The base registered the floor on which both sides must stand, and multiple entry points made controlling a single storehouse no longer equivalent to controlling the record; yet materials can be stored and consulted while the controversy still does not automatically converge. The council of the Qingya river valley soon met a third kind of problem: the questioning faction pointed out that the effective date of the emergency regulations in the base differed by four days from the ledgers of the engineering camp, and those four days decided whether a particular requisition was lawful; the committee faction replied that the ledgers had been checked across versions and the discrepancy was registered on record. The problem was not solved; it only changed position: who re-examines this entry, under what conditions, and under what protections? The end of Chapter 20 said that public memory depends on malice paying a price and leaving a receipt; review is the place where the receipt is issued. If this link has no structure, the sharing established in the previous two chapters holds only as long as both sides consent.
This chapter treats the concrete conditions of review, not a blueprint for review institutions. Two myths must be refused: standpoint neutrality, as if finding impartial persons would make review hold automatically; and procedural fetishism, as if the existence of procedure made conclusions credible. The neutrality of review exists in no one's character; it exists only in conditions that can be checked one by one, and each of these conditions can fail. This chapter discusses four conditions and two failures, with the criterion at the end: procedural review still surpasses mutual accusation without procedure, but the criterion is the failure boundary, not the mere existence of procedure.
Why Review Cannot Rest on Neutral Identity Alone
The base of Chapter 19 and the version chain of Chapter 20 both treat the structure of materials at rest; review is the dynamic moment in which these structures are challenged: someone points at a shared entry and says "this entry is wrong," and the community needs a way to re-examine it rather than replay the original dispute.
Review easily slips back into two forms. One is the vote: the dispute is moved into the review chamber to count heads, and review degenerates into a second battlefield; Chapter 19 has already analyzed how this state hands the matter over to resources. The other is mediation: both sides are urged to concede a step, but a date is either correct or incorrect, and the average of two dates is not a third date. The object of review is materials and inference, not the intensity of standpoints: the conclusion may reject the challenge or revise the entry, but either can come only from relations of evidence, not from volume or posture.
The quality of review conditions cannot be judged from the winner's perspective, for the status quo already favors him. Those who truly depend on review are three kinds of persons: those rejected while holding new material, those whose material falls where the base does not reach, and those whose evidence sits behind entry points controlled by the other side; Chapter 19 noted that the last of these are the first to be pushed out under the exclusion mechanism. The test is therefore this: how much it costs a person who has just lost to raise a challenge again, what risks he must run, and whether his dissenting evidence is still safe after the review.
The Myth of Standpoint Neutrality
To demand that a reviewer have no standpoint is to demand that he has never lived in the valley: the committee's engineer surveyed the dam, the questioning faction's teacher signed the petition, and the outside mediator arrives with a framework of his own. Chapter 5 has shown that expression always selects its object and its emphasis; a reviewer's standpoint is not grounds for disqualification, concealment of it is. A reviewer who declares his standpoint and registers materials as they come is more checkable than one who claims to have no standpoint and leaves his sources obscure; this is isomorphic to the judgment of Chapter 20: an archive that knows where it leans is easier to verify.
Chapter 19 said that the neutrality of the base is constituted by procedure rather than declaration; review pushes this principle one layer further: neutrality is not an input but an output. It consists not in selecting neutral persons to place in the review chamber, but in a set of conditions: when shared materials, prior constraint of procedure, and distance of interest all hold, the conclusion has structural support whatever standpoint the reviewer holds; when any one condition fails, the most neutral person cannot save the conclusion. The consequences differ accordingly: a community that spends resources selecting "more neutral persons" sees the whole link collapse when the choice of person is challenged; a community that spends resources strengthening the conditions can replace the person and start over while the conditions remain in place.
Precisely because "neutrality" has market value, it can be performed. The commonest performance is the disguise of formal parity: both sides are given the same speaking time and the same number of rounds of questioning, as if procedural scale were neutrality itself. But when one side's material is a ledger and the other's is the oral testimony of twenty residents, identical procedure in fact favors the former: oral testimony cannot complete the same verification within the same time limit. Chapter 7 has shown that the cost of evidence filters who owns reality; review that is formally equal but unequal in evidentiary burden is that filtering dressed in procedural garb. Recognition does not abolish procedure; it registers each side's evidentiary costs, so that parity appears in the allocation of time and resources rather than in the symmetry of rounds.
The Four Conditions of Cross-Standpoint Review
The first condition of review sounds the most ordinary and is the least often met: the materials on which the reviewer is to judge must be visible to both sides, in the same version. Chapter 19 required that entries carry verifiable entry points, and Chapter 20 required that entry points state their version history; the two requirements converge in the review chamber: the list of materials should set out each item's provenance, version, and time and manner of acquisition, and be open to both sides. Without this list, review degenerates into trust in the reviewer's private impressions. The list converts review from trust in persons into structural trust.
Chapter 2 established that ten retellings are not ten observations, yet what a side submits in the review chamber is often not the material itself but "someone said" or "some document states." Retellings should not be expunged; sometimes the original really is lost, and Chapter 19's gap entries exist precisely for this. But they must be registered at a downgraded status: who retold it, when, the alleged original source, and whether it can be traced. Downgrading is not a half discount; it changes the retelling's position in inference: an original can support a confirmed entry on its own, while a retelling can support only one party's claim. A review that treats retellings as originals replaces the relation of evidence at the very point where it matters most.
Some materials cannot physically be shared: originals still in the other side's storehouse, records touching the privacy of living persons, documents whose copying is restricted. Review need not halt for this, nor pretend that sharing has occurred; the workable treatment is to register the restriction itself as the boundary of the review: which materials could not be obtained, why, and how far the conclusion reaches. Chapter 20's layered restricted storage applies here: materials in the appeal layer and the sealed layer are examined under conditions, and access leaves traces. A review that declares its boundaries and reaches a limited conclusion is still an honest review; a review that declares no boundaries and reaches a comprehensive conclusion uses the tone of conclusion to conceal a gap in the materials.
Condition Two: Protection of the Dissenting Witness
Those who bring adverse evidence, the bookkeeper within the committee who confirmed the four-day discrepancy, the member of the questioning faction who acknowledged that a certain photograph was questionable, bear risks arriving from both sides at once: their own side regards them as traitors, and the other side wants them to speak beyond what the material supports. Chapter 7 said that skeptics need to disclose their positions of interest; here the other face must be added: the protection of disclosure must be borne by the review structure, not rest only on the courage of the discloser. In a valley where the dissenting witness loses his place on both sides for testifying truthfully, the next time the accounts fail to match, the bookkeeper will choose silence; silence is not a matter of character but the product of a structure of risk.
The intuitive protection is anonymity, but anonymity has its price: an accusation cannot be cross-examined by the other side, its contact with the material cannot be verified, and it is easily forged; banning anonymity outright presses the risk back onto the discloser. The intermediate structure is registration without publication: the witness's identity is registered with the review structure, which verifies his actual relation to the material and then issues a certificate of "verified, identity confidential, questioning conveyed through the review structure." What the opposing side loses is the convenience of face-to-face interrogation; what it keeps is the right to question through conveyance. What the witness loses is the completeness of invisibility: if he gives false testimony, the registered relation can still be pursued.
Chapter 19 distinguished three kinds of objection to the base, and the same classification protects the dissenting witness in review. One who brings adverse evidence is not a traitor; one who is later shown to have misremembered is not a liar; only forgery and persisting in what one knows to be false enter the sphere of responsibility. The review structure should explicitly register this distinction and state separately, in the announcement of results, each witness's status of integrity: testimony borne out, memory deviation, unverifiable, suspected forgery, each assigned to its class. Without this step, "erring after testifying" and "giving false testimony" merge into a single stain in the community's memory, and the stain is a warning to the next witness.
Condition Three: Prior Constraint of the Adjudication Procedure
The review procedure must be published and take effect before the concrete case; it cannot be tailored to a case while the case is under way. If procedure can be modified mid-course, the side holding the power of modification has in fact adjudicated the case, for it can always reshape the procedure into the form that legalizes the desired conclusion. Chapter 2 analyzed how changing the definition of success after the fact allows every result to be described as expected; mid-course amendment is the same mechanism re-enacted at the layer of adjudication. Modification of procedure should therefore have a procedure of its own: disclosure of the relation between the motion and the case under review, a deferred effective date, and, in principle, the old procedure for cases already under way.
Once procedure is published, equal application matters more than correctness of text. Chapter 19 found that capture usually begins not with overturning rules but with selective enforcement: the same rule applied swiftly to one side and left to accumulate for the other. The review procedure must carry two observable indicators of its own: the distribution of acceptance times by the filer's camp (carrying over Chapter 19's tool) and completion records for each stage. Time limits are not merely efficiency but fairness: a review request suspended for half a year has the same effect as rejection, yet leaves no reason and offers no place for appeal. Delay is therefore the cheapest covert weapon inside review procedure; it will reappear later on a larger scale.
Chapter 19 established for the base that the adjudicator cannot simultaneously be the principal beneficiary; review pushes this requirement to the level of conclusions: if a review conclusion directly determines the adjudicator's own responsibility, funding, or reputation, the credibility of the conclusion is capped by that interest. The check does not ask whether the adjudicator is honest; it asks what he would lose if the conclusion went against him: the more concrete the loss, the shorter the distance. An interested adjudication can receive initial review from a role farther from the interest, with the parties appending objections; what is canceled is the structural convenience obtainable without cheating, and here this is extended from entry admission to review conclusions.
The output of review should not be a final-seal verdict. The conclusion carries its supporting structure: which materials it relies on, which stretch of procedure it applies, which questions were set aside for lack of material, and what new material would trigger reopening. Chapter 7 built stopping rules for doubt; the counterpart here is reopening conditions for review: stopping rules keep doubt from extending without limit, reopening conditions keep conclusions from hardening without limit; the two are faces of one coin. A conclusion that states "so found on the materials presently available" is more credible than a finding without conditions, for it acknowledges itself as part of a process.
Condition Four: The Reviewer's Distance of Interest from the Conclusion
Distance of interest is not moral caliber, nor does it automatically bring a lower error rate. A reviewer with no stake in the conclusion may still lack knowledge, and an interested participant may supply the most crucial material. What is proposed here is a conditional risk of incentive: if a conclusion directly determines the reviewer's gain, while method, process, and countervailing evidence go unchecked, he has both the opportunity and the motive to lean toward that conclusion; whether a shift actually occurred must still be settled by checking the materials. Chapter 19's scorekeeper is a fictional device for illustrating this risk, not a statistical proof. A review arrangement should examine relations of interest, professional competence, and conditions of accountability separately and preserve mutual questioning; the status of "outsider" cannot substitute for methodological reliability.
What the check of distance examines is not the reviewer as an individual but the chain of benefit: once a review conclusion is issued, whose economic position, political position, or reputation will change, through which links the chain passes, and at which point the reviewer stands. Every link should be registrable: who made the appointment, where the funding comes from, who decides on renewal, what historical relations exist with the parties. Chapter 7 required skeptics to disclose their positions of interest; this goes one step further: relations of interest rest not on self-declaration but on structural checkability. Declarations can always omit; appointment records and funding flows are registered in other hands.
Distance of interest is not the greater the better. An outside reviewer who knows nothing of valley affairs is farthest away and also most prone to error: ignorant of ledger notation, of the conventions of the water-level stations, of the difference between the two camps' vocabularies; the translators he depends on come from one side, and the distance is quietly shortened at the layer of translation. The workable structure is layered distance: fact-checking goes to reviewers who are distant yet professionally competent; contextual explanation goes to the registration of the materials themselves (the catalog and guide requirements of Chapter 20); the adjudicating stage keeps its distance of interest, and the understanding stage does not pretend to distance. Distance is a quantity distributed across stages, not a single number to be maximized.
Review Authority and Procedural Attrition
The first failure this chapter must confront head-on is caused by no one's negligence; it is the attack surface that necessarily confronts review authority once it is established. The first time the review structure issues a ruling against one side, the losing side has a nearly unfailing response: declare the review structure captured by the other side. The declaration needs no evidence; Chapter 19's signs of capture are for third-party observers, and the losing side needs only a motive. The other side can use the same weapon symmetrically: the next time it loses a ruling, it too declares capture. Every ruling manufactures ammunition for the next charge of "capture"; the more rulings, the thinner the authority. This is the soft spot of any cross-standpoint adjudicating structure.
The attraction of the capture charge lies in its cost structure: raising it requires no proof, refuting it requires proof. To clear himself, the accused must open the procedure, the materials, and the registers of interest; the more he discloses, the more detail the next charge can target; he can also refuse to respond, and silence itself has the appearance of evidence. The accusing side bears almost no cost: after declaring capture, unfavorable rulings are voided wholesale while favorable rulings continue to be cited, and this selective citation is precisely the strategic denial Chapter 19 identified. The capture charge is thus a weapon that only wins; it keeps review authority in continuous depreciation.
Two kinds of actors profit from this dissolution. One is the side that does not want to be reviewed: with review authority paralyzed, no one can move the status quo, and funding the attack is far cheaper than losing a review. The other is the operators of the standpoint industry: the weaker the review authority, the more disputes depend on volume and mobilization, and the more secure the position of those who manage mobilization. Those who bear the cost are the same people as always: the weak who depend on review for redress, the dissenting witnesses whose material is true yet cannot be heard, the ordinary residents who need public conclusions. Dissolving review does not require destroying it; it is enough to let "who knows whether review has been bought" become a catchphrase.
The capture charge is not always false; otherwise the word "capture" would not recur throughout this volume. Distinguishing real capture from strategic accusation relies not on the accuser's emotion but on the structure of the accusation. A charge of real capture points at concrete links: one acceptance that ran past its deadline, the chain of benefit behind one appointment, a long-term drift in the pass rate of one class of materials, and it is willing to submit to third-party examination; the strategic charge is marked by totality and selectivity: it never points at a concrete link, never submits to testing, and appears only when a ruling goes against its side. The cheapest touchstone: those who truly believe in capture do not cite the conclusions of the captured body; those who cite it are merely using it.
Procedure as a Weapon of Attrition
The second failure likewise arises from the review institution's own success. Once procedure exists, procedure itself can be weaponized: apply for review of an unfavorable ruling, then for review of the acceptance of that review, then for review of the qualifications of the reviewers of the review. Each request has procedural warrant, for review is in principle open to all, and this is where its fairness lies; yet stacked together they keep any conclusion from ever taking effect. Chapter 19 already handled stopping rules for meta-level disputes; here the same problem is amplified at the procedural layer: the entry to review must be open, and the number of layers of review must be bounded. Unlimited layers of review are not rigor but a machine for suspending every conclusion.
The asymmetry of procedural attrition lies in this: delay means different things to the two sides. For the side challenging the status quo, each additional layer of procedure adds a stretch of uncertainty, an expense, an attendance; for the side maintaining the status quo, delay is itself gain: the old conclusion stands unrepealed, and he continues to hold everything it brings him. The superficially neutral proposal of "one more review" almost always favors the status quo in its structure of interest. Recognizing delay cannot rely on inference from motive, for Chapter 3 warned against assigning motive directly from patterns of behavior; it relies on registering the distribution of time: on average, at how many layers do the reviews filed by one side end, and how many for the other; set the two numbers side by side in public, and the dividend of time shows itself in the figures.
Chapter 7 showed that the cost of evidence filters who owns reality; stacked procedure exponentiates this filtering. Many layers of review mean many layers of preparation: copying materials, checking versions, time in attendance, professional assistance. For the well-resourced side these are budget lines; for the side funded by pooling residents' money, the third layer of review may be the end. Procedure designers are often without malice: each procedure looks reasonable on its own, and no one answers for the cost curve after stacking. The quantity of procedure itself should therefore be audited as a burden: after all the stacking, how much money and how many days must an ordinary resident spend to reach the end; a procedural system whose answer exceeds what he can bear is in fact open to only one side.
Conversely, repeated application for review is not always abuse; the effort to exhaust procedure cannot be treated as troublemaking. A person genuinely harmed by an erroneous conclusion who appeals layer by layer is acting responsibly; the line of distinction lies not in the number of applications but in their structure. Abusive applications repeat arguments already heard, bring no new material, and play games with deadlines and jurisdiction; persistent applications each bring new material or point to stages not yet heard. Procedure should treat the two differently: the former may be consolidated, speedily rejected with reasons recorded; the latter must be received in full. A procedural system that refuses this distinction will either be dragged down by abuse or drive away the persistent along with the abusive.
What Review Can Resolve
Seen together, the two failures yield this chapter's core criterion. The existence of procedure proves nothing: a procedure declared captured in turn by both sides, a procedure consumed by unlimited stacking, a procedure formally equal but skewed in evidentiary burden are all procedures, and none constitutes review. Conversely, the absence of procedure is not freedom either: mutual accusation without procedure, you shout capture, I shout conspiracy, is likewise "review," only with volume substituted for material and position substituted for procedure; Chapter 19 showed how this state hands the dispute to the side with the most resources. The criterion therefore lies not in the presence or absence of procedure but in which stretch of the failure boundary the procedure operates within.
The failure boundary is not abstract; it is drawn by a set of observable indicators: whether the distribution of review acceptance times by filer is balanced; whether the distribution of review layers runs systematically high for one side; the proportion of rejected requests that attached new material, and the types of reasons; whether capture charges point at concrete links and accept examination; the changes in the dissenting witness's position after testifying. Any single indicator can have a chance explanation; when several drift together, the procedure remains intact in its text while failing in its function. These indicators are exactly the kind of receipt Chapter 20 spoke of: a receipt for degeneration, not for malice.
To admit that authority can be competitively dissolved and procedure weaponized does not lead back to mutual accusation. In the state of mutual accusation, capture charges equally exist, and are emptier and harder to test; delay equally exists and needs no procedure, only refusal to appear; the cost of evidence equally filters, and no registration makes the filtering visible. The value of procedural review lies not in eliminating these attacks but in forcing the attacks to leave traces, pay costs, and expose their structure. Chapter 20 said the point of checks and balances is to make malice leave an accounting; this chapter's conclusion is isomorphic: review procedure forces the capture charge to point at testable links, forces delay to appear as countable latency and evidentiary burden as registrable cost. The attacks remain, but the attacks themselves become indicators; procedure is the minimal structure that carries the indicators, and the indicators are the life of review.
What Review Cannot Handle
Review has places it cannot reach. Chapter 19 already noted that the core of some conflicts lies not in fact but in the ordering of values, and review is powerless before them: whether compensation should take priority or repair, whether loyalty or truth weighs more, are not questions materials can adjudicate. To stuff value disagreements into review procedure only produces standpoint choices dressed in the robes of adjudication, with the reviewer's ordering of values masquerading as a public conclusion. The method of recognizing the misplacement is the same as Chapter 19's: read the citation structure; when the facts cited by both sides agree to a high degree yet the dispute persists, review should stay its hand, for the problem lies elsewhere. Review's restraint is part of its credibility: a review structure willing to adjudicate anything cannot be believed about anything it adjudicates.
Some disputes are exhausted in the sense of material: every version checked, every witness registered, the gap entries honestly marked, and the disagreement remains, not because someone is dishonest, but because the materials end here. For such disputes the correct output of review is Chapter 19's disagreement entry: register both claims, register the boundary of the materials, leave the matter to the future. This is not a failed review but an honest one. To demand that review yield a unique conclusion for every dispute is to demand that it fabricate certainty, and fabricated certainty, once exposed, recoils upon the authority of the whole system.
Review can restore the reliability of conclusions; it cannot automatically restore relations. That review confirms the bookkeeper's testimony does not mean his position in the two camps automatically improves; that review clears the slandered does not undo a slander whose circulation has long exceeded the reach of the correction, and Chapter 14's problem of correction reach is re-enacted here. The rebuilding of trust needs the observable time Chapter 20 spoke of: whether both sides continue to cite conclusions after publication, whether the dissenting witness testifies a next time, whether the indicators stay public. These are not by-products but conditions of survival: a valley whose trust is exhausted loses not some particular review but the public space in which review can still be proposed.
The Boundary of Rebuilding After Review Fails
The order of recovery after the review system fails is isomorphic with Chapter 20's recovery of entry points: begin with the cheapest step. First restore the publication of indicators: the distribution of acceptance times, the distribution of review layers, the records of examinations of capture charges; then the registration of protection for dissenting witnesses; then the registration of chains of benefit and the rearrangement of distance of interest; only at the end comes the reorganization of institutions. The causes of both failures lie in structure; to touch persons first is to concede that the problem lies in persons. The cost of publishing indicators is too low to be refused on budgetary grounds, and at that point refusal itself is evidence. The recovery period should tolerate dual tracks: old conclusions annotated, disagreement entries expanded, parallel reviews coexisting; this is the observation period that the rebuilding of trust must pay for.
The boundary is thereby clear. Cross-standpoint review is not the gift of neutral persons but the continuous operation of four conditions: materials shared symmetrically, so that judgment can be pulled back to the list; dissenting witnesses protected, so that someone is still willing to bring adverse evidence; procedure constrained in advance and equally applied, so that rulings do not deform with the case; the reviewer's distance of interest from the conclusion registrable, so that error does not drift toward power. Each of the four conditions can fail, but failure has signs, the signs can be registered, and registration is itself the starting point of repair. Review cannot adjudicate values, cannot conjure materials out of nothing, cannot substitute for the time of trust; what it achieves is one limited thing: it makes re-examination independent of anyone's good will, makes both the raising and the rejection of challenges leave traces, and makes the cost of maintaining an erroneous conclusion higher than the cost of correcting it. Disputes will not subside because review exists, and this is not failure but precisely the boundary: what review sustains is the common conditions of dispute, not the end of dispute; holding this line, review neither swells into a court that adjudicates everything nor shrinks into a stage where the two sides take turns leveling accusations.