Miles Bron burned a napkin in front of everyone, and with it, Helen Brand’s proof. With the napkin gone and his friends still backing him, she could not get him convicted of murdering her twin sister, Andi. So she started smashing his glass sculptures, one crash after another. Somewhere in the wreckage she found the sample of Klear, Miles’s unstable prototype fuel, and threw it into the fire.
The fire caught the island mansion, Glass Onion, which was made almost entirely of glass. In the heat the panels cracked, small sounds first, then whole walls shattered. The Mona Lisa hung in the middle of the blaze, on loan from the Louvre (Figure 1). “Your fuel of the future just barbecued the most famous painting in the world,” Helen told him, watching the painting burn.
Figure 1. The real painting stayed in Paris for the film: the Mona Lisa (left), and the Louvre’s Salle des États, where it hangs alone behind glass (right). Painting: Leonardo da Vinci, public domain. Photo: Shonagon (2022), CC0. Both via Wikimedia Commons.
That is the ending of Glass Onion: A Knives Out Mystery. The fire is not the twist. None of that ending would have happened if a courtroom had gotten one thing right years earlier. Andi had sued Miles for stealing an idea. She sketched it once on a napkin at a bar called the Glass Onion.
Some real inventions started on napkins too. A NASA engineer, Bill Burcham, sketched one on an airline napkin: a way to steer a plane with its engines alone when its controls fail. In 1995 his team landed an MD-11 airliner that way (Figure 2).
Figure 2. Frank W. “Bill” Burcham’s napkin sketch of the Propulsion-Controlled Aircraft (left), and the MD-11 flight team at Edwards, California, in 1995, Burcham in the back row (right). Photos: Dennis Taylor (EC94-42805-1) and T. Landis (EC95-43251-4), NASA Dryden. Public domain, via Wikimedia Commons.
Miles built Andi’s idea into Alpha, a technology company worth billions, and forced her out. In court, the friends who owed him their careers swore his version was the truth. Andi could not find the napkin, and all she had was her own word. She lost. Years later the napkin turned up. She emailed the friends a photo, and Miles killed her before she could act. Helen found the same napkin hidden in his house. That was the napkin Miles burned, the small fire Helen answered with a much larger one.
That napkin was the proof the courtroom never had. The trial asked whether Miles was innocent of stealing Andi’s idea, given the evidence. Statisticians call a starting claim like that the null hypothesis: the default that stands until evidence overturns it. In a courtroom it is innocent until proven guilty; in the lady tasting tea it is a taster who is only guessing. The burden of overturning that default fell on Andi, not Miles. The opposite claim, that Miles took the idea, is the alternative hypothesis.
No court could watch that night at the bar again. A court gets a sample instead, the way a cook judges a pot of soup from one spoonful. Here the spoonful was word of mouth: Miles’s word against Andi’s, backed by his friends.
A verdict on that evidence works like a choice of two doors: walking through one closes the other. Statisticians call the first reject the null, which overturns Miles’s story, and the second fail to reject it, which lets the story stand. A test has no third door. Failing to reject only means the evidence was not strong enough yet. The friends took the second door, and the story stood.
You probably have a story standing the way Miles’s did. Some evening this week, scrolling for another number, your thumb stops on a name you have not texted in three years. The old version comes back at once: they stopped making an effort, so you did too. You settled on that version the month the texts stopped, and never checked it. Three birthdays have passed unmarked on both sides, so what would it take to know whether your version is true? The story stands until something overturns it, and every month it stands is another month without them.
How a Story Can Be Wrong
The film shows its audience Andi’s sketch on the napkin, a view of the truth a court does not get. The truth had two possible states: Alpha was Miles’s idea, or it was not. The court’s decision had two possible states too: for Miles, or for Andi. Crossing what was true with what the court decided gives four possible outcomes. Two of them match the truth. Two of them do not.
What was true, crossed with what the court decided. Tap a box.
(fail to reject)
(reject)
Two ways to be wrong. Two of the boxes are errors. Rejecting the null when it is true is a Type I error, a false positive: ruling for Andi when the idea really was Miles’s. Failing to reject the null when it is false is a Type II error, a false negative: ruling for Miles when the idea was Andi’s. The boy who cried wolf holds both. Villagers who run to a false alarm make a Type I error, and villagers who ignore the real wolf make a Type II (Figure 3).

Figure 3. The shepherd boy and the wolf, from The Aesop for Children (1919). Illustration: Milo Winter. Public domain.
A verdict never reports which of the four boxes was the real one.
The same four boxes sit under the name in the phone that has gone three years without a text. The settled version, that the friend stopped making an effort, is the standing story. Keeping it when it is wrong is the courtroom’s mistake, and it has cost three birthdays on both sides. Dropping it when it was right is the other mistake, and it costs one awkward message. The silence never says which box the friendship is in.
Deciding how wrong to be. Naming the two errors does not keep either one in check. Take two imaginary friend groups. One would never turn against Miles, whatever the evidence. It never makes a Type I error, and it never catches a false story either: perfectly safe, and perfectly useless. The other would always turn against Miles. It catches every false story, and it also rules for Andi when the idea really was Miles’s.
Keeping both errors in check is the job of Jerzy Neyman and Egon Pearson’s framework (Figures 4 and 5), and it works in two steps. First, fix the maximum tolerable Type I error at a threshold called the level of significance, written alpha. Then, among every test that keeps its Type I error at or below alpha, pick the one with the highest power. Power is the probability of rejecting the null when the null is false. A Type II error lets a false story stand. Its chance is written beta. Power is one minus beta. The never-turn group has a power of zero. The always-turn group buys a power of one with a Type I error rate of one. That is why the framework caps false positives first and only then reaches for power.
Figure 4. Jerzy Neyman at Berkeley in 1969 (left) and lecturing in Warsaw in 1973 (right). Photos: Konrad Jacobs, Oberwolfach Photo Collection. CC BY-SA 2.0 de, via Wikimedia Commons.

Figure 5. Egon Pearson, about two, on the knee of his father, Karl Pearson, himself a statistician, with his mother, Maria, and his sister, Sigrid, around 1897. Photographer unknown. Public domain, via Wikimedia Commons.
No threshold can keep a test from being wrong. What alpha caps is how often the test may overturn a true story, a limit set before the first friend takes the stand.
How Many Voices Are Enough
Ten imaginary friends wait on a hard wooden bench outside a courtroom, called in one at a time. None of them hears what the others said. Each answers one either-or question: does the idea belong to Miles or to Andi? No reasons, just a name. The count of friends who say Andi is the whole of the evidence.
The null, restated as a count, is that the friends are no more likely to name Andi than Miles. The alternative is that they lean toward Andi, and it tells the court which tallies would count in her favor. Before any friend speaks, the court decides how many of the ten naming her would hand her the company. That decision is the rejection rule. Like the threshold for passing a bill, it is set before the votes come in. Five of ten is an even split, evidence for neither, so the rule belongs somewhere in the top half. The question is where.
Start with the strictest rules: all ten name her, at least nine, eight, seven. If the friends were really a 50/50 split, how often would luck alone hand up that many Andi-votes or more? That probability is the rule’s Type I error rate. Under that split, each friend answers like a coin flip, and ten flips can land 1,024 ways, all equally likely. A rule’s rate is its share of those ways: at least nine naming Andi covers 11 of them (Figure 6).
Figure 6. Ten friends, counted: each rule’s Type I error rate, its power, Neyman and Pearson’s two steps, and the arithmetic. Built for this piece.
Instinct draws the line lower. Six of ten naming Andi is already a majority, and most people would stop there and call the company hers. But a 50/50 split produces six Andi-votes or more about 38 times in 100, by luck alone. A line at six would rule for Andi more than a third of the time with nothing pointing to her. That is why six is never a candidate.
Neyman and Pearson’s first step narrows the list: fix alpha. Set it at 0.05, a 1-in-20 chance. A rule’s Type I error rate is its size. A rule stays a candidate only if its size falls at or below alpha. Two rules qualify, all ten and at least nine. At least eight, at 0.0547, is over the line, barely, and barely over is still over. So the threshold cannot fall below nine naming Andi, three votes past a bare majority. That rule’s size sits under 0.05, not on it: counts come in whole friends.
Where alpha should sit is still argued. In 2018 a large group of researchers proposed 0.005 for any claim of a new discovery, and fields still differ on the line they use.
The second step keeps the survivor with the most power. To put a number on power, suppose that when the idea truly is Andi’s, each friend names her nine times in ten.
Of the two survivors, at least nine has the higher power, 0.74 against 0.35. When the idea is hers, at least nine rules in her favor about 3 times in 4, and all ten about 1 in 3. So the court keeps at least nine: rule for Andi only if nine or more of the ten name her.
Double the tolerance to 0.10, and at least eight gets in and wins, with power 0.93. Twice the false-positive risk buys a lower line and a better chance of ruling for Andi when the idea truly was hers. And with fewer than ten?
Why Four Voices Were One
In the courtroom there were not ten friends. There were four: Claire Debella, Lionel Toussaint, Birdie Jay and Duke Cody. Build the same kind of rule for four, and the strictest one possible is unanimity, all four naming Andi. Its Type I error rate is 0.0625, or 1 in 16 (Figure 7).

Figure 7. Every way four friends can answer if each is a coin flip, gold for Andi, with the one all-gold way boxed. Drawn for this piece.
Held against the same 0.05 threshold, 0.0625 is over, the way at least eight was over for the ten. Unanimity is the most any rule could demand, and it already fails the first step. With four friends no rule keeps the Type I error at or below 0.05, so no rule runs at all. Four friends make the never-turn group real: every possible tally, from none naming Andi to all four, lands on Miles. At 0.10, unanimity alone gets in. At least three of four is 0.3125, about 1 in 3.
And in the film the four never spoke as four anyway. They spoke as one. Miles paid the group together, and they answered together, each with a reason to protect the man who paid them. The 0.0625 assumed four friends answering independently, each on their own. That assumption was the mental model under the whole test, its picture of how the evidence behaves: a count where every voice is its own. What the arithmetic counted as four voices for Miles was one decision spoken four times. When the answers are not independent, the count is not four pieces of evidence. It is one piece wearing four faces.
The same miscount happens at family tables. A story comes up on the Sunday call with a parent, as it has for six years: a cousin walked out of a wedding for no reason. “She just got up and left,” an aunt says, and five relatives agree. If all five heard it from that aunt, the family holds one account repeated five times. The check is one question: who saw it happen? If the answer is one person, the cousin has spent six years of holidays outside a story never tested.
Back in the courtroom, the verdict for Miles worked like a two-headed coin: it lands the same way whatever happens. Carrying Andi’s case would have taken many more friends answering on their own, most of them naming her.
The threshold alpha and the company Alpha share nothing but a name. Here, though, alpha was the line the friends’ word had to clear before the company named Alpha could change hands. Their word never came near that line. Only the napkin might have: the one piece of evidence Miles had not paid for. It surfaced years too late, then burned at Miles’s hand before the Mona Lisa did.
A Closing Reflection. A napkin is the proof that turns up after the verdict, and most standing stories never get one. This week, notice which ones you are letting stand.
- Spot the standing story. Right now, notice the name in your phone you keep scrolling past, the version that comes back, and what your body does with it: a held breath let out, the pull to keep going. Is that story true, or only never overturned? How many birthdays has it cost, and how many could one text save?
- Trace the voices. On this Sunday’s call with a parent, take the family story everyone agrees on: the cousin who “just left,” the uncle who “never paid it back.” Ask who saw it happen, and listen for the first name. How many holidays has someone missed on one person’s word?
- Set the line first. Before Monday’s meeting, where a hire gets picked or a late project gets blamed on someone, write down what would count as enough evidence, before anyone speaks. One loud teammate, or three people who saw it separately? Who pays if your line sits in the wrong place?
Andi’s napkin burned in seconds, and a glass house burned after it. A message to the friend you stopped texting is a napkin of your own: small, late, and still able to overturn a story.
Every verdict risks two mistakes: jailing the innocent or freeing the guilty. A court decides first which one it can live with, because it cannot avoid both. Before you judge someone, ask which mistake you would rather make.
Where This Framework Came From
Jerzy Neyman and Egon Pearson built this framework together, and their 1933 paper set out how to find the most powerful test once alpha is fixed. The term null hypothesis is Fisher’s, from the same tea test. The alternative needed to unseat it, alpha fixed before a verdict, the power a test gets once alpha is set, the false positive and the false negative all came out of the pair’s work. It is still the framework this piece runs on, and the one much of science still runs on too.
Intellectual Honesty Note. Glass Onion: A Knives Out Mystery (2022, directed by Rian Johnson) is a real film telling a fictional story. The account of Miles Bron, Andi Brand, and the friend group given here follows its plot. The film presents the painting burned in its ending as the real Mona Lisa, on loan from the Louvre. The production burned a replica, and the real painting was never lent to the film. Null hypothesis testing, the alternative hypothesis, alpha, power, and the four-outcome table, along with the terms Type I and Type II error, are standard statistical vocabulary, not invented for this piece. The shared name between Miles Bron’s company and the statistical alpha may or may not be a pun the film intended; the link this piece draws between them is its own reading. The ten witnesses are an illustration built for this piece, not part of the film. A real trial turns on the substance of testimony, not a tally, and the film’s case was not decided this way.
References
Benjamin, D. J., Berger, J. O., Johannesson, M., Nosek, B. A., Wagenmakers, E.-J., Berk, R., … Johnson, V. E. (2018). Redefine statistical significance. Nature Human Behaviour, 2(1), 6-10.
Fisher, R. A. (1935). The design of experiments. Oliver and Boyd.
Jacobs, K. (1969). Jerzy Neyman [Photograph]. Oberwolfach Photo Collection. CC BY-SA 2.0 de, via Wikimedia Commons. https://commons.wikimedia.org/wiki/File:Jerzy_Neyman2.jpg
Jacobs, K. (1973). Jerzy Neyman [Photograph]. Oberwolfach Photo Collection. CC BY-SA 2.0 de, via Wikimedia Commons. https://commons.wikimedia.org/wiki/File:Jerzy_Neyman.jpg
Johnson, R. (Director). (2022). Glass Onion: A Knives Out Mystery [Film]. Netflix.
Landis, T. (1995). MD-11 PCA research flight team photo [Photograph, EC95-43251-4]. NASA Dryden Flight Research Center. Public domain, via Wikimedia Commons. https://commons.wikimedia.org/wiki/File:MD-11_PCA_-_Research_flight_team_photo_EC95-43251-4.jpg
Leonardo da Vinci. (c. 1503-1519). Mona Lisa [Painting]. Musée du Louvre. Public domain, via Wikimedia Commons. https://commons.wikimedia.org/wiki/File:Mona_Lisa,_by_Leonardo_da_Vinci,_from_C2RMF_retouched.jpg
Neyman, J., & Pearson, E. S. (1933). On the problem of the most efficient tests of statistical hypotheses. Philosophical Transactions of the Royal Society of London, Series A, 231, 289-337.
Perezgonzalez, J. D. (2015). Fisher, Neyman-Pearson or NHST? A tutorial for teaching data testing. Frontiers in Psychology, 6, 223. https://doi.org/10.3389/fpsyg.2015.00223
Shonagon. (2022). Salle des États, salle de la Joconde [Photograph]. CC0, via Wikimedia Commons. https://commons.wikimedia.org/wiki/File:Salle_des_%C3%89tats_-_salle_de_la_Joconde_-_Mona_Lisa_room_-_after_2109_renovation.jpg
Taylor, D. (1994). A simple sketch on a TWA napkin by NASA Dryden engineer Frank W. “Bill” Burcham [Photograph, EC94-42805-1]. NASA Dryden Flight Research Center. Public domain, via Wikimedia Commons. https://commons.wikimedia.org/wiki/File:A_simple_sketch_on_a_TWA_napkin_by_NASA_Dryden_engineer_Frank_W_%22Bill%22_Burcham_led_to_development_and_validation_of_the_Propulsion-Controlled_Aircraft_concept_(EC94-42805-1).jpg
Unknown photographer. (c. 1897). Egon Sharpe Pearson; Karl Pearson; Maria Pearson (née Sharpe); Sigrid Letitia Sharpe Pearson [Photograph]. Public domain, via Wikimedia Commons. https://commons.wikimedia.org/wiki/File:Egon_Sharpe_Pearson;_Karl_Pearson;_Maria_Pearson_(n%C3%A9e_Sharpe);_Sigrid_Letitia_Sharpe_Pearson.jpg
Winter, M. (1919). The shepherd boy and the wolf [Illustration]. In The Aesop for Children. Rand McNally. Public domain, via Wikimedia Commons. https://commons.wikimedia.org/wiki/File:The_Boy_Who_Cried_Wolf_-_Project_Gutenberg_etext_19994.jpg