5.1Missing propositions
Consider an invented original containing two assertions: that the reading room closes at 18:00 and that the north entrance has a step. A summary reports only the first: that the reading room closes at 18:00. The omitted proposition is that the north entrance has a step. Comparing the texts locates a specific absence; the summary itself neither mentions the entrance nor indicates that this second assertion was present.
The two texts share one proposition but differ in content. Retaining the closing-time assertion preserves that particular claim; it does not preserve all of the original’s content. The omission is also different from a contradiction. The summary does not assert that the north entrance has no step. It says nothing about that entrance. A reader who turns its silence into a negative answer would add a claim that neither the retained sentence nor the comparison supplies.
This comparison concerns what the texts assert. It does not independently verify either the closing time or the entrance description. Nor does the omitted assertion, even if accurate, establish that every entrance is inaccessible. Its content concerns a step at one specified entrance. Naming the lost proposition at that level avoids replacing a precise textual difference with a larger claim about the building.
An omission record can therefore identify the original assertion, its location and its absence from the summary without yet deciding whether the summary fails its intended use. Locating the loss and assessing its consequence are separate operations. The first has a determinate answer in this small example. The second requires a purpose against which the retained and omitted content can be assessed.
The same summary can now be tested against two tasks. For a lookup of the stated closing time, its retained assertion supplies the requested information. For a question about the step at the north entrance, it does not. The texts remain fixed while the information needed from them changes.
Take the task of finding the stated closing time. The retained proposition answers it directly: the reading room closes at 18:00. The omitted assertion about the north entrance plays no part in this task, so its absence causes no loss of the sort the task cares about. The summary, as far as this application is concerned, is adequate. Adequacy here is a relation between content and purpose, not a property the summary carries wherever it goes.
Now change the task to checking whether the described north entrance has a step. The same summary supplies no yes-or-no answer to that entrance question. Nothing in it says whether that entrance has a step, and the missing proposition was precisely the one the question targets. The identical omission that was inert under the first task becomes decisive under the second. This is the sense in which scope must be declared: a summary may be fit for one application and unusable for another without contradiction.
The result is local to these lookups. Other uses might require the date of the announcement, the authority behind it or details not stipulated here. None of those facts can be supplied by declaring the summary adequate for closing-time lookup. Equally, failing the entrance question does not make the retained closing-time assertion inaccurate. The useful assessment names both the task and the missing information, so that a later reader can distinguish a change of purpose from an error in the content that was actually retained.
5.2Collapsed distinctions
Consider two stipulated fictional contributors to a discussion of the reading room, whose current closing time is 18:00. Both request that the room stay open until 20:00 on Wednesdays. Contributor A gives one stated reason: a class ends at 19:00, so its students cannot enter before the current closing. Contributor B gives another: a work shift ends at 19:00, so its workers cannot enter before the current closing. The requested action is identical, and so is the timing problem each describes. Yet the reasons concern different activities, and their timing conditions can change separately. The example need not assume that the student and worker groups are disjoint.
Now suppose the class is moved to end at 17:00 while the work shift is unchanged. The class no longer ends after closing, so that particular timing obstacle is removed. Whether its students can actually reach the room before 18:00 also depends on facts, such as travel time, that the example does not supply. The worker-related obstacle survives intact. A recommendation that looked uniform at the level of action therefore divides at the level of justification. One supporting reason has lost its purchase on the situation; the other continues to apply exactly as stated.
That divergence is informative, but only up to a limit. A's reason losing force does not prove that A now opposes later opening. A may continue to support the request on grounds not yet stated, or may judge that the earlier end to class leaves the request worthwhile for other purposes. The stipulation records only the stated reason, so any stronger attribution goes beyond it. Symmetrically, the persistence of B's obstacle shows that B's stated reason remains available; it does not by itself fix what B would say about every other possible change.
Thus a shared recommendation does not settle how its supporters would respond when conditions shift. Here the stated conditions distinguish the two reasons even though the requested action is shared. They do not reveal either contributor’s complete position.
Separate entries can retain A’s class-related reason and B’s work-related reason. This arrangement reduces repeated wording while leaving the grounds for the request available for a different kind of reading. A person looking only for the proposed action need not first read every reason; a person examining its justification needs a clear route to them.
Those entries need to preserve which condition belongs to which reason. If the class schedule changes, the reader should be able to locate A’s original timing claim without treating B’s unchanged work schedule as though it changed too. Recording the altered class time addresses the stated after-closing obstacle; it does not establish all conditions for actual access or determine A’s current preference. The summary can keep that distinction visible without putting the entire history into its heading.
What the reader must be able to recover is therefore narrower than a complete record, but more than a bare request: the action, the distinct reasons offered for it, and the condition on which each reason turns. That is enough to notice that one stated obstacle has fallen away while the other remains. It is not enough to conclude that the first contributor now opposes later opening, since the record does not establish whether that obstacle exhausts their position, nor to establish that either contributor lacks further, unstated reasons.
A short action heading can accordingly stand in front of separately inspectable reasons, naming the request once and leaving the grounds for examination. The heading should make clear that reasons are there to be retrieved and what distinction they will show. It cannot guarantee that a reader will reach them, and it should not assume that every relevant context is already known; its contribution is to preserve the distinction if the reasons are consulted.
5.3Disappearing speakers
Three message records do not establish three speakers. In the stipulated example, M1 and M2 carry the same contributor identity, A, while M3 carries B. The record therefore holds three messages and two distinct contributors. The two counts answer different questions: how many utterances were recorded, and how many identities produced them. Neither count, on its own, establishes how much of the relevant population is covered.
The distinction matters because message count does not translate into person count. A produced two of the three messages, so reporting the three messages as three distinct contributors would inflate the contributor count even in this small, identity-stipulated record. Where contributor identity is not stipulated, message count is weaker still: it cannot tell whether one person spoke repeatedly or several spoke once. The A/B labels here preserve sameness and difference only within the example. They are not proof of a privacy guarantee, and they carry no sampling claim. Message count, in short, measures record volume; contributor count measures attributed identity; coverage would measure relation to the population.
The distinct-contributor count still needs population information before it supports a coverage claim. The relevant population is all reading-room users, but its size is not provided, and nothing establishes how the three messages were drawn from it. Two contributors could be every user, a small minority, or anything between; the counts alone cannot distinguish these cases. A coverage fraction would need a population denominator and a defined relation between the contributors and that population; neither is supplied. Nor does the presence of two identities show that the voices represented in the record match the population in age, role, or any other respect.
Accordingly, the warranted statement is narrow: the record contains three messages by two stipulated contributor identities. Claims about population coverage require separate facts about the population and about how contributors entered the record. Until those are supplied, coverage remains undetermined.
When a collection is opened to readers under a privacy constraint, the analyst need not choose between naming every contributor and erasing every speaker relation. One possible arrangement withholds names while preserving stable distinctions among contributors inside the collection. In the stipulated example, messages M1 and M2 carry contributor A, whereas M3 carries contributor B. Assigning A and B as persistent labels keeps M1 and M2 together and keeps M3 apart, so the collection still supports questions about repeated contribution and contrast without disclosing either name.
Accordingly, a reader can use the labels to answer a relation question: whether M1 and M2 come from the same contributor is yes, and whether M3 comes from a different one is yes. The labels do not answer who A or B is, nor whether these records represent reading-room users generally. Withholding the names is a narrower operation than guaranteeing anonymity. A stable label is a relational identifier within the example: it records that two records share a contributor and that one differs. It does not show that readers cannot re-identify anyone, that the labels would be safe under linkage with outside information, or that every speaker context can be discarded. Context may matter for interpreting who could have contributed, and omitting it can change what a record can support. The labels therefore should not be treated as a privacy certificate.
The choice of what to retain can therefore be made relation by relation. For a reader tracing repeated contributions, the within-collection A/B distinction does useful descriptive work. Public names need not accompany that distinction in the display. Other tasks may need context that the labels do not convey, and the earlier coverage question remains unanswered. A proposal to withhold names should specify which relations remain visible and why, then examine the disclosure risks and interpretive limits of that particular arrangement.
5.4Flattened stakes and urgency
Consider an invented scheduling process in which the original request asks the reader to book the room by Thursday at 12:00 for a talk on Friday at 17:00. A summary of that request asks only that the room be booked for the Friday talk. The summary retains the object and the general occasion, but drops the booking deadline. What remains looks like a complete instruction, yet it no longer conveys the deadline for pursuing the stated option.
The omission of the booking deadline matters because the deadline is not decorative. In this example, the booking window for the talk closes Thursday at 12:00. Booking by that deadline is a necessary condition for using this stated route. The deadline alone does not guarantee a room assignment if a request arrives in time; capacity, acceptance or other conditions have not been specified. Without the deadline, a reader can mistake a time-bounded option for an open one. The practical question becomes simply whether the room can be booked, rather than whether it can be booked in time. A reader relying on this summary alone has no stated booking cutoff against which to assess a proposed delay. The omission need not mislead every reader, but it removes a timing condition available in the original.
The loss is therefore not merely a shortage of detail. It can change a reader’s assessment of which responses remain practical by concealing the timing condition. Restoring the deadline restores that discrimination: the request identifies the scheduled time of the talk and the deadline for the booking action.
This example supports a limited conclusion. The Thursday deadline affects this booking option; it does not rank the request above other requests. Nor does it show that the Friday event becomes impossible under every alternative. Another arrangement may still be possible. What is concealed is the condition attached to the stated Friday-room option, not the status of the event as such.
The booking deadline establishes a condition for pursuing one option. It does not guarantee that acting in time secures a room, and it does not determine which request deserves the room if several compete. These are different questions even when the decision must be made quickly.
That is a claim about timely feasibility, and it should not be allowed to migrate into a claim about priority. Whether this request can still be made in time is one question; whether it should be served before competing requests is another. The deadline helps answer the first when the current time is known; it supplies no answer to the second by itself. An urgent label may tempt the reader to treat the second as already settled, but the label adds no ranking relation that has not already been stated.
Nor does urgency become an automatic priority merely because no alternative arrangement is available. Losing the sole way to hold this talk could be a serious consequence, but its weight relative to other requests would still need justification. Where another arrangement is possible, that fact may alter the assessment without making the deadline irrelevant. The example leaves both the alternatives and the governing allocation criteria open; neither should be invented by the summary.
Where competing requests are present, allocation therefore requires considerations the deadline does not provide: the terms on which the other requests were made, the resources they require, and the windows still open to them. The deadline's contribution is more modest but still concrete. It marks the point after which this option is gone, and so it tells the decision-maker when to act if this option is to be kept available at all. Treating the urgency label as a trump card overstates what has been stipulated; treating the deadline as weightless understates the window it genuinely closes.
5.5Lost uncertainty
Consider an invented repair forecast: repairs may finish on Friday, with inspection pending. The first clause presents Friday completion as possible. The second reports the status of an inspection. A summary that instead states that repairs will finish on Friday strengthens the predictive commitment and removes the inspection status. Nothing supplied with the forecast warrants either treating the inspection as completed or presenting the more definite prediction as the source's wording.
The stronger sentence remains a prediction about a future event. It does not become an established outcome merely because “will” replaces “may.” Calling it a settled fact would therefore obscure the particular change made by the summary. The source allowed Friday completion without asserting that it would occur; the summary makes that assertion. Neither sentence supplies a numerical probability, and the example provides no later outcome with which to assess the forecast's accuracy.
There is also a limit to what can be inferred from the two original clauses. Their juxtaposition does not specify whether inspection determines the completion date, checks completed work, or serves some other purpose. An editor can preserve both the tentative prediction and the pending inspection without selecting one of those explanations. Assigning the inspection a decisive causal role would add another unsupported commitment.
The local loss thus concerns the strength and content of what is asserted, not a measured forecasting error. A shorter version should be examined for those changes before its smoothness is taken as evidence that the uncertainty survived.
A shorter uncertainty expression is not automatically a loss of uncertainty, and a verbatim requirement is the wrong test. What matters is the confidence asserted: does the shorter wording present the claim with the same degree of commitment that the original qualification warranted? Retaining the relevant qualification is one condition of adequacy for this comparison; other content may still matter for a different task. The comparison should therefore run between the qualification carried by each wording, not between their lengths or their shared vocabulary.
Returning to the repair forecast from 05.5.1, any compression must be checked against its tentative modality and its separate report of the pending inspection.
A proposed shorter wording preserves uncertainty and mentions inspection: Friday completion remains uncertain pending inspection. It may, however, suggest that inspection is what will settle the timing question. The original does not specify that relationship. A more conservative compression keeps the two claims separate: Friday completion is possible; inspection is pending. This formulation avoids inventing a confidence percentage. It is still a candidate for the stated comparison rather than a guarantee of identical interpretation in every context.
The check therefore concerns the commitments a reader is invited to take from the words. It need not demand verbatim repetition, but it should ask whether a modal qualification became stronger, whether separate claims became conditional on one another, and whether relevant information disappeared. When the relation between two claims is unknown, brevity is a poor reason to supply one. Retaining their separation can be more faithful than producing a smoother sentence with an unsupported connection.
5.6Lost routes to correction
Accuracy at the moment of composition is not a channel through which later information can travel. A summary may faithfully report what was then announced and still become unsuitable for a current closing-time lookup once the announcement changes. The problem is not that the earlier wording has turned false. It remains an accurate report of a past announcement. Its past accuracy does not establish a route by which a later correction can reach people reading a reused copy.
Consider an invented notice process. Notice N1, version 1, accurately states the then-announced closing time of 18:00. The announcement later changes to 19:00. A copied summary identifies neither its notice version nor a responsible reviewer. The copy supplies no direct account of which announcement it reflects or how a reader should report a discrepancy. Other evidence, such as its wording or a separate publication record, might help identify its origin; the missing label does not make reconstruction logically impossible. The difficulty is that the correction route is left unstated in the copy.
Historical accuracy and current suitability can then come apart. Preserving the old notice as a record of what was announced is compatible with marking it as superseded for present lookup. Replacing every historical occurrence with 19:00 would erase that distinction rather than solve it. What needs to change is the representation offered as current and, where appropriate, the way an older representation directs readers toward the later announcement.
A later correction also has to reach the places where the old information is still offered for use. Accurate copying at the first publication does not establish that this later step will happen. A reader may continue to encounter the 18:00 summary even after the maintained notice has changed. The correction problem therefore concerns the relations among versions, current displays and the people able to update them, not a retroactive claim that the original announcement never existed.
For this proposed notice arrangement, a correction process needs to identify the text at issue, direct the issue to a responsible role and carry accepted changes to maintained copies. Those functions can be described without prescribing one technical implementation. An explicit version identifier is one way to distinguish revisions; another reliable identification method could serve that function. A contact address may provide a point of receipt, but by itself it does not establish what happens after a message arrives.
In the N1 example, the proposed route would first determine which announcement the copied summary reflects. The reviewing role would then consider whether the copy is presented as a historical record or as current information. That distinction affects the remedy: a historical version can remain intact with its status identified, while a maintained current summary may need the new closing time. The process should also specify which copies it actually maintains rather than imply control over every copy a reader might have made.
Naming these functions describes a design, not its operation. Someone must actually receive the issue, assess it and make the accepted update in the maintained locations. Evidence of that activity would be needed before claiming that a particular correction had propagated. The role receiving a report need not be the person who changes every copy, but the handoff cannot remain an unexplained assumption. A process description should make it possible to ask where an accepted correction stopped and which maintained displays still use the superseded information.
5.7Tradeoffs without one magic score
A single score for communication risk is tempting because it would make comparisons quick. But the losses involved here are not naturally commensurable. Losing a distinction between reasons differs from losing a contributor relation or a route for updating a reused summary. For that reason, this manuscript proposes six analytic dimensions—propositions, distinctions among reasons, speaker relations, stakes and timing, uncertainty, and routes to correction—as a way to organize review questions, not as a validated measurement instrument.
Each dimension suggests a different question. Which assertions survive the summary? Can distinct reasons for the same action still be recovered? Which contributor relations remain visible? Are relevant timing conditions or consequences retained? Does the wording keep the qualification of an uncertain claim? Can a reader identify and pursue an appropriate correction route? These questions concern different features of a representation, and a satisfactory answer to one supplies no automatic answer to the others.
These questions cannot simply be added together. Losing clarity about a proposition and losing a route to correction are heterogeneous losses; summing them would require a declared measurement rule stating how one kind of harm trades against another. Without such a rule, a total obscures more than it reveals. Nor do these six dimensions claim to exhaust every possible loss a communicative situation can generate; further dimensions may be warranted by cases not considered here.
The practical implication is modest but useful. Rather than asking whether a statement is good or bad in the aggregate, reviewers can ask which dimensions matter for the decision at hand, how each fares, and where the most consequential failures cluster. Keeping the answers separate makes the losses available for comparison; deciding what to trade still requires priorities.
Suppose a comparison is reduced to a weighted sum. The following is an invented two-dimensional illustration. Two loss flags are tracked for each candidate: whether a requested closing-time statement was omitted, and whether the distinction between the two contributor labels was collapsed. On these two flags, A is (0,1) and B is (1,0): 0 means the named loss is absent, 1 means present. Under the stated rule, a lower weighted sum is preferred.
Assign weights (2,1), so the omitted closing-time statement counts twice as much as the collapsed contributor distinction. Then A sums to 0×2 + 1×1 = 1, while B sums to 1×2 + 0×1 = 2, and A is preferred. Now assign weights (1,2), reversing which loss counts more heavily. Then A sums to 0×1 + 1×2 = 2, while B sums to 1×1 + 0×2 = 1, and B is preferred.
The choice of weights supplies the priority; the arithmetic derives the ranking from those weights and the stipulated loss flags. The calculation does not discover which priority ought to govern. The ranking reverses while the candidates themselves stay exactly as they are. That is what the chosen priorities contribute: an ordering of the two named losses, stated openly rather than buried inside a total.
The scope is correspondingly narrow. The binary flags and the weights are choices for this example, not empirical measurements, and the sum is not an objective total of all six loss types. A candidate with a low weighted sum may still carry losses outside the two dimensions, since the calculation never counted them. Declaring the weights and counted scope makes this ranking interpretable, but does not establish that the selected losses or priorities are adequate for a real decision. A separately imposed hard requirement may rule out a candidate regardless of its score.