Scenario Testing — Case Study

16% approved.
The crowd was “monstrous.”

On 14 June 2026, a $60 million UFC cage-fighting card ran on the White House South Lawn. National polling put approval at around 16%. The promoter called the audience “monstrous.” Both are true. We took the event apart with a panel of 399 digital twins to show why — and what a single poll number hides.

399
Digital twins
23
Stacked scenarios
2
Questions per twin
9,173
Reasoned responses
01 — The puzzle

One event, two numbers that don’t fit together

If only one in six Americans approved of the event, who made up the “monstrous” audience the promoter bragged about?

The instinct is to assume one number is wrong — that the poll under-counts, or the hype is inflated. Our reading is that both can be right at once, because approval and viewership are answers to two different questions running on two different engines. Approval keys off legitimacy — the cost, the venue, the vanity, whether the whole thing feels appropriate. Viewership keys off spectacle — curiosity, sport loyalty and the pull of a car-crash you can’t look away from.

A topline poll measures the first and quietly assumes it predicts the second. It doesn’t. To show the gap rather than argue about it, we rebuilt the event from its parts.

02 — Method

Two questions, 23 layers of reality

Rather than ask one population one question once, we gave a panel of 399 digital twins — synthetic respondents modelling a real US population — the same event in 23 versions, stacking one layer at a time, from a single bare facet up to the full assembled spectacle. Every twin answered the same panel, so we can follow the same person across every layer.

Do you approve?

The attitude question. Whether the event is legitimate, appropriate, a good use of the building and the money.

Would you watch?

The behaviour question. Whether you’d actually tune in — a function of curiosity, sport interest and spectacle.

Each twin also wrote out its reasoning, which lets us read why an answer landed where it did, not just the true/false. Two answers × 399 twins × 23 scenarios × written rationale = the raw material below.

Calibration: the twins recover the real-world number

Before trusting the twins on un-pollable, stacked versions of the event, we checked them against ground truth. On the framings that most closely match what a national pollster actually described — a $60M White House event billed as the 250th but landing on the president’s 80th birthday — the twins land at 16.0% approval. Across the comparable pre-event framings they sit in a 14.5%–17.5% band that brackets the published ~16%.

The twins reproduce the real poll (~16%) before we ask them anything a poll can’t. That recovery is what licenses the rest of the analysis.

Reference approval for matched framings — idx 14: 16.0% · idx 0: 16.8% · idx 5: 17.5% · idx 13: 14.5%. Published national approval ~16% (Reuters/Ipsos, June 2026 — a poll of 4,531 US adults found 16% considered the event appropriate).

03 — The data

Every layer, both questions

All 23 scenarios, grouped from single ingredients to combinations to the whole. Approve is the share who support the event; Watch is the share who’d tune in; Gap is watch minus approve. Panel of 399 twins per row.

Scenario layerApproveWatchGap
Single facets — one ingredient at a time
Bud Light as headline sponsor43.1%48.4%+5.3
Celebrating America’s 250th anniversary39.8%37.1%−2.7
Card headlined by champions (Topuria, Gaethje)34.3%33.3%−1.0
A plain UFC cage fight31.6%29.1%−2.5
An event hosted by Trump at the White House30.2%36.8%+6.6
Soundtracked by the Village People (“YMCA”)24.6%34.8%+10.2
Staged during Pride Month19.5%29.3%+9.8
To celebrate Trump’s own 80th birthday17.5%29.1%+11.6
Oiled, scantily clad men grappling in a cage15.8%15.8%0.0
Host appears to fall asleep during the main event3.5%27.3%+23.8
Cost 50% more to stage than it generated1.8%24.8%+23.0
Winning fighter declares “Michelle Obama is a man”0.3%1.5%+1.2
Combinations — layers stacked together
Live UFC at the WH, marking the 250th25.1%34.1%+9.0
UFC at the WH + YMCA + Pride Month19.5%31.6%+12.1
2nd year running: a birthday under a national-milestone guise16.8%29.9%+13.1
$60M event, billed 250th but held on his 80th16.0%29.8%+13.8
UFC 80th birthday + boycotted Bud Light sponsor15.5%31.1%+15.6
$60M UFC at the White House South Lawn14.5%31.8%+17.3
UFC at the WH + host falls asleep14.5%20.1%+5.6
$60M UFC at the WH, for his 80th birthday12.8%27.3%+14.5
$60M, 250th-but-80th, + 50% cost overrun7.0%19.3%+12.3
UFC at the WH + cageside “Michelle Obama” slur1.3%4.0%+2.7
The whole — the full event, every layer at once
The complete event, all layers stacked1.5%8.3%+6.8

Binary approve / watch outcomes, 399 twins per scenario (397–399 valid responses). At this panel size treat differences under ~5 points as noise.

04 — Parts vs whole

The parts don’t add up. The worst ones take over.

This is the core reason to stack scenarios rather than test a single topline. The full event approves at 1.5% — lower than every individual ingredient except the slur. A plain fight polls 32%. The 250th framing, 40%. Bud Light, 43%. Assemble them and the result isn’t an average of the parts — it’s dragged down to the floor by the two facets that independently destroy legitimacy.

X — the charge

Each facet alone tells you the value of every ingredient. A plain fight is an asset (32%). A 50% cost overrun is a landmine (1.8%).

X+Y — the why

Stacking shows which layer detonates the others. Add the $60M cost to the framing and approval collapses while viewing barely moves — causal attribution a poll can’t give.

X+Y+Z — the outcome

The whole lands at 1.5%, below any single part. Sum-of-parts ≠ whole; the gap between them is the actual insight.

Test the parts → you learn the levers. Test the whole → you learn the outcome. Test the path between → you learn the why.
05 — The split

Approval collapses. Viewing doesn’t.

The facets that destroy approval are often the ones people most want to see. For spectacle, controversy is fuel. Two facets make the decoupling unmistakable:

Host falls asleep mid-event
approve vs watch
3.5%
27.3%
Cost 50% more than it made
approve vs watch
1.8%
24.8%

A president falling asleep at his own event is indefensible to approve (3.5%) — and exactly the clip people tune in for (27.3%). A loss-making vanity project is almost impossible to endorse (1.8%) but doesn’t dent curiosity (24.8%). The things that sink legitimacy can leave spectacle untouched — sometimes feed it.

The one exception: the “Michelle Obama is a man” slur is the only facet that suppresses both — approval to 0.3% and viewing to 1.5%. A line that crosses from controversial to repellent doesn’t draw a rubbernecking audience; it drives people away. Not all controversy is created equal.

06 — The audience

The crowd is mostly made of disapprovers

Because every twin answered both questions, we can sort them into four quadrants for any scenario. Do that for the realistic $60M White House card and a counter-intuitive picture appears: the people who’d watch outnumber the people who approve — and most of the audience disapproves of the very thing they’re watching.

Realistic $60M White House UFC cardTwins
Approve & would watch50
Disapprove — but would watch anyway77
Approve, wouldn’t watch8
Disapprove, wouldn’t watch264
Of everyone who’d tune in, 61% disapprove of the event. Push to the full, ugliest version and that rises to 88%. The audience is built out of its own critics.

This is the mechanism behind the paradox. The “monstrous” crowd and the 16% approval aren’t in conflict — the crowd is largely the same people who told a pollster they disapproved. A topline that measured approval and stopped would have predicted a small audience and been badly wrong.

07 — Segmentation

The gap isn’t everywhere — it lives in fight fans

The disapprove-but-watch gap is not spread evenly across the population. Splitting the panel shows where the “watch anyway” behaviour concentrates — and where it barely exists. Figures below are for the realistic $60M White House card.

SegmentApproveWatchGap
By sport interest
Fight fans (would watch a plain UFC card) · n=11624.1%66.4%+42.2
Non-fans · n=2838.1%11.3%+3.2
By gender
Men · n=12016.7%44.2%+27.5
Women · n=27511.3%20.4%+9.1
By generation
Millennials · n=10911.9%34.9%+22.9
Gen X · n=13710.9%25.5%+14.6
Baby Boomers · n=13515.6%23.7%+8.1

Fight fans carry the paradox almost single-handed: only a quarter approve, but two-thirds would watch — a +42-point gap. Among non-fans the gap nearly vanishes (+3): they don’t approve and they don’t tune in. The “monstrous” audience isn’t the general public holding its nose — it’s a committed sport audience that watches the fights and files the politics separately.

08 — Why twins, not a survey

The twins admit what people won’t

“I disapprove of this… and I’ll watch it anyway” is an awkward thing to tell a human interviewer. It sounds hypocritical, so real respondents under-report it — the classic social-desirability bias that makes intent-to-watch one of the hardest things to measure honestly. The twins have no ego to protect, so they state the contradiction plainly and then explain it. Reading the reasoning of the disapprove-but-watch twins, a single pattern dominates:

“They explicitly list UFC as a passion… they also say they don’t like mixing politics with sports, so they’d downplay the Trump/White House angle but still watch for the sport itself.”Female, Gen X — disapprove, would watch
“He consistently prefers watching sports over most TV… while he dislikes celebrity/political spectacle, the Trump-hosted framing likely wouldn’t stop him from watching the fights.”Male, Gen X — disapprove, would watch
“She’s an occasional UFC viewer… bored and annoyed by constant Trump news, but she’d still tune in out of curiosity, the way she treats the Super Bowl.”Female, Baby Boomer — disapprove, would watch

Two channels, cognitively separate. Disapproval lands on the politics; the watch decision keys off the sport and the curiosity. Crucially, when these twins explain why they’d watch, they almost never cite their disapproval as a reason to stay away — the two judgements simply don’t talk to each other. That decoupling is the whole paradox in miniature, and it’s exactly the admission a self-conscious human survey tends to bury.

09 — What it means

Measure one number and you’ll mispredict the other

Approval keys off legitimacy. Viewing keys off spectacle. They are separate outcome channels — not one number worn two ways.

For anyone forecasting how a charged event will land — a brand activation, a controversial campaign, a live broadcast, a product tied to a polarising moment — the lesson generalises past the UFC. A single approval poll would have told you this event was dead on arrival. The behaviour data says the opposite: a large, engaged, mostly-disapproving audience showed up anyway. If your decision depends on attendance, sponsorship or viewership, the attitude number is the wrong instrument.

Model both channels

Attitude and behaviour move independently. Get both, or you’re guessing at the one you didn’t ask.

Stack the scenario

Test facets alone, in combination, and as the whole. The gap between sum-of-parts and whole is where the decision lives.

Read the reasoning

The why behind a true/false is where the strategy is. Twins surface contradictions humans hide.

Method & caveats

How to read these numbers

The panel. 399 digital twins — synthetic respondents modelling a US population — each answered all 23 scenarios for both approval and watch-intent, with written reasoning. One row per twin × scenario.

Noise floor. Outcomes are binary at n≈399. Treat differences under ~5 points as noise; we lean on the large, directional gaps, not decimal places.

Don’t average across scenarios. There is no single “topline” here — support is a function of framing, so a mean across 23 scenarios is a meaningless artifact. Each row is its own question.

Stimulus honesty. Scenarios are written to be factually defensible; narrative/perception items (e.g. the “asleep” framing) are tested as stated claims, not asserted facts. The ~16% real-world benchmark is the June 2026 Reuters/Ipsos poll figure.

Brox.AI

Don’t measure one number.
Model the whole picture.

Honest answers. Stacked scenarios. The full, nuanced picture — broken down so you can understand it.

See what a poll can’t.
Brox.AI — Scenario Testing case study · UFC White House event, 14 June 2026 · Panel: 399 digital twins × 23 scenarios × 2 questions. Approval benchmark: June 2026 Reuters/Ipsos poll (~16%).