36.1The editor’s legitimate discretion
An archive of any usable size cannot run on passive accumulation. Transcriptions contain errors introduced in copying, material arrives in an order that obscures relationships, and displays decay as links break or labels drift. Someone must be able to correct these faults as they arise. In the proposed small archive, that capacity is assigned to a single editor who may correct transcription and keep the display navigable. The point is not that an editor is always right, but that a duty to maintain the record implies a permission to alter it: without discretion, the assigned task cannot be performed at all.
This establishes discretion as a necessary condition of maintenance, not as a sufficient one. Two further things must be specified independently. The first is the remit: which classes of change count as transcription correction or display maintenance, and which do not. The second is a record of what was changed, so that later readers can inspect whether an edit stayed within its class. Both can be supplied in the design of the small archive; neither follows automatically from granting the editor a role.
The scope claim should also be read carefully. Discretion within a remit makes certain errors correctable; it does not make the resulting record reliable. Reliability depends on how often the editor notices faults, how consistently the remit is applied, and whether the record of changes is itself kept current. Those are possible sources of failure, not inevitable features of the arrangement. A single editor could be careful and transparent, or careless and opaque, while holding identical formal powers.
What the arrangement therefore compares is not edited against unedited output, as if those were two fixed qualities. The task is held constant — maintaining the same archive’s transcription and displays — and the inputs remain the same materials. What varies is only whether a named actor has permission to perform the necessary corrections, and whether that permission is bounded and recorded.
Take a contested deletion in the proposed small archive. The editor treats a marginal fragment as a transcription error and removes it from the public display; a scholar says the fragment is evidence of an alternate reading and should remain visible. The dispute is not settled by calling the act a correction, since correction and selection can produce the same visual result. What distinguishes them is the object affected: correcting changes how a retained item is rendered, whereas selection determines which items enter or leave the represented set. So a deletion that leaves the fragment in the underlying record and recoverable in an alternate view is compatible with delegated correction; a deletion that removes the fragment from the record altogether is a substantive selection, whatever editorial label it carries.
This distinction between correction and selection is a necessary test, not a sufficient guarantee of legitimacy. Even a record-preserving edit may mislead if the display makes the absence invisible, and even a reversible edit may do harm before the steward reviews it. Conversely, deleting an item need not be abusive: the remit may authorize exclusion of duplicates or corrupted material, and that authorization is doing the work rather than the editor's judgment alone. Conditional supposition: if the archive's rules require the fragment's status to be documented and the advisory panel's objection recorded, the editor's choice becomes contestable without transferring the decision right to the panel. If those conditions are absent, the same discretion quietly enlarges into selection power.
The steward's power to reverse a disputed deletion answers a different question from the panel's power to recommend. Reversal corrects an outcome after the fact; it does not bound the class of edits the editor may attempt. Thus the concrete case fixes the boundary: transcription correction stays within remit when it alters rendering rather than the represented set, and the set-changing act requires the separately specified reversal route, not merely an advisory comment.
36.2An advisory role without decision power
In the proposed small archive, three rights are separately assigned: the editor may correct transcription, an advisory panel may recommend revisions, and a steward may reverse a disputed deletion. These rights touch contested content in different ways. The panel supplies a second reading and can challenge the editor's transcription or framing; the steward may undo the edit. The panel thus may improve the information on which a decision rests without acquiring the decision itself.
The distinction matters because counting the parties involved does not show who shapes the outcome. Arnstein's ladder, developed for historical United States program contexts, distinguishes degrees of participation by the power citizens actually hold in shaping the result (Arnstein, 1969), and this study does not treat it as a validated universal ranking of every arrangement. What follows locally is that the panel's advisory status is defined by the absence of a decision right. A recommendation is not a sufficient condition for a revision, just as the editor's authority over transcription is not sufficient authority over deletion, which the design assigns to the steward. Conversely, none of these powers is necessary for the others: the steward could reverse without prior advice, and the editor could correct without consulting anyone.
The advisory role's epistemic contribution is therefore conditional rather than automatic. If panel challenges are never recorded, or never reach the steward when a deletion is disputed, the assigned right produces no additional information at the point where it matters. A workable design must specify how a recommendation travels: whether it accompanies the edit, whether the editor must respond to it, and whether the steward sees the disagreement. These are procedural questions distinct from who holds the final word. Possible outcomes range from a panel whose challenges have no practical reach to one whose reasoning reliably structures what the steward considers; which obtains depends on those accompanying duties, not on the advisory label alone. None of this describes an observed institution.
The label "participation" describes a relation — being asked, heard, answered — not a share of authority. In the proposed small archive, the advisory panel's competence stops at recommending a revision: the editor may correct transcription as an operative act, and a separately specified steward holds the right to reverse a disputed deletion. These are designed rights of a prospective institution rather than observed ones, so what follows concerns what such an allocation would mean, not how any archive has behaved.
Two inferences follow. First, consultation is not a sufficient condition for influence over an outcome. Even advice that improves information and challenges an edit leaves the result contingent on uptake by whoever holds the decision right; the possibility of change is real, but not a guaranteed correction. Nor is consultation necessary for a decision to be well founded, and holding the decision right is likewise not sufficient for a well-founded decision. Equally, nonbinding names the absence of a claim to compel, not the absence of effect. Reasons that cannot compel may still be decisive in fact, and a recommendation can settle a question before it hardens into a dispute — a practical value no power to compel would by itself supply.
Second, the scope of the advisory role should be stated matter by matter. A panel competent to recommend on revisions is not thereby competent to restore a deletion, and the right to be consulted on transcription is not a right to author the corrected text. Calling all of these "participation" risks comparing unlike things: the label captures voice and access to reasons, while operative power to approve, reverse, or maintain remains with someone else. The limit is one of jurisdiction and force, not of value.
36.3An identified decision owner
Ask first not who has been consulted but who may finally approve, reverse, or maintain the disputed change. In the small archive sketch these powers are assigned separately: the editor may correct transcription, the advisory panel may recommend revisions, and a separately specified steward may reverse a disputed deletion. They are designed rights in a prospective scheme, not observed institutions, so no claim follows here about how such an arrangement would actually behave in use.
Distinguishing the three powers clarifies what ownership of a decision means. Approval settles the change; reversal undoes it; maintenance leaves it standing because someone with the relevant power has declined to disturb it. The owner is the actor whose defined power reaches the outcome at issue. That qualification matters because the powers are issue-specific. Holding a right to reverse a deletion is not sufficient for authority over transcription, and the editor's power to correct transcription does not extend to deletions. A reversal right is likewise not necessary for taking part in the argument; it is necessary only for changing that particular result rather than merely discussing it.
A granted power and its exercise also come apart. If the steward holds the reversal right, the deletion may be maintained when that right goes unused, so reversal is possible but not inevitable. Supposing the steward were additionally charged to review every disputed deletion, non-action might then count as maintenance rather than silence. Absent such a charge, an unused power leaves the editor's decision in place by default, not through any further approval. Identifying an owner therefore means mapping each contested question, rather than the archive as a whole, to the body holding the power over its outcome.
In the proposed small archive, authority should be classified by operative rights rather than by the number of names attached to the process. An editor corrects transcription, an advisory panel recommends revisions, and a separately specified steward reverses disputed deletions. Three constituencies therefore appear on the page, but headcount is the wrong measure. If only one actor's act can change a given result, the remaining participants belong to a procedure, not to a distribution of power over its outcome.
The distinction separates necessary from sufficient conditions. A panel recommendation is not a sufficient condition of revision, since revision follows only if someone with editing rights adopts it; nor is it necessary, if the editor may correct transcription without asking. Shared authority would require at least one case in which another actor's act is sufficient to alter or hold the result. In this design that condition arises only for the steward, and only when a deletion is disputed.
The scope of the conclusion is accordingly narrow. The steward's power is genuine but domain-limited: it reaches disputed deletions, not transcription or ordinary revisions. Authority in the archive is thus distributed by subject matter rather than uniformly across the editorial process, and an account that describes the panel as a co-owner would be measuring attendance rather than capacity. Conversely, the absence of binding power does not make the panel's influence impossible or its marginalization inevitable; recommendations may still shape what editors adopt. That influence is real, but it should not be relabeled as authority when the panel cannot itself change the result. Since the arrangement is a proposal, these are conditional claims about designed rights, not findings about an operating institution.
36.4Several centers of review
A list of names is not the same as a distribution of authority: if one actor alone can change the result, the others are consulted rather than sharing the decision. Placing review in several centers alters this structure. Each center becomes a separate route by which an interpretation can be contested, because a party dissatisfied with one forum can seek a ruling from another that is not subordinate to the first. The gain is procedural, not epistemic. A second center does not make an interpretation more accurate; it makes the interpretation revisable.
Whether that route is real depends on what the centers can do. A center that may comment but cannot require reconsideration adds a venue for objection without leverage. One that can send an interpretation back for amendment, or attach a recorded dissent that travels with the decision, changes who must answer. Multiple forums alone are sufficient for nothing in this respect, and they are not strictly necessary either, since a single center empowered to reopen would also admit challenge. What multiplicity adds is a second access route when the first is closed to the party.
In an invented illustration, two reviewers disagree over whether a conditional objection belongs in the summary. An escalation rule names one decider and assigns that person responsibility for including or omitting the objection. The rule settles who answers for the choice; it does not settle the interpretive question, which remains open between the reviewers. That separation is the point. Assigning responsibility is necessary for the summary to be treated as decided, but it is not sufficient to show that the objection was rightly placed or rightly excluded.
If the escalation record retains the disagreement rather than absorbing it into the ruling, the unresolved question stays available to later review centers. That is possible under such a design, not inevitable: a rule may name a decider while quietly discarding the dissent.
Several centers of review multiply the points at which a decision can be made, and that multiplication carries two costs: decisions may conflict, and reaching them takes more work than a single determination would. Conflict is a possibility the arrangement permits, not an outcome it compels. It grows likelier when centers consider the same material under criteria that rank considerations differently, and stays latent when their scopes barely overlap. The exposure to conflict therefore varies with overlap and with how far criteria are shared; it is not a fixed property of having several centers.
The burden claim also needs a precise comparison. Holding the review task and its inputs fixed, what additional centers add is coordination: deciding who encounters the question first, transferring the record, and sequencing or reconciling conclusions. That is distinct from the work of review itself, which several centers may divide rather than duplicate. Nor does the coordination cost of several centers show that one center would be better overall; the comparison sets that cost against whatever distributed authority is taken to provide, which remains a separate matter.
In the illustration already on the table, two reviewers disagree about whether a conditional objection belongs in the summary. An escalation rule can name who decides, and that is sufficient to fix responsibility: someone is answerable for what the summary says. It is not sufficient to settle the interpretive disagreement, which stays open for anyone who revisits the objection. The rule thus answers an allocation question while leaving an interpretive one unresolved, and treating the first as a solution to the second would mistake an answer about who for an answer about which reading is correct. A procedure that records the disagreement as unresolved preserves the distinction; one that treats escalation as endorsement of the winning reading conceals it.
36.5Conflict between reviewers
Consider the following hypothetical case: two reviewers agree that the source supports a conditional objection, yet disagree about whether that objection belongs in the summary. The dispute is therefore not about the text's content but about what a summary owes its readers. The shared source constrains both judgments but does not determine either. Both reviewers can therefore preserve the condition while defending different choices about inclusion and prominence. Agreement on the reading is thus necessary for agreement on placement, but it is not sufficient.
That distinction matters because the two kinds of disagreement respond to different remedies. If reviewers diverge about what the source supports, closer attention to the passage may narrow the gap. If they converge on the reading yet diverge about summary policy, further evidence about the source cannot decide the question, because what is at issue is the function of the summary rather than the state of the record. Treating the second dispute as though it were the first mistakes a policy judgment for an interpretive failure.
Nor does having two reviewers make conflict inevitable; it makes conflict possible rather than guaranteed. What follows is not that one reviewer must be wrong. Both placements can be defensible given the shared reading, and selecting one does not show that the other misread anything. Accordingly, omission from the summary should not be recorded as a finding that the objection lacks support. The contested item remains source-supported even after a placement decision excludes it, and the authority of that decision extends only to what the summary displays. Noting which question was contested keeps that boundary visible for later readers.
An escalation rule answers a question of responsibility: who decides, and by when. It does not answer the interpretive question at issue. In the prospective design already sketched, two reviewers disagree about whether a conditional objection belongs in the summary. Escalation can name the person or body empowered to settle placement, but that settlement is a decision, not a demonstration that one reading of the objection is correct. Conflating the two turns an allocation of authority into evidence about meaning.
The distinction is one of necessary and sufficient conditions. A named decision-maker may be necessary for the summary to be produced at all; that person's involvement is not sufficient for the summary to be accurate or fair. Likewise, the end of procedural blockage is possible while interpretive disagreement persists; it is not inevitable that disagreement will disappear once responsibility is assigned. What the rule removes is uncertainty about who acts, not uncertainty about what the record supports.
The rule should therefore be applied within a stated scope. It governs who has the last word on a particular editorial act—whether the objection is included, and perhaps how it is framed. It should not be cited as settling what the source-supported interpretation requires. If the same interpretive question returns at a later revision, reopening it is consistent with the rule rather than a violation of it. By contrast, a rule written as though the decision also certifies the interpretation makes later reconsideration look like insubordination, which quietly converts procedural closure into substantive closure.
The practical application follows from that scope. Escalation is for reaching a workable output despite disagreement; interpretation remains open to argument on its own terms. Treating the former as proof of the latter would make the rule do work it cannot perform.
36.6Maintenance and resources
Maintenance asks a different question from initial approval: who does the continuing work, and who can take it over. The competing simple design of one accountable editor with a visible change log places responsibility in one place and leaves a trail of edits. That arrangement can reduce confusion about whose decision prevails, but it also concentrates the continuing workload in a single role. If the editor leaves, the design's prospects turn on whether someone with available effort and sufficient continuity of context can assume the task. A change log records what changed; by itself it does not carry forward the reasoning behind decisions, nor does it make a successor appear.
These requirements are therefore necessary rather than sufficient. Available effort, continuity, and a route for handover each have to be present, but their presence does not by itself guarantee that maintenance will occur or that it will be done adequately. Conversely, an editor's departure does not make breakdown inevitable: a workable handover route can transfer the task, provided the successor can reconstruct the state of the record and identify what remains unsettled.
That qualification matters where the escalation rule considered above resolved responsibility but left substantive disagreement standing. The rule settles who may act, not which instruction is correct. In the prospective design, if additional review centers have supplied conflicting instructions, maintenance inherits the work of deciding which directives still bind and of keeping that determination visible to whoever comes next. Extra review centers thus add not only effort but a stream of competing guidance that a maintainer must reconcile or retire.
The change log shows which edits were made, but whether it suffices for handover depends on what a successor must reconstruct in order to resume work. That question is about continuity of the record, and it is distinct from the question of who nominally holds the editing role.
An assignment of authority states who may act; it does not supply the time, attention, or standing required to act. Take the design that names one accountable editor and records every change in a visible log. The chart makes the mandate legible, but the mandate can remain inert: the named editor may have no scheduled effort for reviewing submissions, no capacity for the upkeep the log itself demands, and no practical way to refuse competing demands. Formal office is thus not sufficient for maintenance, even if it is necessary for attributing responsibility when maintenance fails.
Conversely, actual capacity cannot be read off the organizational chart. A diagram of nested review centers may display many nodes of attention without showing whether anyone can attend them, whether their instructions agree, or whether conflicts among them are ever settled. Additional review centers in this competing design add both instructions and effort. Their existence is possible evidence of distributed attention but not inevitable proof of it. The chart records claimed authority; capacity is a different quantity, namely the effort actually available to a named role for a specified task over a specified interval.
The qualification matters for application. As a conditional supposition, suppose two designs carry identical formal assignments but differ in whether the editor's hours are funded. Any difference in what is maintained should then be traced to the resourcing difference rather than to the assignment. That does not make resourcing sufficient either: a well-funded editor facing contradictory instructions may still fail. The narrower point is that comparisons require a fixed task and inputs. "Authority allocated" compares permission structures; "maintenance performed" compares effort delivered on a given task. Treating the first as evidence of the second mistakes a statement of office for evidence of work done.
36.7A responsibility map that can be challenged
A responsibility map should record, for each role, the actions it may perform rather than the status it holds. Proposing an edit, authorizing it, implementing it, reviewing it afterward, and reversing it are distinct acts; assigning a role to one of them does not settle the others. The map should also mark where an action requires a second actor and where it does not, since a veto is not the same as a duty to consult. This distinction matters because a formal assignment without resources may sit inert: the map establishes permission, not capacity to act.
A second layer should state the route for challenging the allocation itself. That route needs standing (who may object), grounds (which defect counts), a forum (which actor receives the objection), a remedy (reversal, reassignment, or escalation), and a record of the outcome. Without a remedy, an objection is only a complaint; without a record, the allocation can be silently altered. Neither layer is sufficient on its own: a permission list makes scrutiny possible but does not ensure the authority is exercised, and a challenge route makes objection possible without making any objection succeed.
Applying both layers to the competing design of a single accountable editor with a visible change log shows the map's limits. There, proposing, authorizing, and implementing may fall to one actor, but the change log makes prior changes available for review and reversal. Adding further review centers can produce conflicting instructions and extra effort; this is possible, not inevitable, and the count of review centers compares how many bodies must be consulted before an edit proceeds, not the correctness of what results. The map therefore tests whether a claimed division of authority matches the actions and the challenge route, while leaving the practical question of which arrangement suits a given collection open.
The allocation of actions and challenge routes just set out should itself be treated as a proposal, not a settlement. Its authority is conditional on the collection remaining within the scope for which it was drawn. Making the map revisable is necessary if participants are to contest the allocation rather than only the decisions produced under it, but revisability is not sufficient: a revision route also needs a specified trigger, a decision point, and someone answerable for the outcome. Otherwise the map can be nominally open while its terms stay fixed in practice.
A competing simple design fits a small, low-stakes collection: one accountable editor, supported by a visible change log. Where the collection is small and the consequences of error are modest, this may be preferable to adding review centers, since each additional center can introduce conflicting instructions and duplicated effort. That is a conditional judgment about scope and stakes, not a claim that simplicity always yields better decisions. Nor does consolidation eliminate disagreement; a single editor may disagree with an earlier self, and a change log records what changed without necessarily recording why. The log therefore supports traceability, which aids scrutiny but does not by itself supply grounds for reversal.
The comparison must fix the task and inputs before the arrangements are compared. Holding the collection, its decision tasks, and the material available constant, the question is what each arrangement costs in delay, instruction clarity, and error detection, and who bears those costs. It is not a comparison of structures in the abstract. Complexity in a responsibility map is a cost to be justified, not evidence of greater fairness; equally, the absence of visible contestation in a simple design may reflect quiet acceptance or the absence of any practical route to object.