4.1Finite attention and discovery
Consider a hypothetical archive of six reports. The task is to find reports about entering a fictional reading room. R01 records that the north door opened late on Monday morning, and R02 records that the door-control unit failed on Tuesday evening and the north door did not open. The remaining reports concern a failed reading lamp, a cold afternoon session, a catalogue entry pointing to the wrong shelf, and a cleaning cart returned late. For this task, R01 and R02 are stipulated as relevant, and the corrected entry index lists exactly R01 and R02. These assignments define the constructed example; they are not measured findings about readers.
The conditional benefit is a reduction in candidate reading. If the corrected index reliably identifies the reports relevant to this task, a reader can restrict the first inspection to two candidates rather than reading all six before deciding what counts as relevant. The set that must be inspected shrinks from six to two, so four reports need not be read first. Whether that smaller set adequately serves the task depends on its coverage of the relevant reports. Reducing the number of candidates is not enough if the reduction omits information the reader needs.
Using the index also involves work. A reader must consult it and decide whether its selection criterion fits the task. Grounds for reliance might come from a prior assessment or from checking the original material; a fresh examination of every report need not occur on each use. If this reader must read all six reports to validate the selection, the smaller final candidate set has not removed that initial reading. The distinction matters when comparing discovery arrangements: who checks the index, and when, affects the work that remains for later readers.
The example establishes a candidate count, not a measured time saving or total-effort comparison. Its benefit for an actual reader would depend on the work needed to use and assess the selection.
Now consider a faulty variant of the same invented index. It selects reports labelled 'entry' and therefore returns R01 alone. R02 remains in the original six-report archive, but its assigned label is 'equipment'. The change is in the selection rule, not in the text or stipulated relevance of R02.
The difference between the label and the report's content matters. R02's text records that a door-control unit failed and that the north door did not open. The equipment label captures one aspect of the report but does not exhaust its relevance to a question about entry. The record itself concerns entry into the room, and it is therefore relevant under the stipulated task even though its category is not. A selection rule that follows labels alone excludes a report stipulated as relevant.
Two safeguards can expose such an omission. First, a retained original archive lets a reader move beyond the index and inspect what the labels summarize. Second, a challenge mechanism lets a user contest a selection and ask for revision. Either route can bring R02 into view, provided someone exercises it.
Yet the availability of originals makes checking possible; it does not guarantee that anyone checks, nor that every omission is discovered. A reader may accept the index as authoritative, a challenge may never be raised, and the missing report may persist in practice. The safeguards widen the opportunity for correction without determining its outcome. A challenge also needs to reach someone able to review the selection. Retaining records and allowing objections support possible revision, but do not establish that the index will be corrected.
4.2Reducing duplication without erasing reasons
A repository containing two hypothetical submissions can reduce visible duplication without collapsing the submissions themselves. Suppose records Q01 and Q02 each contain the sentence "Could the reading room open until 20:00 on Wednesdays?" in relation to the same reading room and the same upcoming Wednesday schedule. The interface may present that wording once as a navigational anchor, accompanied by a submission count of two and links to both records. Nothing is deleted or merged underneath: each record retains its identifier, its provenance metadata, and whatever other information the collection holds about it.
The grouping asserts a limited textual correspondence. It does not convert a question about later opening into an endorsement of that change. Repeating the sentence in a list would likewise supply no additional reason for adopting it merely by repetition. With both records and their links retained, the reader can inspect the entries separately when context matters. Whether the compressed display makes those links sufficiently noticeable is a further design question; preservation underneath the display does not guarantee equal attention in use.
The count of two is likewise a fact about submissions, not a translation into people or endorsements. Two records do not establish that two distinct individuals wrote them, that either author endorsed the other's message, or that the proposal enjoys any particular level of support.
The exact-match check that licenses this grouping is only a comparison of surface text. It can be performed automatically and reported as a textual correspondence. Its result concerns matching wording, while any substantive interpretation remains a separate task. Whether the two submissions share an intention, whether one borrowed from another, and whether they should be counted separately in some substantive assessment are questions requiring evidence beyond the match itself. Navigation grouping therefore compresses display while leaving those questions open, and preserves a route back to each record whenever a reader needs to inspect its context.
The same collection also contains two conditional questions. Q03 asks for later Wednesday closing if paid staff are available; Q04 asks for it provided no additional unpaid work is required. Both concern the same room and schedule. A shared heading can place them together, but cannot replace either condition with the other or present both as an unqualified request.
The distinction is between a record and what it warrants. As 04.2.1 showed, record existence establishes no more than the fact of submission. Likewise, the fact that Q03 and Q04 carry different conditions is a property of their wording, not evidence that their authors disagree, and certainly not evidence about whether staff are actually available or whether extra unpaid work would in fact be required.
The conditions are the very parts that block merging. Availability of paid staff and the exclusion of additional unpaid work are separate constraints, and they are compatible: both could hold, only one could hold, or neither could. Treating them as contradictory manufactures a conflict the records do not contain.
Retaining both conditions preserves distinct checks. Q03 directs attention to whether paid staff can be assigned; Q04 directs attention to whether the arrangement would add unpaid work. Neither condition is established merely by formulating a check: an availability inquiry can return a negative answer, and a workload review can reveal added unpaid duties. For this navigation task, a parent heading can group the related records while leaving both conditions visible or readily inspectable. In that arrangement, Q03 and Q04 can share a heading while each remains answerable on its own terms.
4.3A map for later reading
A summary can do more than shorten a body of material: it can function as a map, provided the categories it preserves point to specific source material that a reader can then follow. The map's value lies not in its compactness alone but in whether its labels lead somewhere inspectable.
Return to the corrected entry index, which points to R01 and R02 in the invented archive. A reader can follow those pointers to examine the reports rather than rely only on the category heading. That route matters if the reader needs the details of Monday's late opening or Tuesday's door-control failure. The category helps identify where to begin; the linked text supplies the actual account to be read. A summary that displays the heading but drops the pointers offers a different form of assistance, because it does not provide the same direct route back to the selected material.
The example also clarifies what a category is. A category is a description attached to a record; it is not itself an inspection of the record, and it is not a claim that the record is accurate. That R01 reports a late opening is a claim about its contents. That the door actually opened late is a claim about the event; the archive's mere existence does not establish it. Similarly, a working link in the corrected index establishes where R02 is located, not that R02 is accurate and not that the category assignment is complete. The index makes later reading possible; the reading itself remains to be done, and its results remain to be assessed.
Suppose a later reader receives the corrected entry index for the six-report archive described above as an invented illustration, but cannot open the linked originals. In that circumstance, the map cannot support the full inspection promise associated with access to the records themselves. A link displayed alongside an entry shows where an original was said to be located; it does not transmit that original’s wording to the reader. Without the records, the reader cannot compare a summary with its source, examine context omitted from the summary, or decide independently whether a record belongs under the assigned heading.
The index may still provide orientation by naming candidate records and the criterion used to select them. If it also carries the brief descriptions given in this example, a reader can see why R02 was included despite its equipment label. That is a basis for provisional reliance on the description, not independent confirmation of the report's wording. If the descriptions are absent as well, only the identifiers and visible categories remain available. The useful question is what this particular reader can inspect, rather than whether a link exists somewhere in the system.
The narrower claim is that the index can help a reader identify material to request or a selection decision to question. Its accuracy may be supported by other evidence, but cannot be checked against these inaccessible originals by this reader. Access might later permit that comparison, without guaranteeing that the reader undertakes it or accepts its result. The summary should therefore disclose the access limit where it matters to the intended use. Doing so preserves a possible navigation benefit while leaving the evidential gap visible.
4.4Comparison across many submissions
Consider two hypothetical, invented proposals for a library's evening service. P1 would extend Wednesday closing to 20:00 on the condition that a paid shift is available. P2 would retain the current closing time and instead offer reserved seats within existing opening hours. Placing them under a common set of dimensions declared in advance makes comparison possible without requiring that either proposal be judged in advance.
Three dimensions are declared here: closing time, stated staffing condition, and way of obtaining access. On closing time, P1 proposes an extension, while P2 leaves the hour unchanged. On staffing condition, P1 states a conditional requirement, namely the availability of a paid shift; P2 states no new staffing condition; keeping the existing hours does not prove that reservation handling needs no additional work. On obtaining access, P1 offers access during later opening, while P2 offers access by reservation within the existing schedule. These dimensions organize the terms each proposal states. They record what each submission proposes and what condition each attaches; they do not establish whether any condition will in fact obtain. A claim that a paid shift will be available would be a claim about the world, and it has not been tested here.
The arrangement lets the reader ask more focused questions. Would the paid shift required by P1 be available, and what work would the reservations in P2 require? Neither answer is supplied by putting the proposals side by side. If P1's condition fails, its proposed extension cannot proceed on the stated terms; that does not imply that existing access disappears. P2's reservation route also cannot be called the only possible way to obtain access merely because it is the other proposal listed here. The comparison exposes different matters to investigate without narrowing the world to the two entries.
The three dimensions are proposed for this comparison, not demonstrated to be exhaustive, and declaring them does not prove that either proposal is feasible. The comparison organizes stated terms; it neither measures nor assigns a winning score.
The three declared dimensions leave another possible concern unrepresented. In the same hypothetical comparison, a contributor asks that an existing quiet area remain available during any new service. Closing time, the stated staffing condition and the way of obtaining access do not, as defined here, specify what happens to that area.
The concern can receive an explicit place beside the original dimensions. The entry might state the request and mark the effects of each proposal on the quiet area as not yet established. Treating it as zero on an unrelated scale would conceal the missing information; inventing a favourable score would be no better. A reader could then ask what each proposal would change in the room and which arrangements might preserve the quiet area. Those are questions raised by the additional entry, rather than conclusions produced by recording it.
The omission of quiet-area availability from the original dimensions does not by itself prove bias or bad faith. Nor does choosing to record the objection show agreement with the contributor or settle the issue. The shared framework is limited to the terms it declares, and adding a concern as a separate entry acknowledges that limit without pretending to resolve it. In this use, a comparison table organizes stated terms and identified objections; it does not decide between the proposals.
4.5The labor shifted to editors
An edited summary may lower the effort a reader needs to understand a record, but that reduction is only one component of the work involved. Four costs must be kept separate: initial preparation, maintenance after the underlying record changes, contributor clarification when an editor cannot determine what the record means, and the reader’s effort in interpreting the summary. Preparation occurs before publication; maintenance extends across later revisions; clarification may occur at either stage and involves at least two participants. Reader effort is incurred whenever the summary is used. These categories can overlap in time without belonging to the same task, so an editorial session that combines drafting and clarification must be allocated before it enters a total.
Any total therefore needs a declared unit, a common measurement period, and explicit allocation rules. The unit might be person-time, but no particular rate is supplied here. The period must cover initial preparation and subsequent updates, or else state that it measures only one phase. Clarification work should be divided between editor and contributor according to who performed it, while a joint exchange may require recording both parties’ effort. Maintenance should be assigned to the revision that prompted it, rather than counted again as preparation. These rules prevent the same activity from appearing in two categories merely because it serves more than one purpose.
Consider a hypothetical edited digest whose readers report that locating relevant material became easier. That observation supports a claim about reported reader effort; it does not by itself establish lower total effort or a net saving. Such a conclusion would additionally require measurements of preparation and maintenance, plus any contributor clarification. The necessary measurements are not supplied, so the stronger inference must remain conditional. Even a valid reader-effort reduction becomes a total saving only if the other costs over the specified period do not offset it.
A single prepared summary can serve many readers, so its preparation cost is shared rather than incurred afresh for each consultation. That arrangement can make compression useful at scale, but sharing alone establishes no conservation of effort. Time not spent by one reader is not thereby converted into editor time, and the saving realized by a later reader need not equal that realized by the first: familiarity with the project, the reader’s purpose, and the parts consulted can all alter the effect. Any claim of net benefit therefore remains a conditional projection until both sides of the comparison have been measured.
For the invented archive, the same index could be consulted by several readers after one preparation effort. That possibility changes the comparison with each reader beginning from the full collection. It does not establish how much any reader saves or whether preparing the index was worthwhile. Some readers may need to examine every report even after consulting it; others may use only one category. An update to the archive may also require work before the prepared index can be reused. Sharing one representation is thus a feature of the arrangement whose consequences still need measurement.
Establishing net benefit relative to full reading would require a measurement design rather than a narrative account. In one effort unit over a defined period, it would need to record initial preparation, maintenance after changes, contributor clarification, and reader effort, with overlapping tasks assigned once so that they are not counted twice. It would also need observed uptake: how many readers used the summary, which portions they used, and what they would otherwise have read. Comparison with a defined full-reading baseline would then assess whether reduced reader effort exceeds the additional preparation, maintenance and contributor effort attributable to the summary arrangement. Observed reuse helps establish who actually used the summary, but neither reuse alone nor an assumption that future readers will resemble earlier ones proves net benefit.
4.6When full text is the simpler solution
Consider a hypothetical collection built to a different specification from the six-report archive. A folder contains two notices, printed in ordinary type and placed adjacent to one another: one announcing that the reading room is closed Wednesday for cleaning, and one announcing that it reopens Thursday at 09:00. The stated task is to find the reopening time. No navigation barrier, formatting defect, or access obstacle is stipulated.
Under these conditions, reading the folder is already sufficient to answer the question. The second notice contains the information directly, and the task requires no extended search or reconstruction from scattered passages. The reader still identifies the relevant notice and interprets its wording. A summary might reproduce the reopening time of Thursday at 09:00, but it would add another representation to maintain without removing any obstacle that the task presents.
The answer established by that reading concerns the stated reopening time. Whether the room actually reopens then would require evidence about the event. This distinction leaves the original lookup intact: the task here is to find what the notices announce. An additional representation could have another purpose, but should be assessed against that purpose instead of being presumed useful merely because it is shorter.
What this example establishes is narrow: there is at least one case in which the original text is brief, organized, and directly responsive to a stated task, so that compression is not needed to make the answer available. It does not follow that every short collection needs no assistance. Two notices might still require translation, an accessible presentation, or a summary for a reader who cannot use the format in which they arrive. Nor is any claim made about how quickly a reader finds the answer; no reading time has been measured. The point is only that, where the text already supplies the requested item on its face, an added representation must justify itself by solving some further problem.
Now add a synopsis and a revision record to the same two-notice folder. The notices are already adjacent and readable, and the assigned task remains finding the stated reopening time. The proposed synopsis can repeat that time in a shorter form, but the example supplies no navigation obstacle that only this extra representation would remove.
The synopsis repeats the reopening time from the second notice, adding one copy of that statement. The first notice only announces Wednesday's closure. These are two current presentations of the time; a revision note may separately retain a history of changes. If the Thursday time changes, the relevant original notice and the maintained synopsis need to be brought into agreement, while a separate revision record documents the change. The closure notice need not change unless the closure information also changes. Relative to keeping the two original notices current, the extra representation introduces a synchronization obligation. If inconsistent versions remain visible, a later reader may also need to establish which one reflects the current announcement.
This case does not show that summaries are generally harmful, and the added obligation is not a measured net cost. It shows only that, under stipulated conditions of adjacency, readability, and a simple lookup task, summary maintenance can arise without any corresponding reduction in reading. The comparison could change sharply if the notices required translation, accessible presentation, or simplification for readers who cannot use the notices as written. Nothing in this example establishes that two notices never need such treatment; those needs are simply absent here, so their absence is part of the case's scope rather than a general conclusion about brief documents.
4.7Declaring the purpose before the method
Compression is not a single operation with a single ideal setting. The amount and form of compression that serves one task may undermine another. It is therefore better to declare the intended use before selecting a method than to choose a compression technique first and hope that its output proves useful. Three purposes merit distinction: finding relevant records, comparing proposals, and making a decision. Each asks a summary to support a different activity, and those demands should shape the summary in advance.
Finding relevant records asks whether the summarized material is worth retrieving and examining in full. Here compression may emphasize coverage, salient terms, and enough context to guide a search. A record may contain claims about itself—its date, author, or stated subject—and claims about the world, such as an asserted outcome or cause. A finding-oriented summary needs to preserve enough of both kinds to support follow-up without committing to their truth.
Comparing proposals asks how alternatives relate on dimensions that matter to the comparison. A summary for this purpose should preserve comparable fields, handle differences in scale or framing, and expose gaps where one proposal is silent. It need not endorse a proposal or resolve conflicts among them.
Making a decision asks which course to take under stated goals and constraints. A decision-oriented summary should make trade-offs visible and identify what would count as a reason to prefer one option. Even so, a useful finding or comparison tool does not by itself decide for its users. A summary can expose trade-offs without establishing who has authority to decide. The person who commissioned it and the person entitled to make the choice need not be the same.
The declaration is provisional rather than permanent. A record found for retrieval may later become one item in a comparison, and a comparison may feed a decision not yet framed. Changing purpose calls for revisiting the compression, because what was safely omitted may become material.
A purpose need not be a fully specified decision task in order to discipline a summary. An exploratory index may be built before anyone knows what later choices it will inform, provided that its immediate purpose is declared as discovery. That declaration is not a placeholder standing in for rigor. It names a real job—organizing records so that patterns, gaps, and anomalies become inspectable—and it excludes other jobs for which the output has not been prepared. Provisional purpose is still purpose.
The declaration should carry enough scope information for the proposed discovery use. Where relevant, this includes the records covered, known exclusions or incomplete entries, and the transformations used to produce the fields. The needed detail depends on the inferences a reader is being invited to make. The index may relay claims about the world, but their warrant depends on the underlying evidence and how it was transformed, rather than on the index being compact. A cluster in the index is evidence about the indexed material and its construction, not yet evidence that the corresponding pattern exists beyond the sample or that it predicts any outcome. Because discovery is exploratory, candidate patterns remain provisional unless a separate inquiry tests them.
When a later use becomes specific, the declaration must be revisited rather than merely reused. The new decision task requires its own assessment: whether the index's records cover the population at issue, whether its fields are adequate and reliable for that task, whether omitted information matters for the choice at hand, and whether the index's construction introduced biases that become consequential once stakes attach. Validation appropriate to discovery may be insufficient once a decision depends on the output. Limits recorded during the exploratory phase are therefore inputs to this reassessment, not guarantees that survive a change of use.
A compact representation can make structure visible and retain an account of its sources. Compactness alone does not establish that the structure supports a later inference. The next question is therefore more specific than whether a summary is short: which distinctions, qualifications or routes back to evidence would a particular use lose, and which could it safely leave aside?