Abstract
Reception scholarship regularly reports what a marketing campaign “did to” a film — what audiences knew going in, which surprises were forfeited, which structural effects the promotional apparatus foreclosed. The evidence for these reports is drawn almost entirely from the population that the campaign reached and that left traces of having been reached. The conclusion is then stated as a claim about the audience. This paper argues that the move is a frame error: a sampling frame constructed by an instrument is substituted for the target population the instrument was never able to survey. The error is not random in direction. Because public availability of information is the ceiling of possible exposure, and because availability is the quantity actually observed, the error runs one way — it inflates estimated foreknowledge and correspondingly deflates the film’s estimated remaining capacity to operate on a viewer as designed. I distinguish four states that the literature routinely collapses (availability, exposure, uptake, salience at the moment of viewing), give five mechanisms by which the minimally informed viewer is rendered uncountable, argue that this viewer’s vantage is also unreconstructable after the fact, and propose a set of bounded reporting practices, including prospective pre-registration of first-viewing readings. A worked case is drawn from the 2026 adaptation of Project Hail Mary, whose central non-human character was disclosed in promotional material more than a year before release. The paper closes by turning the argument on itself: the one class of minimally informed viewers who can be recovered is, by the same logic, the class least representative of the population it is asked to stand for.
1. The claim in the wild
A familiar sentence appears across trade commentary, critical reviews, and academic reception work in roughly this form: the trailer gave away the reveal, so audiences went in already knowing. Variants attach to particular structural features — a casting disclosure, a plot beat shown in a television spot, a poster that discloses a creature design a film withholds for forty minutes. The sentence has two parts. The first is a claim about a promotional artifact and is generally correct and easily checked: the trailer did disclose the thing. The second is a claim about a population of persons and their mental states at a particular moment, and it is almost never checked at all, because the apparatus that would check it does not exist.
The two parts are joined by an inference so quiet that it is rarely written down: that what a campaign made available to a moviegoer was thereby known to the moviegoer. Stated plainly, the inference is obviously invalid. Left unstated, it does the load-bearing work in a large body of scholarship about how contemporary promotional practice reshapes narrative experience.
My interest here is not to catch out a careless sentence. It is to identify what kind of error this is, why the field’s instruments make it nearly unavoidable, and what a disciplined account would say instead.
2. The error named
The error is a mismatch between sampling frame and target population, of the classic sort, with an unusual feature: the frame is not drawn by the researcher but supplied ready-made by the industry being studied.
Marketing research measures reach. It measures reach because reach is not total — that is the entire reason the measurement exists. A campaign report that says a trailer achieved a given penetration among a defined demographic is a report with a denominator, and the denominator is the point. Awareness tracking, presale telemetry, search-volume curves, social engagement counts, exit polling of opening-weekend theatrical attendees: each of these is a well-formed measurement of a specified subpopulation, and each is candid within its own discipline about whom it did not touch.
The error occurs in transit. When these findings cross from marketing research into reception scholarship, the qualifier is dropped and the residual is retained. “Sixty-one percent aided awareness among frequent moviegoers aged 18–34” becomes “audiences knew.” What was a bounded statement about a measured slice becomes an unbounded statement about everyone who saw the film. The frame — persons the campaign reached, who were disposed to be reached, and who left a recoverable trace of having been reached — is then reported under the name of the target population, which is every person who watched the film in any window, on any surface, for any reason.
This is a definitional failure before it is an empirical one. The category “the audience” is being defined, silently, by the reach of the instrument that measures it. A population defined by the apparatus is then treated as a natural kind, and findings about the apparatus are read as findings about the world.
3. Four states, routinely collapsed
Clarity requires separating states that the collapsed usage runs together. For a given item of information i about a film, and a given viewer v:
Availability. i was published somewhere a member of the public could in principle encounter it. For a wide-release trailer, availability is effectively unity. It is also the only one of the four states that can be established from the documentary record alone.
Exposure. i actually crossed v‘s perceptual field — the trailer played in front of them, the thumbnail appeared in their feed, someone said the thing aloud in their hearing.
Uptake. v encoded i as information about the film, in a form retrievable later. Exposure without uptake is the ordinary condition of modern promotional life: a trailer plays in an adjacent browser tab, before a video the viewer wanted, in a theater lobby, and leaves nothing behind. Retention decays; a June 2025 disclosure is not a June 2026 memory.
Salience at viewing. v had i available to consciousness at the moment the film’s structure depended on its absence. A viewer may retrieve, on being asked afterward, that they had once seen a promotional image, while having had no access to it during the scene the image disclosed.
The ordering is monotone. Salience is bounded above by uptake, uptake by exposure, exposure by availability. Where i is publicly released promotional material, availability equals one. The literature estimates the last term with the first.
Because the substitution replaces a quantity with its own ceiling, the resulting bias has a fixed sign. Foreknowledge is systematically overstated. What is systematically understated is the residual population for whom a film’s withheld information was still withheld — and therefore the film’s remaining capacity to do what its structure was built to do.
4. Five reasons the minimally informed viewer cannot be counted
Suppose a researcher accepts the argument and sets out to size the residual population directly. Five features of the object frustrate the attempt, and they compound.
(a) Absence generates no artifact. Knowing produces traces: a search query, a presale, a forum post, a preorder of the source novel, a subscription to a channel that covers the production. Not-knowing produces nothing. The unloaded viewer is a null in every dataset that exists, and in observational data a null is indistinguishable from a nonentity. The population is not underrepresented; it is unrepresented, and its unrepresentation looks exactly like absence.
(b) The trace-producing disposition is the variable of interest. The behaviors that make a viewer visible to reception research — reviewing, posting, responding to a survey, joining a tracking panel — correlate strongly with the same engagement that produces foreknowledge. This is not ordinary sampling noise that a larger n would wash out. The selection mechanism and the measured quantity are near-identical, so increasing sample size increases precision around an estimate whose bias is structural.
(c) The category “moviegoer” is drawn by the industry’s own instruments. Theatrical attendance is measured because it is monetized at a point of sale. The viewer who selects a film from a seatback console at thirty thousand feet, or from a hotel menu, or from a catalog tier three years after release, is not a moviegoer under the operative definition, and is invisible accordingly — yet is precisely the sort of viewer most likely to arrive unloaded, since the selection is made without prior intent, in an environment that offers no opportunity for research and often no reliable connectivity to conduct any.
(d) The window closes. Even if the residual population were reachable, the state under study is temporary in one direction and regenerating only weakly in the other. Information saturation increases with time from release. The pool of unloaded viewers refreshes only through people who do not engage with discourse about films — which is to say, through exactly the people who will not answer.
(e) Measurement destroys the state. Asking “did you know about the reveal?” supplies the reveal, or supplies the fact that there was one. Prospective study of unloaded viewers requires informing them that a study is occurring, which loads them. This is not an inconvenience to be engineered around; it is a property of the object. The state of interest is the state of a person to whom no one has yet drawn attention to anything.
5. Why the vantage cannot be reconstructed afterward
If the state cannot be caught live, can it be recovered from testimony? Only in a degraded form, for reasons that are worth separating from the counting problem above.
A viewer does not store a record of what they did not know. Knowledge has an object; ignorance has none, and so it is not filed. What the viewer stores instead is the experience of the reveal, and once that experience exists, the prior state is only accessible through it. The reconstruction is performed by a mind that now possesses the outcome and cannot suppress it — the standard hindsight effect, operating here not on probability estimates but on autobiographical content. Asked whether they had suspected, viewers will produce an answer, and the answer will be built from present knowledge.
The error runs in both directions and cannot be corrected by a constant. Some viewers overclaim foreknowledge, because retrospective competence is flattering and because a dimly recalled promotional image is easily promoted to a memory of having known. Others underclaim it, because the surprise was pleasant and the memory of surprise is the thing they wish to preserve. Nothing in the testimony distinguishes the two.
Add to this the professional problem. Film reception writing is produced overwhelmingly by critics and scholars, whose working conditions make them the most heavily loaded viewers alive: press materials, embargo terms, festival context, source texts read in preparation, months of production coverage as occupational background. When such a writer reasons about what the audience knew, introspection is available and is not obviously misleading — the writer did know. The generalization from a maximally contaminated sample to a general population is thus made under conditions where it feels like reporting rather than inference. This is a structural explanation for the error’s persistence, and it predicts that the error will be most severe precisely in the most expert venues.
6. What the deflated claims look like
Applying the distinctions, several standard formulations should be rewritten. “The trailer spoiled the reveal for audiences” becomes “the trailer disclosed the reveal; the disclosure was available to any member of the public and reached an unmeasured fraction of eventual viewers, of whom an unmeasured further fraction retained it in an operative form.” That is not elegant, but it is the sentence the evidence supports, and it has a different implication: the film’s structural gamble may have paid off for a large number of people, and the literature has no standing to say otherwise.
Two further corrections follow.
First, an aggregate claim is not an individual claim, and interpretive claims about films are individual claims. Suppose the modal viewer did know. A film seen by fifty million people that retained its surprise for one viewer in five surprised ten million people. Reception writing consistently treats the modal viewer as the viewer, which converts a distributional statement into a universal one and then draws aesthetic conclusions — about whether a structural device “still works” — that only the universal statement would license.
Second, universal claims are cheap to defeat. A single documented case of a viewer who watched without the disclosed information is logically sufficient to falsify “audiences knew,” and existence claims of this sort are the one thing testimony can establish reliably, since a person who reports surprise at a reveal is not reconstructing whether they knew but reporting what happened to them. What testimony cannot establish is frequency. The asymmetry is worth stating explicitly, because it defines the useful scope of first-person evidence in this area: it bounds the universal claim without sizing the population.
7. A worked case
The 2026 adaptation of Project Hail Mary provides a clean instance, because the disclosure is documented and the structural stake is high. The film’s central relationship is between the human protagonist and a non-human character whose existence, appearance, and role were shown in promotional material released more than a year before the film opened. On the standard account, the film therefore could not use the encounter as a discovery; the audience arrived knowing there would be an alien, and knowing roughly what he looked like.
That account is correct about the artifact and unsupported about the persons. I can supply one counterinstance from my own record, made under conditions that reduce the usual reconstruction problem: I selected the film from an in-flight catalog on the recommendation of others, with minimal knowledge of the plot, no familiarity with the trailers, and without having read the novel. I did not know the second character existed, and the encounter functioned as a discovery. The case sits in the ancillary window, was selected without prior intent, and produced no trace of foreknowledge because there was none to produce — which is to say it is an instance of exactly the class described in §4, and it exists.
What the case establishes is bounded, and the bound should be stated rather than left to the reader. It falsifies the universal form of the claim. It does not size the residual population, does not establish that the film’s structural gamble was correct in expectation, and does not license any inference about the modal viewer. It does establish that the ancillary window — in-flight, hotel, catalog streaming, recommendation-driven selection without research — is a locus where the unloaded viewer persists long after the theatrical window closes, and that this locus is systematically outside the measurement apparatus that generates reception claims.
I would add one further observation from the same case, which bears on the salience distinction in §3. Even where a viewer has had some exposure — a recalled fragment of a poster, a friend’s passing remark — the operative question is whether the information was present to the mind during the scene that depended on its absence. A film watched on a small screen in a noisy cabin, chosen on impulse, is watched under attentional conditions that reduce retrieval of weakly encoded background information. The ancillary window does not merely contain more unloaded viewers; it contains conditions that push partially loaded viewers toward functional unloadedness. This is a testable proposition and, so far as I can determine, an untested one.
8. What a disciplined account would do instead
The object here is a negative-space object: defined by the absence of the traces that would ordinarily constitute evidence of it. Such objects can be handled without being counted, provided the handling is honest about which operations are available.
Report the frame, not the population. Every claim about audience foreknowledge should carry its denominator. “Among opening-weekend theatrical attendees who responded to exit surveys” is a usable phrase. “Audiences” is not.
Name the residual and specify its properties. The unloaded viewer should appear in the account as a named class with stated characteristics — ancillary-window selection, recommendation-driven choice, low discourse engagement — rather than being absorbed into the estimate by silence. A class that is named can be argued about; a class omitted cannot.
Pre-register first-viewing readings. Where a researcher or collaborator will see a film unloaded, the reading should be recorded before any contact with the source text, the discourse, or the production history. Prospective capture is the only method that produces uncontaminated testimony, since it does not ask the viewer to reconstruct anything. It is available only opportunistically, and it should be treated as a discipline for structuring the later analysis rather than as an inferential engine: its output is a fixed record that later readings can be checked against, not a finding.
State the sensitivity threshold. For any interpretive claim that depends on foreknowledge, the writer should ask what fraction of viewers would have to be unloaded for the claim to fail, and say the number. Frequently the answer is a small percentage, and stating it makes plain how much weight is resting on an unmeasured quantity.
Use ancillary-window structure as an indirect bound. Total viewership across in-flight, hotel, catalog streaming, library circulation, and late linear broadcast is partially documented and is systematically excluded from reception evidence. It does not identify unloaded viewers, but it bounds the size of the invisible region and shows that the region is not small.
Distinguish disclosure analysis from reception analysis. Study of what a campaign disclosed is a legitimate and well-evidenced enterprise. It becomes unsound only when its findings are restated as findings about mental states. Keeping the two under separate headings would remove most of the error at no cost to either.
9. The argument turned on itself
A methodological paper should state where it is vulnerable, and this one has a specific vulnerability that follows from its own premises.
The class of unloaded viewers who can be recovered consists of those who, after viewing, entered discourse and reported. They are therefore recruited from among the discourse-entering. The single case in §7 is offered by a person who writes papers about films — which is to say, by a member of the trace-producing population whose overrepresentation §4(b) identifies as the source of the error. The recoverable unloaded viewer is, by construction, the least typical member of the class he is asked to represent. Every testimony of this kind will have the same defect, and the defect is not removable by collecting more testimonies, since each additional one is drawn by the same mechanism.
I do not think this dissolves the argument, for the reason given in §6: an existence claim survives an unrepresentative source. What it does establish is a limit on how far the corrective can go. This paper can show that the estimate is biased and can name the direction of the bias. It cannot supply the corrected estimate, and a reader looking for one should be told plainly that the object does not admit it.
That limitation is worth accepting rather than engineering around, because the alternative is worse. A field that will not report an unmeasurable quantity as unmeasurable will report it as zero, and reporting it as zero is how the error under discussion was made in the first place.
10. Conclusion
The sentence “audiences knew” is doing work it has not earned. Its evidentiary base is a population defined by the reach of a promotional apparatus and by the disposition to leave traces of having been reached. Its subject is every person who watched the film. The gap between those two is occupied by a class of viewers who select films casually, in ancillary windows, without research, and who leave no record of their ignorance because ignorance leaves no record. That class cannot be counted, and its vantage cannot be reconstructed afterward, because the state in question is destroyed both by measurement and by the experience of the reveal itself.
The correct response is not to assume the class away, which is the current practice under a different name, nor to pretend to size it, which the evidence will not support. It is to name it, state its properties, carry the denominator on every claim, record first viewings prospectively where the chance arises, and say how much of the argument depends on the number nobody has.
