On August 29, 2026, two searches launched on the same day on a single issue from 1963 produced two incompatible medians: $129.99 for the first formulation, $39.99 for the second, in the unverified segment. A ratio greater than three, on an identical object, at the same time. On the verified copies side, the difference falls to $8,499.99 against $7,949.99. The conclusion is in one sentence: a median does not describe a fascicle, it describes the query that produced it. Any figure cited without its original formulation is unusable. The amounts below are prices claimed by sellers on that day, not completed transactions, and a measurement taken on a single date does not indicate progress.
The experience is banal and its result is brutal. A booklet published in 1963, two ways of designating it in an ad search engine, the same day, a few minutes apart. The first formulation returns 149 unverified listings, ranging from $3.00 to $8,800.00, with a median of $129.99. The second returns 99 unverified listings, from $1.50 to $33,000.00, median $39.99. Nothing has changed between the two statements: neither the market, nor the booklet, nor the date. Only the typed words have changed.
This observation is not an anecdote of measurement. It ruins a very widespread habit, that of citing “the median” of an issue as one would cite its date of publication, that is to say as a stable property of the object. The data from August 29, 2026 shows that there is no median of the fascicle: there is a median of formulation A and a median of formulation B, and they are not of the same order of magnitude. Understanding why means understanding what an ad list really is — a sample constructed by a query, never a census.
Two surveys, one object, two different populations
The most instructive detail of the survey is not the difference between the medians: it is the difference between the numbers. Formulation A brings back 149 unverified ads and 45 verified copy ads. Formulation B returns 99 and 76. If the two queries targeted exactly the same set of objects, these numbers should coincide. They do not coincide, and not by a hair: fifty unverified announcements of difference on one side, thirty-one verified announcements on the other, in the opposite direction.
In other words, the two queries do not just sort the same packet differently. They constitute two distinct packages, which partially overlap without merging. The median calculated on one has no mathematical reason to resemble the median calculated on the other, since it does not relate to the same population. We compare two samples, not two measurements of the same quantity.
This observation reverses the usual order of questions. We usually ask for « combien vaut ce numéro ? »; you should first ask « quels objets ma requête a-t-elle rassemblés sous ce nom ? ». The answer to the second question entirely determines the answer to the first, and it is rarely documented. A median screenshot, shared without his query, is an orphan answer to his question.
We must add a precaution that the following will not repeat in each paragraph: all these amounts are sums displayed by sellers on August 29, 2026. No one paid these prices; These are demands, not regulations. And because the statement is unique, dated from a single day, it allows absolutely no words about an increase, a decrease or a trend — there is no second point to connect to the first.
Where the Gap Comes From: What Each Formulation Picks Up
Les rééditions absorbées
A broad formulation captures subsequent recirculations of the same story: collections, facsimiles, reprints. These items are plentiful and cheap, and they pile up in the bottom half of the distribution. They don't change the price of the original; they move the midpoint downward by inflating the lower tail. This is the simplest mechanism to explain that a median of $39.99 coexists with a median of $129.99 on the same subject.
Related booklets
Ad engines tolerate guesswork. A query formulated with the name of the series and an approximate number indication brings back adjacent numbers, which have neither the same rarity nor the same demand. Each neighbor added is an observation which does not belong to the object sought, and which nevertheless weighs as much as the others in the calculation of the environment.
Lots and groupings
A batch of several booklets displays a single price for multiple contents. Depending on the wording, these announcements are included in the count or not. When they enter, they confuse the reading in both directions: a jackpot can display a high amount without the desired item being particularly valued, and an ordinary lot can display a derisory amount while containing the targeted object.
Derivative objects and spare parts
Around an old booklet gravitate advertisements that are not: single covers, isolated pages, reproductions, promotional items using the same title. Permissive wording picks them up. They almost always occupy the bottom of the scale, in very large numbers, and a median calculated on a group that they dominate no longer measures anything that we thought it measured.
None of these four mechanisms is a failure. These are normal behaviors of a general search engine, designed to maximize recall rather than precision: it is better, from its point of view, to show too much than to miss a relevant ad. The problem only appears when someone applies a statistic to this result assuming the result is clean. The median does not know where its observations come from; she tidies and cuts in the middle, whatever is given to her.
A difference of more than three times between two medians does not mean that one of them is false. Both are accurate: each correctly describes the set she received. What is wrong is to believe that these sets were the same.
The consequence is direct: the question “what is the correct median?” » has no answer. The only question that has one is “what population does this median summarize?” ”, and it requires knowing the wording used.
Why the verified segment moves much less
The second lesson of the survey is more discreet and more useful. While the unverified segment goes from $129.99 to $39.99 depending on the wording — a division by more than three — the verified segment goes from $8,499.99 to $7,949.99. A difference of about 6.5%, on a basis of several thousand dollars. Both segments underwent the same query change; They didn't react the same way at all.
The explanation lies in the nature of what makes up each segment. An announcement of a verified copy carries an explicit identification: the number, the edition, an identifier specific to the copy. This identification mechanically limits the ability of an approximate query to bring in foreign objects. A recent reissue does not present itself as a verified copy from 1963; an isolated page either; a heterogeneous lot either. The verified segment is therefore, without anyone intending it, a much better delimited whole.
This observation must be read for what it is, and not for what it is not. He does not say that verification would have any particular virtue in terms of value, nor that verified copies would be a better investment. He says something strictly methodological: when the objects of a set are described with precision, two different formulations converge; when they are described vaguely, they diverge. The observed stability is a description stability, not a market stability.
We can also turn the observation into a diagnostic rule. If you run two queries on an item and the results diverge wildly, it's not the market that's unstable, it's your definition of the item that's fuzzy. The extent of disagreement between formulations measures your imprecision, not the volatility of what you are looking for. This is perhaps the most productive use that can be made of two contradictory pieces of research.
One final note on the extremes of the verified segment: both formulations cap at the same amount, $79,999.99. It's the same seller, or the same handful of sellers, found by both routes. This amount is a stated request, nothing more, and it does not enter into any reasoning here: it simply illustrates that the two requests do indeed touch on a common core, while diverging sharply on what they add around it.
Why we cannot formulate “well” once and for all
Salespeople's vocabulary is not standardized
Each ad is handwritten, in the words of the seller. No query can anticipate all the ways in which the same issue will be named, nor all the extraneous words that other sellers will have added. Broadening the wording catches more relevant ads and more noise; restricting it does the opposite. There is no setting that has it both ways.
The year indication separates poorly
Specifying 1963 seems decisive. In practice, the year also appears in advertisements for reissues which mention the original publication, and is missing in perfectly relevant advertisements written economically. The filter therefore excludes valid objects while letting in objects that are not valid.
Verification status moves the boundary
Depending on the wording, the same verified copy falls into the count or leaves it: 45 advertisements on one side, 76 on the other. The boundary between the two segments is therefore not given by the object, it is produced by the query. Two people in good faith can divide the market differently without making an error.
Batches refuse automatic sorting
A lot has no unit price. No wording can decide for you whether it should count as one observation, as many, or as none. This choice is a human arbitration disguised as a search parameter, and it changes the median without ever appearing next to it.
The number alone is ambiguous
A serial number does not identify anything outside its series, and the same numerical designation is found in many titles. Any query built primarily on a number therefore causes objects to collide that have no relation to each other. This is one of the easiest sources of inaccuracy to introduce without realizing it.
Content and medium are confused
A query searches for words, not physical objects. The story published in 1963 exists today in multiple forms which share the same title. Distinguishing the original booklet from its reissues by the text of the announcement alone is a problem that the wording does not resolve; it is resolved by the eye, announcement by announcement.
These six obstacles have one thing in common: none of them can be corrected by adding a word to the query. They are due to the very structure of a market for freely written advertisements. This is why the quest for the “good formulation” is a false problem. There is no perfect query waiting to be found, there are only differently biased queries, and the only progress possible is to know the bias of the one you are using.
What to record and what to note
Note the wording along with the number.A statement that retains "median $129.99" without retaining the query that produced it is irrecoverable: in three months, you will be unable to tell whether this amount described 149 announcements mixing reissues and lots, or a better kept set. Record the exact wording typed, to the nearest character, in the same line as the result. This is the only information that makes the figure readable later.
Record the number, not just the median.The statement of August 29, 2026 is telling because we know its numbers: 149 versus 99, 45 versus 76. Without them, the difference between $129.99 and $39.99 would remain inexplicable and would give rise to all fanciful interpretations. The number of announcements is what allows us to understand why two medians diverge; it costs three characters to note and is worth twice the number it accompanies.
Systematically separate verified and unverified in your notes.The two segments do not react the same way to wording changes — a few percent on one side, a factor of three on the other. Mixing them into a single box is like adding a robust measure and a fragile measure, then treating the total as if it were reliable. Two distinct columns, always, even when one of them is empty.
Date and qualify the nature of the amount.Write “requested on 08/29/2026” and not “value”. These amounts are what sellers hope to obtain at a given time; none attests to an agreement between a buyer and a seller. Without this mention, a statement turns into an estimate in the memory of the person who rereads it, and the estimate turns into certainty as soon as it is shared.
Refuse to conclude on a movement from a single day.August 29, 2026 is a photograph. A photograph shows no movement, no matter how carefully the photo is taken. As long as you do not have a second statement, carried out with the same wording - otherwise the comparison will measure your change in request and not the market - any sentence containing "rises", "falls" or "tends" is an invention. The discipline is to leave the trend box empty until it is legitimately filled.
The only defensible practice: cross, then round
If no formulation can be declared good, a method remains usable. It consists of launching several deliberately different formulations on the same object, observing where they come together, and retaining only this area of overlap. On the statement of August 29, 2026, the two requests agree to place the verified segment in the order of magnitude of a few thousand dollars requested, and the unverified segment in that of tens of dollars. It's not much. This is also all that the data allows us to assert without lying.
The corollary is more demanding than it seems: anything more precise than this order of magnitude must be treated as noise. The cents displayed in the medians — these 99 and these 95 terminals — are not information, they are pricing habits copied from seller to seller. Choosing “about forty dollars” rather than “$39.99” is not lazy rounding, it is a refusal to give a figure a precision that its method of obtaining does not give it.
This practice has a beneficial side effect on the way we read other people's numbers. An amount presented without its wording, without its number and without its date should not be discussed: it should be discarded, because it is not verifiable, even in good faith. This is not distrust towards the person who puts it forward, it is the simple observation that the information transmitted is incomplete to the point of being unusable - the reading of August 29, 2026 shows that by only changing the words, we obtain three times less without anything else having changed.
Finally, there remains a consequence on the pace of work. Cross-referencing several formulations costs time, and this time does not buy precision: it buys solidity. We leave with a wide range that we know is defensible, instead of a narrow figure that we cannot justify. For a collector who documents his collection over years, this is the best possible exchange — an order of magnitude that holds is infinitely more useful than a precision that collapses at the first cross-check.
Neither in isolation. Each is accurate for the set of ads it summarizes, and those sets are not the same — 149 ads on one side, 99 on the other. The only honest reading is that asking prices in the unverified segment, on August 29, 2026, were in the tens of dollars order of magnitude, with high sensitivity to how you search.
Because a query doesn't select a subject, it selects ad texts. Depending on the words used, it brings in or out of reissues, related issues, lots and derived objects. The 149 and 99 unverified announcements are therefore not two counts of the same set, but two partially overlapping sets, constructed by different criteria.
More stable from one formulation to another, yes: around 6.5% difference compared to more than three times that. But this stability comes from the accuracy of ad descriptions, not a property of the market. These are still amounts requested on August 29, 2026, not transactions, and the difference in numbers between 45 and 76 announcements shows that even this segment is redivided by the request.
No, and no adjustment of the calculation would change anything. All observations are from a single day. A single point does not define direction: it would take at least a second reading, carried out later with exactly the same wording, for a comparison to have any meaning - and even then, it would then measure the prices asked, not the prices obtained.
Two are enough to reveal instability, as here; three or four begin to draw an exploitable overlap zone. The objective is not to accumulate figures but to identify the range where the results come together despite deliberately dissimilar formulations. Anything beyond this common range, in one direction or the other, should be considered an artifact of the query.
Keep the query next to the number
A statement only has value if it retains its wording, number and date. My Comics Collection allows you to record your observations book by book, with their context, so that the amount noted today remains interpretable in a year instead of becoming an orphan figure.