, she can alter it, is where the evaluation system is itself evaluated.