47.1Declare the task and population
The proposed workflow begins by fixing three things at once: the task to be performed, the readers the output serves, and the population of cases and records to which the task applies. For the later-hours question, the task is to assess the conditional proposal under the committee's declared remit and the known source version; the readers are committee members who must decide whether the ordinary summary-plus-links baseline is adequate. The population is the bounded set of cases falling under that remit and traceable to that source version—not a tally of the people who might be scheduled differently, and not the wider set of staffing questions the committee could conceivably entertain.
A declaration is useful only if it is operational. It should let a later reader decide whether a given input belongs to the declared population without re-reading the whole argument. The disputed removal of a staffing condition tests exactly this. If that removal is an edit to a record already inside the declared population, the task stands and the removal must be handled as a contested input within scope. If, instead, the removal changes the remit by making staffing conditions irrelevant to what counts as a later-hours question, then the task itself has changed and must be redeclared. Absorbing such a change silently would quietly replace the original standard of success with a new one.
Two limits matter here. First, declaring the task and population is necessary but not sufficient: it does not select an output form, certify that claims are supported, grant revision rights, or name who decides. Those elements come next. Second, a loose boundary makes later misassignment possible rather than inevitable; the risk is realized only when an input is admitted or excluded without checking it against the declaration. Keeping the boundary explicit therefore does real work even before any output is produced, and it gives the disputed removal a place to be argued about rather than a place to disappear into.
An exploratory task may remain provisional. What it may not do is change its target without notice. The declaration therefore fixes three things: the question asked, the readers who will rely on the answer, and the population the answer covers. Within that frame, provisionality is a licence to learn, not a licence to move the goalposts afterward.
The later-hours question supplies the constraint. The committee's declared remit settles what it may ask; the known source version fixes the material available; the ordinary summary-plus-links baseline gives a standard form for the output. On those inputs, an exploratory answer may report what the evidence supports and where it is thin. If later work were to widen the question — suppose it reached a different shift pattern or a wider group of staff — that would be a new task, and the change would have to be entered in the record with its reason and timing. Only then could a reader tell whether the result held under the original conditions of success or under altered ones.
A durable record of revision is therefore necessary for honest provisionality, but it is not sufficient. The record must also separate a change of encoding from a change of task. Holding the question, readers, population and source version fixed, the summary-plus-links baseline and the alternative produced by dropping the disputed staffing condition are two renderings of the same inputs; comparing them compares encodings, and the conclusion that one fits the task better is a claim about fit. If the removal instead changed who is covered or what counts as success, the comparison would no longer be like for like, whatever the record said.
Provisionality also constrains the strength of the outcome claimed. An exploratory result makes a favourable later decision possible; it does not make it inevitable, and it does not extend the committee's remit. The declaration thus licenses cautious findings while keeping every later redefinition visible, which is what the next step will need in order to preserve the relation to the source.
47.2Preserve the source relation
Treat the later-hours question as a design illustration rather than a documented case. Its protocol record identifies the task and population as declared in the preceding unit, names the committee's declared remit, and fixes one source version of the question as the object being transformed. Against that baseline, the ordinary summary-plus-links output remains comparable only if the links resolve to the same source version; if the linked material has moved on, the comparison silently changes inputs. Fixing the task, population and source version therefore precedes any comparison of output formats: the encodings may differ, but what they encode must be held constant.
One transformation in the illustration is disputed: a staffing condition is removed from the later-hours question. Preservation here means keeping enough of the relation between the source version and the revised output to inspect that removal — the retained source text, the revised text, and a record identifying what the removal was taken to do. That relation is necessary for inspecting the transformation. It is not sufficient for judging the removal: inspectability shows what changed, not whether the change was warranted or under whose authority it was made. A reviewer who can see the removal may still lack the committee's reasoning, and a complete relation record does not by itself establish that the staffing condition was removable.
The countercondition is narrower. If the summary-plus-links baseline is judged by how much it compresses, the quantity compares lengths of outputs, not fidelity to the source version; low compression may coincide with a broken relation, and high compression with a preserved one. What must be compared instead is whether the task-relevant transformation remains inspectable at the same declared task and inputs. The relation record supports that check; it does not ensure that the resulting decision is sound.
Inspection claims must match the access actually available. Suppose, in the prospective later-hours design, the record remains internal to the committee: the known source version, the declared remit, and the disputed removal of the staffing condition are retained in a log that only members and an appointed reviewer may open. The retention claim does real work. The reviewer can compare the earlier and later formulations and identify which alteration removed the condition. What cannot be claimed is that any reader can perform that comparison.
This qualification matters because unrestricted publication is neither necessary nor sufficient for the relation to be preserved. Publication of a summary-plus-links package, without the earlier formulation, leaves the alteration uninspectable even though the material is open. Conversely, a closed record holding both formulations preserves the comparison for those with access. Restricted access is therefore sufficient for some purposes (revision by the decision authority, challenge within the committee's remit) and insufficient for others, such as a public challenge by someone outside the committee. The procedure should name the intended purpose before deciding how much access the record must carry.
Comparisons should also hold the task fixed. Set the question as whether the committee may extend hours given staffing, and hold the source version and remit constant; the open record and the access-controlled record are then different encodings of the same inputs. An open encoding that substitutes a revised source version is not a stricter version of the same thing. A count of preserved links measures how many references survive, not whether the transformation from the earlier wording can be examined at all.
Applied to the disputed removal, the bounded claim is this: an authorized reviewer, not the public, can inspect how the staffing condition disappeared from the proposed wording. Whether that suffices depends on who is entitled to question the decision; the protocol should record that entitlement as a scope limit rather than settle it by requiring publication.
47.3Produce alternatives where needed
Produce several formulations when one sentence would conceal a conflict in what the answer is required to do. The signal is not the number of clauses but the presence of demands that cannot be met by the same wording: one requirement may ask for a qualified answer while another calls for a plain one, or two audiences may need different levels of detail. A single sentence faced with such a split tends to settle it silently, and the lost demand becomes visible only when the answer is used. Where the demands can in fact be met together, the split is only apparent and the remedy is unnecessary.
That gives a criterion, not a quota. Alternative formulations can preserve distinct requirements, but they are not necessary whenever a text is long or careful; a single formulation suffices when the requirements can be met together without suppression. Nor is plurality inevitable where requirements differ: it remains a possible design response, chosen when separate wordings keep each demand inspectable. The test applied to a set of formulations is therefore whether each preserves a relevant requirement the others would drop. Removing one that adds no distinct requirement costs nothing; removing one that does creates the concealed conflict again.
When alternatives are compared, hold the task and its inputs fixed and vary only the formulation. Otherwise the comparison measures a changed question or changed material as much as a changed wording, and no conclusion about which formulation preserves more follows. What such a comparison names, then, is the difference in what each formulation lets the answer meet under the same demands. Nothing assumes in advance that one will prove superior. This keeps the conflict visible and locates it in the answer requirements rather than in the mere existence of more than one sentence.
The paired move establishes an upper bound: where one sentence conceals a conflict in answer requirements, separate formulations are warranted. The present move draws the lower bound. Plurality is justified by a distinction that matters to the answer, not by a display quota or a preference for variety. Where two entries would say the same thing for the same purpose, one display entry can carry both, provided the records retain separate source identities. That separation matters later if one source is revised, withdrawn, or found unsupported; collapsing the identities would make the shared entry harder to check or split.
In the invented illustration, a harmless duplicate therefore meets the test for consolidation. The test is not whether two wordings can be told apart, but whether an answer, decision, or later review would have to treat them differently. If it would, preserve separate formulations; if it would not, shared display is sufficient. This keeps the procedure minimal: produce alternatives when the distinction has a practical consequence, and reduce them when it does not.
A countercondition qualifies the lower bound. Separate source identities do not by themselves require separate entries, and separate entries do not by themselves establish an important distinction. What matters is whether the distinction survives contact with the intended use. If a single sentence would conceal conflicting requirements, that is a reason to produce alternatives; if duplicates are harmless, it is a reason not to. The scope of production is therefore conditional on use: the same records may require one entry in a summary and several in a review that must track their independence. Thus the output count is subordinate to the preserved distinction, and the distinction is subordinate to the task actually being answered.
47.4Check support and omitted distinctions
Having produced alternative outputs, the protocol turns to two separate checks. The first asks which claims in each output are carried by the assigned material: the known source version and whatever the committee's declared remit licenses. Support is claim-specific. A claim may rest on the source version yet fall outside the remit, or sit inside the remit while that version does not establish it. Either case marks a gap in the output rather than licensing a silent inference.
The check therefore attaches support to propositions, not to the presence of references. For the present audit, the useful point of comparison with ALCE (Gao et al., 2023) is the question asked of each citation: what claim does its passage support? The benchmark does not certify every reference, so each claim needs its own basis identified.
The second check asks what the output has dropped. Omitted distinctions matter when a reader could act differently if the distinction were visible. In the prospective later-hours design, a summary-plus-links baseline may support a conditional recommendation, yet the disputed removal of a staffing condition leaves unclear which version of that recommendation is before the committee. The dispute does not make every claim unsupported; it leaves the condition's status unresolved.
Together the checks yield an inference neither yields alone: supported claims and adequate disclosure are different tests. An output can be supported in everything it asserts and still mislead by suppressing a contested condition, since the reader cannot tell which version of the recommendation is before them. Exposing the omission does not settle which removal is correct.
A labelled conditional supposition shows the limit. If the disputed removal affected no claim the output makes—if the recommendation were stated without reliance on the staffing condition—the distinction would be material to the record rather than the output text. Where the recommendation does rely on that condition, omission becomes a substantive failure route. Support is necessary for each claim; disclosure of the contested condition is necessary for the output's fitness; neither is sufficient alone.
The two questions fixed earlier — which output claims the known source version supports, and which material distinctions the summary-plus-links baseline omits — are judgments about the record, not properties that a single score can proxy. A citation count compares how many references are attached to the output; it says nothing about whether any particular claim is supported by the cited material, and an omitted distinction leaves no citation to count. A consistency score compares claims with one another or with the source for agreement; an absence is not a contradiction, so the disputed removal of the staffing condition can pass unnoticed.
Neither indicator is sufficient for the judgment it resembles. Attached citations need not bear on the claim in question; internal consistency is compatible with a silently dropped condition. Nor is either indicator necessary in a straightforward way: a single passage may support a claim regardless of how many citations are listed, and an omission may be flagged without disturbing the output's coherence. Any comparison among indicators therefore holds the task and inputs fixed — the same source version and the same declared remit — and varies only what is being recorded.
Suppose both measures look favorable in the later-hours case while the committee's remit remains silent on whether staffing is material. The favorable measures would not resolve the question; they would merely fail to display it. Such cases belong in a separate record of unresolved items, which names the disputed distinction, the source version consulted, and what remains undecided, rather than folding the case into either aggregate. The register also fixes the scope of the check: it records only what the current source version and declared remit leave open, so that later revision can close an item without reopening the whole output. Keeping that register distinct preserves the open questions for the assignment of decision authority and challenge that follows.
47.5Assign responsibility and contestation
The checks so far specify what a record must contain; they do not say who carries the resulting duties. In the illustrative later-hours case, take the disputed removal of the staffing condition. Responsibility for the edit attaches to the actor who removed the condition, not to the committee at large: the record entry names that actor, the source version edited, the condition removed, and the reason offered. Responsibility for keeping the entry answerable is a second assignment, to the office that maintains the record, and it is not decision authority. The committee's declared remit fixes which body may finally decide the hours question; neither the editor's authorship nor the record-keeper's custody transfers that power.
A reasoned challenge needs a defined object and a defined target. It identifies the record entry and the source version against which the removal can be re-examined, states why the removal was wrong — for example, that the staffing condition fell within the declared remit and had support in the cited material — and states the disposition sought: restoration, revision, or referral to the decision authority. A challenge of that form goes to the actor responsible for the edit, who must record an answer. Refusal is permissible; silence is not an answer, and a refusal must rest on the remit or on the support check rather than on the fact that the edit stands.
Two qualifications follow. Naming a responsible actor is necessary for the route to run but not sufficient: a challenge may be answered and still fail on the merits. Nor does the route's availability make success inevitable. Supposing, as a conditional case, that the same body both removed the condition and decided challenges to its removal, the route would remain formally open while losing the independence that gives its answer weight. The design question is therefore whether the authority answering a challenge is distinct from the authorship it revisits, except where the declared remit expressly merges the two.
A complaint channel satisfies the proposed function only if it has both a reconsiderable object and a responsible actor. The object must identify the disputed output, the source version against which it was prepared, and the particular support or omission at issue. The actor must be someone entitled to reconsider that output and to record either revision or refusal. Without these two elements, a complaint can register dissatisfaction while leaving the representation unchanged and unaccountable.
Applied to the conditional later-hours question, the disputed removal of a staffing condition is therefore a proper object only when the complaint can point to the known source version and to the omitted condition within the committee’s declared remit. Naming the editor who removed it supplies the responsible actor; naming the source and omission supplies what can be reconsidered. The ordinary summary-plus-links baseline does not by itself answer the complaint, because the dispute concerns whether a stated condition was supportably removed.
This requirement is necessary but not sufficient. A named actor with a precise object may still decline revision after checking support. The decision must then record why the removal stands, or narrow the representation claim so that it no longer suggests the condition was addressed. Conversely, even a recorded decision does not establish that the original edit was correct. Contestation creates a route for a reasoned challenge; it does not guarantee a particular outcome. The channel succeeds when the challenge reaches a decision that can be traced to the scoped record, whether that decision revises the output or refuses the requested change.
47.6Revise or refuse the representation claim
When the maintained output claims to represent a source for a declared task, revision has a determinate trigger: either the source or the task has changed. Equally, the absence of a detected change does not establish continued fit. The question is whether each remaining claim is still supported within the declared scope.
Where the gap can be repaired on the existing record, revise the output and log the particular edit, the person responsible, and the source passage relied on. Where it cannot — where the record does not reach the current source or task — the representation claim must be narrowed rather than patched. Narrowing is not a softer revision: it reduces what the output is said to stand for, from the whole task to a defined part of it, and the reduction is itself an output claim that has to be recorded and contestable.
The proposed operations remain prospective. A form, graph or named role shows that inspection, challenge and revision are intended channels; it does not show that any inspection occurred, that a challenge altered a maintained entry, or that the designated person exercised the assigned authority. If one supposes that the inspector cannot read the source passages cited in a challenge, then the route is nominal, because the contested entry cannot be checked against the record. That is a possibility of the design, not an inevitable outcome of every such design.
Comparisons must fix the task and the inputs before the encodings are varied; otherwise a difference attributed to encoding also includes changed workload. The same caution governs a named quantity such as coverage: it compares supported claims with claims made under one scope, not the output's accuracy against the source.
Refusing an overstated mandate does not require refusing every output. The refusal blocks a claim—that the maintained artifact speaks for the whole record, or that its edits settle the underlying question—but an index of located items or an explicitly provisional summary can still serve a declared, bounded purpose. The output then has to be described as what it is: a navigation aid, a list of candidates, or a summary whose authority extends only over the sources and task actually inspected.
The distinction between usefulness and representativeness matters because they are different conditions. For the proposed protocol, a bounded purpose, access to the relevant sources, and authority to inspect, challenge, and revise are necessary for a defensible provisional output; they are not sufficient, since the named operations may never occur. A form showing an edit trail, a dependency graph, or a designated role shows only that these operations were provided for, not that anyone inspected a source, challenged a particular edit, or changed the maintained output. A provisional output earns its label only if its stated coverage tracks what was inspected; a larger claim is not made true by keeping the artifact small.
The countercondition is therefore concrete: if the index is used to support a claim beyond its scoped purpose, or if the provisional summary is read as a settled account, the refusal has been undone at the level of application even though the artifact remains formally modest. Conversely, refusal becomes useful only when the output carries its own limits—purpose, covered sources, and unresolved points—alongside it. The question is not whether a graph or role can represent the record, but whether the proposed operations were actually performed within the declared scope.
47.7The conventional-workflow counterproposal
Hold the task and inputs fixed before comparing two ways of organizing the same work. Suppose the task is to maintain a document together with the sources supporting it, and the inputs are one draft, one set of source links, and one disputed edit. Both workflows receive identical materials; what differs is how the work is recorded, assigned, and checked. Any difference in outcomes can then be credited to the arrangement rather than to better inputs, though this constructed comparison does not establish arrangement as the only possible cause.
The conventional alternative is a competent editor working through an annotated document. Marginal comments mark the passages in dispute, source links stand beside the claims they support, and the editor accepts, revises, or refuses the particular edit while leaving the reason on the page. The annotation then serves as a scoped record of the challenge; the links permit anyone with access to retrace the support; the maintained draft is the output.
Beside this, the fuller proposal registers a declared purpose, scoped records, provision for plural outputs, support checks, routes for revision, and settled decision authority. Each names an operation that must actually be available — someone able to inspect the source, to challenge this edit, and to revise the maintained output. A form, a graph, or a named role does not establish that any of these operations occurred. That gap matters on both sides: the editor's annotation counts as evidence only if it records the challenge and its disposition, just as the proposed protocol secures nothing merely by reserving a slot for review.
Necessary conditions can therefore be separated from sufficient ones. Competence and access are necessary for the conventional route to work at all, but they do not make support checking inevitable; a careful editor can still pass over a weak citation. Conversely, the proposal's explicit failure routes cannot prevent refusals or stale records — they make such failures identifiable and attributable. The comparison thus fixes the burden of proof: a difference between the two arrangements counts only if it appears in the performance or recording of these operations on the same task.
The conventional workflow deserves to be treated as a serious rival rather than a ceremonial baseline. A maintained document with annotations, source links and a revision trail, tended by someone able to inspect and amend it, already supplies much of what the fuller protocol demands. Extra structure therefore needs a demonstrated task need: a graph of dependencies, a richer record taxonomy or additional role names earns its place only when the conventional arrangement cannot perform some operation the task requires.
The test is operational. The counterproposal requires that a designated person be able to inspect the source, challenge a particular edit, and revise the maintained output. Those are proposed operations, and each must be possible and attributable. As the paired unit notes, assigning responsibility is not sufficient for the corresponding activity. Nor is a source link sufficient for inspection: it makes inspection possible, while leaving whether inspection happened an open question.
Distinguishing the two conditions also clarifies the failure routes. A structure that permits challenge does not make challenge inevitable, and a missing trace makes verification impossible rather than merely inconvenient. What a count of records, links or annotations compares is the extent of documentation, not the quality of revision or the exercise of decision authority. Any such comparison should fix the task and its inputs, leaving only the encoding of the workflow to vary; otherwise differences attributed to structure are differences in the work being asked.
Applied this way, the conventional workflow is neither dismissed nor overbuilt. It is the baseline that additional structure must outperform for a specified task need.