The morning of April 19, 1997, my two partners and I gathered at our headquarters and stared at the phone. Our company had a network of professors and was still looking for the people who would pay to reach them. We were down to our final $10,000, and we had bet all of it on one advertisement in The Wall Street Journal. We had not found our customer, so the plan was to let the customer find us.
The first call came around 9:00. I lunged for the phone. "Russy! How are you honey?" It was my mother. The next several calls were vendors and professors. Around noon a man named Kimball Anderson called from Winston & Strawn in Chicago. He had a patent trial coming and needed computer scientists and economists. He was a litigator, a kind of customer we had not thoroughly considered, and connecting litigators with experts has been my work for the nearly thirty years since.
An advertisement is a door, and the only question is who walks through it. A stranger with a patent trial pays the bills. A mother does not, though my parents covered a few along the way. I could tell the two apart because I answered the phone myself.
I thought of that phone when I read page 8 of a DraftKings investor presentation. DraftKings has spent this year opening a very large door: a prediction market whose sports contracts it offers in states such as California, Florida, Georgia and Texas, where it has no sportsbook. It expects the door to cost $650 million to $700 million this year. It calls the business that came through incremental, meaning new customers and not its own customers moved from one product to another. Its exhibit, on an appendix page, is a pie chart headed "Nearly 70% of Predictions Sports Consumer Volume Comes from States Without Sportsbook". The footnote says the pie is the estimated state mix "for the leading U.S. Predictions Operator per Eilers & Krejcik Gaming". That is another company's volume. DraftKings' exhibit about who is calling is a count of the calls to somebody else's office.
Two questions hide in the word incremental. How many of DraftKings' own Predictions customers were new to the company? And how much additional business did they bring? The answers can differ. A DraftKings fantasy player who spends more after taking up Predictions is not a new customer, and what he adds may still be incremental revenue. A familiar voice can place a new order. The pie answers neither question. It is the leading operator's, and DraftKings calls itself "the clear number three Predictions exchange by Total Volume".
I found the slide because a lawsuit cites it. The idea for this piece came from the public complaint in Chan v. DraftKings Inc., a proposed class action filed July 28, 2026, in federal court in Massachusetts. I read complaints for fun. My day job is running a firm that helps trial lawyers find expert witnesses, all thanks to that 1997 phone call from Kimball, and at Vista Research we bring experts to bear on questions that investors and executives are trying to answer. As of this writing, that firm is not working on this case. The lawsuit is the occasion and not my subject. I take no view of it, or of the arguments in courts and legislatures about prediction markets. The business facts here come mainly from DraftKings' own documents and from eight state regulators.
I also took up this piece because I'm highly intrigued about the business model. My brother Andy used to take me to the Meadowlands when I was a teen in New Jersey where I learned the art and science of horseracing (there is no science), then we upgraded to visits to the Atlantic City casinos (as a young man), and that led to a standing Poker Night with my friends in Parsippany. Once in a while my Dad would join and take everyone's money. So if DraftKings is doing something highly innovative with the games I knew so well in my youth, then I'm intrigued and want to learn more.
DraftKings' filings cannot say who is calling either, because they now count sportsbook and prediction customers together. So we went to the one public source that still counts the sportsbook alone: the state regulators, who publish every operator's bets each month. They can check one piece of the second question: whether the new business came out of DraftKings' own sportsbooks, in the states where it has them. In eight states, DraftKings' share of those bets has not moved beyond chance since the door opened. The newest months point two ways at once. And what I think is the larger part of both questions sits where no regulator can see it.
DraftKings was not always a sportsbook. It was founded in 2011 with, in its own words, "a singular focus" on daily fantasy sports. Until 2018, fantasy was the business. In May 2018 the Supreme Court ruled unconstitutional a law that had prohibited most states from authorizing sports betting. That year DraftKings added a sportsbook and an online casino. It has been publicly traded since April 2020. A lottery courier came in 2024 and Predictions in 2025. The fantasy company, with headquarters in Boston, now calls itself "a digital sports entertainment and gaming company". Jason Robins, its chief executive and co-founder, runs it. It takes sports bets online in 27 states, Washington, D.C., and Puerto Rico, home to about 53 percent of Americans. Its sportsbook took $53.6 billion of bets in 2025, by its annual report.
The 2018 case came out of New Jersey, as I did. It began under Governor Chris Christie and ended as Murphy v. National Collegiate Athletic Association, named for his successor. The law dated from 1992, and one of its sponsors was a New Jersey senator, Bill Bradley, a former basketball star. I met him a few times when I was an intern for my New Jersey congressman, the late Dean Gallo, in 1990. The NCAA and four professional leagues sued the state's governor. The Justice Department argued on the leagues' side under President Trump, as it had under President Obama. The vote was six to three, Justice Alito writing. "The legalization of sports gambling is a controversial subject," the Court said, and it left the choice to Congress or, failing that, to each state. By the end of 2025, 39 states, Washington, D.C., and Puerto Rico had some form of authorized sports betting. Of those 41, 33 had it online.
Predictions launched December 19, 2025. In the company's words it offers "federally regulated event contracts under CFTC oversight", the CFTC being the Commodity Futures Trading Commission. A customer trades a contract on how an event turns out, a game included. "We have the entire other half of the country now open to us," Robins said during the first-quarter call. His words from the calls are quoted here from published transcripts.
In February, two months after the launch, Eric Sheridan of Goldman Sachs asked the question this piece is about. Was Predictions more activity from existing users, or a wider pool of players? Robins answered that "for us, it's about incremental customers and other states." He named California, Texas and Florida. Jordan Bender of Citizens asked who the players were. They "tend to be Californians and Texans", Robins said, and otherwise "look a lot like our existing customers."
A new customer who looks a lot like an old one is the customer every business wants. He is also, now and then, an old one. The first voice on my new phone line was one I had known all my life.
The door was expensive. DraftKings began 2026 guiding to $700 million to $900 million of Adjusted EBITDA, its preferred measure of profit. In November it guided to $175 million to $275 million, after an investment in Predictions of $650 million to $700 million. For 2027 it expects $650 million to $750 million of Predictions revenue.
It has something to show for the money. "More than 1.6 million customers have now engaged with our Predictions offering year-to-date," the November letter says.
If those 1.6 million are strangers, the money bought new customers. If many of them are its own fantasy players and its own bettors, it paid to move them, and what matters is whether they spend more than they did. Anyone who owns the shares or lends to the company has a stake in both answers.
The pie hides nothing. Its footnote is DraftKings' own, and where a rival's volume comes from is fair evidence about the market. It does not show who DraftKings' customers are, or what they added.
DraftKings' own count is not in its filings. Its monthly unique payers include anyone with a paid engagement "across one or more of our Sportsbook, iGaming, Fantasy, Lottery, Prediction Markets or other offerings", and each person counts once. A paying fantasy player who starts trading on Predictions is still one payer. Since the second quarter the company reports a single Sports revenue line, which "includes online sportsbook, retail sportsbook, and Prediction Markets revenue". The company's stated reason is that the combined line reflects how it plans to operate. In the filings, that leaves an outsider no way to watch one product take from the other.
DraftKings does make one claim an outsider can check. It says prediction markets have had "no discernible impact" on its sportsbook. It also says its share of the bets across its sportsbook states rose from a year earlier "for the fourth consecutive quarter", by its reading of state regulatory filings. The regulators publish those filings. So we read them.
The usual tool for a question like this is difference in differences. Vista's Decision Library, our catalog of decision methods, calls it before and after, with someone who did not get it. In 1992 New Jersey raised its minimum wage and neighboring eastern Pennsylvania did not. David Card and Alan Krueger surveyed fast-food restaurants in both states before the raise and again after it. Pennsylvania showed what would have happened anyway. I smiled reading about that study, because my very first job was as a burger flipper at Burger King in New Jersey, earning $3.35 an hour.
Predictions has no Pennsylvania. No sportsbook state was out of its reach, since bettors cross state lines, so none can serve as the comparison. We built the comparison out of DraftKings' own past, twice. The first is the same months a year earlier, which takes out the football calendar. The second is the path its share was already on. Through 2025, before Predictions existed, DraftKings' share in the eight states we measure was slipping by about 0.85 points a year, mostly to Fanatics. A fair test must not blame Predictions for a slide that began without it.
Eight of the states we collected publish what we need: New York, Illinois, Ohio, Massachusetts, Maryland, Michigan, Indiana and Iowa. What we need is online handle, the dollars bet, by operator and by month, with DraftKings named or identified by the regulator. In the year before the launch, bettors in those states wagered $80.3 billion online, and DraftKings took 35.8 percent of it.
For each state, S is DraftKings' share of online handle over a set of months, in percent.
S = 100 x (DraftKings' online handle) / (the state's online handle)
FLAT is the share since the launch minus the share in the same months a year earlier. TREND then takes out the slide. Here b is the state's 2025 trend, in points a year, and g is the gap in years between the months compared, which is one year for now. The eight states are weighted by their handle.
FLAT = S(since the launch) - S(the same months a year earlier)
TREND = FLAT - g x b
Since December 2025, FLAT is plus 0.01 points. DraftKings' share is where it was a year earlier. TREND is plus 0.86 points. Against the path it had been on, DraftKings is doing a little better.
Is 0.86 of a point anything? To find out, we lied to the model. We told it that Predictions had launched on fifteen dates in 2024 and 2025 when nothing launched, in New York and Michigan, and ran the same calculation each time. Nine in ten of the fake launches moved DraftKings' share by 2.22 points or less. The largest moved it 2.83. A sportsbook's share wanders by a point or two a year for reasons that have nothing to do with any launch. Plus 0.86 is inside that band. So is plus 0.01.
I had expected to find something. In these eight states, so far, Predictions has left no mark that stands out from chance, in either direction.
There is a catch in "so far". For most of the months we measured, Predictions was small. The company reports its volume annualized, a month's volume times twelve. From December 2025 through June 2026, consumer volume averaged $0.9 billion on that basis. From July through October it averaged $6.0 billion. October alone ran at $10.1 billion a year, 19 percent of what the sportsbook took in 2025. Most of the traffic through the door arrived after most of our measurements.
So take the newest months alone. From July 2026 to the latest month each state has published, DraftKings' share ran 0.60 points below the same months a year earlier. It was lower in six of the eight states. Against its trend, the figure is plus 0.25. The two readings point opposite ways, both sit inside the band, and four months, at most, is not many. They are also the months in which Predictions grew to ten times its earlier size.
All of that concerns one piece of one question. A sportsbook report cannot see inside the states where DraftKings has no sportsbook, and I think most of the answer is there.
DraftKings had customers in such states long before Predictions. It offers fantasy contests in 44 states, many more than the 27 where it takes bets online. In February, Robins described the opening as a sports product that could monetize and engage customers "in ways that we never were able to with fantasy sports." That describes an existing customer moved to a product that earns more from him. He may be additional business. He is not a stranger. It is my mother, calling the new number.
So the data that we're missing is DraftKings' own. How many of its 1.6 million Predictions customers were already its fantasy players or its bettors? What did those customers spend before, and what do they spend now? What is its own state mix, in place of the leader's?
Likely, we can get some of it from primary research, which is what Vista is all about. I'd ask someone who has run customer marketing at a fantasy operator what share of the people it signs up for event contracts were already its fantasy players, and how their spending changed. I'd ask a former analytics lead at a state gaming regulator how much of a border state's online handle came from across the line. Those answers would inform a follow-up. They never decide the test.
Here is our hypothesis, and it is only that. It is DraftKings' own claim, narrowed to what can be checked: Predictions is not taking business from DraftKings' sportsbooks in the states where it has them. The test uses the same eight states and the regulators' reports for November 2026 through June 2027. None of those months had been published when we froze the test on November 14, 2026. Each month is compared with the same calendar month before Predictions existed. If DraftKings' weighted share is at or above its old level, and also at or above the path its 2025 trend pointed to, the claim is confirmed. If the share is 2.5 points or more below both, the claim is falsified. Anything between is cannot tell. We will score it in the first week of September 2027. The full wording, the weights and the conditions that void it are in "For the record", with a fingerprint that changes if a single word does.
The verdict will be no wider than the measurement. It can say whether DraftKings' sportsbook share held up against two benchmarks from its own past, in eight states. It cannot say who the Predictions customers are, or what they added. It can even miss a bite, because share is a zero-sum measure. FanDuel's share, in the seven states that name it, fell from 35.4 percent in 2024 to 31.1 percent since the launch. If it lost about 4.7 more points than it otherwise would, spread in proportion over everyone else, a 2.5-point bite out of DraftKings would not show. A sportsbook's share can hold while its wagers move to Predictions. So the comparison with BetMGM and Caesars, which have launched no prediction product, will be printed beside the verdict, unscored.
Why 2.5 points? It is beyond what nine in ten of the fake launches produced, and it is real money. A 2.5-point fall means DraftKings lost about 10.5 percent of its bets in these states, if the bets left sportsbooks altogether. If it lost the same share in every state where it takes bets, that is about $362 million a year of net revenue, about half the $700 million midpoint of its 2027 Predictions guidance. That illustrates scale. It is no forecast and no call on the stock. The arithmetic is in the record.
We amended the test once, the day we froze it, over a number we had wrong. New York's report for August 2025 prints Fanatics at $565.8 million of bets, against $137.7 million in July. We took it for a misprint and smoothed it. An outside reviewer objected. New York's weekly report shows the bets were real: $147.3 million, $212.7 million and $145.8 million in the weeks ending August 17, 24 and 31, about $505.9 million in three weeks. The published figure now decides the verdict, and it lifts today's readings and the newest months, whose year-earlier months include that August. It steepens DraftKings' 2025 slide and widens the noise band, and since the test will not charge that slide to Predictions, a fall must now exceed about 4.1 points, rather than 3.7, to falsify. The amendment can turn a falsified verdict into cannot tell and never the reverse. We made it because the published data required it, not because it helps our odds. The frozen text stays in the record beside it, so anyone can compare.
Now the odds, in the order they matter. We think there is about a 55 percent chance that the test reaches no verdict at all. If it does reach one, we think there is about an 80 percent chance that it confirms the claim. Both numbers are judgment, and they describe the test, not DraftKings' business. Part of the 80 is the design, which makes falsifying harder than confirming. Confirming takes a share at or above its old level. Falsifying takes that 4.1-point fall, which is 2.5 points beyond where the 2025 slide alone would leave the share. We accept that because the slide began before Predictions existed, and a test that blamed Predictions for it would be measuring 2025. The price is a test that says cannot tell more readily than it says we were wrong. For the claim: the company offers its sports contracts in states such as California and Texas, where it has no sportsbook. Today's readings sit inside the band. The company reports four straight quarters of share gains.
This is our own argument against ourselves. No outside skeptic has signed it.
DraftKings' share may be holding only because FanDuel's is falling. Against BetMGM and Caesars, its handle has run about 3 percent behind its earlier pace since December.
Predictions is also ten times bigger than it was for most of the months measured so far. A bite that grows with the product would land inside the test window. And a 4-point bite, with the share otherwise holding level, would be caught only about half the time.
And if most of the new Predictions customers in California and Texas were already DraftKings fantasy players, our test could come back confirmed and tell us nothing about it.
DraftKings did not open this door alone. Fanatics launched Fanatics Markets on December 3, 2025. FanDuel launched FanDuel Predicts on December 22, three days after DraftKings. BetMGM and Caesars have stayed out, which makes them the nearest thing this industry has to Pennsylvania.
The question is older than prediction markets. Whenever a company opens a new door beside an old one, the new door gets credit for everyone who walks through it, including those who would have come through the old one anyway.
PENN Entertainment has casino floors and an online casino. During its first-quarter 2025 call, its chief executive, Jay Snowden, said that "70% of our standalone iCasino theoretical revenue is coming from incremental sources, including newly acquired customers, digitally reactivated players, and players who were previously retail only customers." Retail is the casino floor. So PENN's incremental includes its own casino customers, new only to the online door. One outside count is in Pennsylvania, the real one this time. Its Gaming Control Board publishes, every month, casino-floor and online revenue side by side.
The Michigan Lottery began selling over the internet in August 2014. A 2022 study, as quoted by Mark Hichar in Public Gaming International, found that Michigan's retail lottery sales grew 91 percent from 2013 to fiscal 2021. The study was prepared for NeoPollard Interactive, an online lottery vendor. The Lottery's own audited report gives its sales every year, with online instant games shown apart from tickets sold in stores. For fiscal 2025 it attributes declines in instant tickets, in stores and online, to competition from casinos offering "sports betting and online gaming" and to inflation.
I ask the same question of my own firm. We are trying a new way to bring in expert searches. Suppose a lawyer I have worked with for years brings us a case through it. The new channel gets to count that search. I still have to ask whether the phone would have rung anyway.
A company's own count cannot answer that. Find a count the company does not control. Set it against the company's own past and against someone who opened no door. Ask how far the number moves when nothing has happened.
If you have run customer marketing at a sportsbook or a fantasy operator, analyzed handle for a state regulator, or built estimates of prediction market volume, and you can see where we are wrong, write to director@vistaresearch.ai. We print what we hear, with the date.
In 1997 I learned who my customer was by picking up the phone. The first voice was my mother's, and the one that mattered came at noon. DraftKings expects to put $650 million to $700 million this year into making the phone ring, and by its own count more than 1.6 million people have called. The first month our test will score is the one I am writing in. By next September the regulators can tell us whether the old phones went quiet. I'd still like to know who is on the new line.
DraftKings says more than 1.6 million customers have engaged with Predictions this year, but its filings count sportsbook and prediction customers together, so they cannot say how many were new to the company. A customer who already played its fantasy contests and spends more through Predictions may be additional business without being a new customer. The public record cannot settle that part; the people who know are DraftKings and those who have run customer marketing at sportsbook and fantasy operators.
DraftKings' own claim that prediction markets have not hurt its sportsbook. State regulators publish every operator's online sports betting handle each month, so DraftKings' share of the bets in the states where it has a sportsbook can be measured against its own past. Vista uses eight states, weighted by their handle.
For November 2026 to June 2027, DraftKings' share in the eight states is compared with the same months a year earlier (FLAT) and with that comparison adjusted for its 2025 trend (TREND). If both are at or above zero, the claim is confirmed; if both are 2.5 points or more below, it is falsified; anything else is cannot tell. The test was frozen on November 14, 2026, and is scored in the first week of September 2027 from the regulators' reports as posted by August 31, 2027.
We think there is about a 55 percent chance that the test reaches no verdict at all. If it does reach one, we think there is about an 80 percent chance that it confirms the claim. Both are judgment. Part of the 80 is the design, which makes falsifying harder than confirming: confirming takes a share at or above its old level, while falsifying takes a fall of 2.5 points or more on both readings.
We first smoothed one New York figure, Fanatics' August 2025 handle; New York's own weekly report shows the spike was real. The amendment, made on November 14, 2026, before any month the test scores had been reported, uses the published figures. It can turn a falsified verdict into cannot tell and never the reverse, and the frozen text is printed beside it.
What follows is the test exactly as frozen, the one amendment made to it, what its two lines mean in practice, how large a move chance alone produces, whether a real bite could pass the test or plain noise could fail it, how the data were assembled, and what was verified.
We froze the test on November 14, 2026, as plain text inside model.py, the program in the folder for this hypothesis that computes every figure derived here. The program refuses to run if a character of the text changes. The fingerprint, the SHA-256 of the text's UTF-8 bytes, is 7816a0c3b4e86c66a8edaf4f6b90c0c8ece0022b3dfc2df648d56bd584db8ef7. Most numbers typed into the wording, from the weights to the dollar figures, are checked in code against the calculation that produced them. The wording follows, as frozen. Where the amendment printed after it differs, the amendment governs.
DRAFTKINGS PREDICTIONS HYPOTHESIS: THE DATED TEST
Frozen 14 November 2026, before any of the eight regulators had published a test month (latest month posted: October 2026 in New York, Indiana, Iowa and Maryland; September 2026 in Michigan, Ohio and Massachusetts; August 2026 in Illinois).
CLAIM. DraftKings Predictions is not taking business from DraftKings' own sportsbooks in the states where DraftKings has them. In November 2026 to June 2027, DraftKings' share of online sports betting handle in the eight states below is at or above both the level and the path it was on before DraftKings Predictions launched on 19 December 2025. The test measures sportsbook handle in sportsbook states only; it cannot show whether Predictions customers in other states are new to DraftKings.
STATES AND WEIGHTS. New York 0.323, Illinois 0.192, Ohio 0.127, Massachusetts 0.104, Maryland 0.081, Indiana 0.070, Michigan 0.069, Iowa 0.034: each state's share of the eight states' combined online handle in December 2024 to November 2025, fixed now. Excluded: Pennsylvania (its regulator's handle reports go by casino license and do not name DraftKings), New Jersey and Tennessee (no handle by operator).
MEASURE. For a state and a set of months, S = 100 x (DraftKings' online handle summed over the months) / (the state's online handle summed over the same months). The test months are November 2026 to June 2027. Each is matched to the same calendar month in December 2024 to November 2025: November 2026 to November 2025, December 2026 to December 2024, January to June 2027 to January to June 2025. For each state, over the test months published by the data date and their matched months only:
FLAT = S(test months) minus S(matched months).
b = S(January to November 2025) minus S(January to November 2024), the state's pre-launch trend in points a year.
g = the average gap in years between those test months and their matched months (1.875 when all eight are published).
TREND = FLAT minus g x b.
The eight-state FLAT and TREND are the weighted sums of the states' values, with the weights rescaled to sum to 1 over the states scored, each rounded to 0.1 point before comparison.
VERDICTS. CONFIRMED: FLAT and TREND are both at or above 0.0 points. FALSIFIED: FLAT and TREND are both at or below minus 2.5 points. CANNOT TELL: anything else. A verdict in either direction must hold whether or not DraftKings' 2025 share trend would have continued. The 2.5-point line is a materiality judgment set beyond the chance moves found at fake launch dates (90 percent of them within 2.0 points). In the base year DraftKings held 36.0 percent of the eight states' online handle. A 2.5-point fall means it lost about 10.4 percent of its handle there if the lost bets left the sportsbooks altogether, as bets moved into a prediction app would, or 6.9 percent if they went to rival sportsbooks; applied to its 2025 sportsbook handle of $54 billion at its 6.4 percent third-quarter 2026 sportsbook net revenue margin, that is about $361 million or $240 million a year, about half or a third of the $700 million midpoint of its 2027 Predictions revenue guidance. A smaller bite can hide in the noise.
PRINTED BESIDE THE VERDICT, NOT SCORED. (1) DraftKings' handle against BetMGM plus Caesars, which had no prediction product: FLAT and TREND computed the same way on 100 x the natural log of the ratio of their handles. (2) DraftKings against FanDuel, which launched its own prediction product on 22 December 2025: two treated operators, not a control. (3) Illinois, Iowa and Michigan, which border states with no legal online sports betting, against the other five.
DATA RULES. (1) Figures are the latest each regulator has posted by 31 August 2027, revisions to the matched months included; the weights do not change. (2) Online handle only. The state's online handle is: New York, the statewide mobile sports wagering handle; Illinois, the sum over licensees of online Tier 1 plus Tier 2 handle in the All Wagering Activity Detail report; Ohio, Subtotal Type A Online; Massachusetts, the sum of the online operators' Monthly Handle (tickets settled); Maryland, Total Mobile (the sum of the mobile licensees where none is printed); Michigan, All Internet Sports Betting Operators; Indiana, the sum of the online skin rows on the sports wagering detail sheet; Iowa, internet handle. (3) DraftKings is: New York, DraftKings; Illinois, Casino Queen, Inc.; Ohio, the Hollywood Toledo row whose service provider is DraftKings; Massachusetts, DraftKings; Maryland, Draft Kings; Michigan, the DraftKings platform of the Bay Mills Indian Community; Indiana, the Ameristar skin Sportsbook.DraftKings.com; Iowa, Crown IA Gaming, LLC. If a regulator relabels DraftKings, the regulator's own brand list decides; a state where DraftKings cannot be identified drops out. (4) New York, August 2025: the Fanatics handle as printed, $565.8 million, 4.1 times its July figure and 2.6 times its September figure, is replaced in the state total by the average of Fanatics' July and September 2025 handle, unless the Commission revises it by the data date. (5) In a test month, an operator's figure more than 2.5 times the larger, or less than 0.4 times the smaller, of its own figures in the months before and after, for an operator with at least six months of prior reports, and moving the state total by more than 2 percent, is replaced by the average of those two months unless the regulator has revised it. (6) A state with fewer than six published test months, or whose regulator changes its handle definition without restating the matched months, drops out.
VOID. The test is void if the states scored hold less than 75 percent of the weight, or if any of these happens before 30 June 2027: (1) DraftKings offers sports event contracts to customers located in one of the eight states (that state drops out first, then the weight rule applies); (2) DraftKings stops offering sports event contracts nationally, or says it has stopped in states that hold most of its Predictions business, for 60 days or more after 1 November 2026, for any reason; (3) DraftKings merges with, acquires or is acquired by a company that runs an online sportsbook in any of the eight states; (4) an operator with at least 5 percent of the eight states' online handle in the base year stops taking online bets in states holding at least a quarter of the weight.
SCORING. In the first week of September 2027, from the regulators' monthly reports for November 2026 to June 2027 as posted by 31 August 2027.
PROBABILITY. We think there is about an 80 percent chance that the test, if it reaches a verdict, confirms the claim, and about a 50 percent chance that it reaches none. This is judgment.
We amended the test once, on November 14, 2026, the day it was frozen and before any month it scores had been reported. The amendment is held in model.py the same way, and the program refuses to run if a character of it changes. Its fingerprint, the SHA-256 of its UTF-8 bytes, is 0b0575e3c301eeb607c670b239d0a3d2d8fe6015e91e0bdb3dba366aba588f07. It follows, word for word.
Amendment 1, dated November 14, 2026, the day the test was frozen. Made after an outside review, and before any month the test scores had been reported: November 2026, the first, had not ended, and no regulator had posted a month later than October 2026. The regulators' figures are used as they print them in the result that decides the verdict. New York's August 2025 Fanatics handle of $565.8 million, and the statewide total that includes it, count as the Commission publishes them, under data rule 1. Data rule 4 no longer replaces them: the replacement it prescribed becomes a sensitivity, scored the same way and printed beside the verdict, and it never decides the verdict. Data rule 5, the spike rule, no longer alters any figure. The stated chance that the test reaches no verdict moves from about 50 percent to about 55, because the judgment mixture behind it, its weights unchanged, gives 55 percent with the figures as printed against 52 with the replaced figure; the chance that a verdict confirms stays about 80 percent. Nothing else changes: the claim, states, weights, measure, months, lines, comparisons printed beside the verdict, void rules, data date and scoring stand as frozen above, and the frozen text is unchanged. Why: the figure is no misprint. The Commission's weekly report for Fanatics records $147,334,258, $212,741,600 and $145,833,764 of handle in the weeks ending August 17, 24 and 31, 2025, $505.9 million in three weeks of the month's $565.8 million, against $137.7 million in all of July. Gross gaming revenue in those weeks was only about $2.7 million, $0.4 million and $14.8 million, which suggests heavy promotional volume at a thin margin, but wagers all the same, recorded week by week. And a rule written after seeing the one figure it catches is the kind of choice a frozen test exists to prevent. The effect, on the figures posted today: August 2025 is not a matched month, so FLAT is unchanged. It is in New York's 2025 trend, which moves from minus 0.07 to minus 0.65 points a year, and the weighted trend moves from minus 0.66 to minus 0.85, so with all eight test months published TREND reads 1.59 points above FLAT, not 1.24. CONFIRMED still turns on FLAT at or above 0.0. FALSIFIED, which also needs TREND at or below minus 2.5, now needs FLAT at or below about minus 4.1, not about minus 3.7. The amendment can turn a falsified verdict into cannot tell and never the reverse, and it can neither create nor remove a confirmation: it makes the test harder to falsify. The frozen weights and the frozen description of the 2.5-point line were computed with the replaced figure and stay as written. With the figure as printed, New York's weight would have been 0.326, not 0.323; 90 percent of the fake launches move the share by 2.2 points or less, not 2.0; and DraftKings' base-year share is 35.8 percent, not 36.0, so a 2.5-point fall is 10.5 or 7.0 percent of its handle, about $362 million or $241 million a year.
Every verdict must hold on both readings, FLAT and TREND. When all eight test months are published, the months compared sit an average of 1.875 years apart, and the weighted 2025 trend, with the figures as printed, is minus 0.85 points a year. So TREND is FLAT plus 1.59. Confirmed therefore turns on FLAT at or above 0.0. Falsified turns on TREND at or below minus 2.5, which is FLAT at or below minus 4.09, unless revisions move the trend. The 2025 slide cannot be mistaken for a bite. The price is that falsified takes a bite of about 4.1 points if the slide had stopped, or 2.5 points if it would have continued. A 4.1-point bite is about 17 percent of DraftKings' handle in these states, if the lost bets left the sportsbooks.
Two dry runs on months already published, which are not the score. December 2025 to June 2026 against a year earlier reads FLAT plus 0.2 and TREND plus 1.0, confirmed. July and August 2026 read FLAT plus 0.6 and TREND plus 1.5, confirmed. In that second run August 2025 is itself a matched month, so the smoothed figure changes the reading, to FLAT minus 0.7 and TREND 0.0, cannot tell. In the test itself it is not a matched month.
Nine stories, scored at their expected values. No bite and the old level holds: FLAT 0.0, TREND plus 1.6, confirmed. No bite and the 2025 slide continues: minus 1.6 and 0.0, cannot tell. No bite and DraftKings recovers half a point a year: plus 0.9 and plus 2.5, confirmed. A 2.5-point bite and nothing else moves: minus 2.5 and minus 0.9, cannot tell. A 2.5-point bite while the slide continues: minus 4.1 and minus 2.5, falsified. A 2.5-point bite during a recovery of half a point a year: minus 1.6 and 0.0, cannot tell. A 1.5-point bite: minus 1.5 and plus 0.1, cannot tell. A 4.0-point bite: minus 4.0 and minus 2.4, cannot tell. A 2.5-point bite hidden by FanDuel losing 4.7 more points, spread over everyone in proportion: 0.0 and plus 1.6, confirmed.
The last story is the blind spot named beside the verdict. In 2025 it did not work that way: FanDuel lost share and DraftKings lost share too, to Fanatics. A bite shared by all three launchers is the share measure's weak side. If DraftKings alone lost 10.5 percent of its handle to something outside these sportsbooks, its share would fall 2.50 points. If FanDuel and Fanatics lost the same 10.5 percent to their own apps, it would fall only 1.05. Against BetMGM plus Caesars the loss would show in full, which is why that comparison is printed beside the verdict.
The two yardsticks today. Measured against BetMGM and Caesars together, DraftKings' handle since December has run 2.98 percent behind its earlier pace, worth about 1.07 points of share. Measured against FanDuel, which launched its own prediction product, DraftKings ran 2.18 percent ahead. The two point in opposite directions, and neither reading is outside chance. The yardsticks themselves moved. BetMGM's share of New York jumped to 10.9 percent in December 2025, from about 7 before. So did the market around them: Fanatics went from 2.8 percent in 2024 to 8.3 since the launch.
The 2.5-point line in dollars, with the figures as printed. From a 35.8 percent share, a 2.5-point fall means DraftKings lost about 10.5 percent of its handle in these states, if the lost bets left the sportsbooks altogether. Suppose it lost the same share in every state where it takes bets. Its 2025 handle was about $54 billion, and the sportsbook's third-quarter margin was 6.4 percent. That comes to about $362 million a year of net revenue. If the lost bets went to rival sportsbooks instead, a 2.5-point fall is 7.0 percent of its handle in these states, about $241 million a year on the same arithmetic, about a third of the $700 million midpoint.
Fake launches in time. The exact calculation, seven months after and an eleven-month trend before, was run at fake launch dates from September 2024 to May 2025 in New York and from December 2024 to May 2025 in Michigan. Fifteen fakes: a standard deviation, the usual measure of spread, of 1.14 points, 90 percent of moves within 2.22, the largest 2.83. The real readings for those two states through June 2026, minus 0.78 for New York and minus 0.48 for Michigan, sit inside. Earlier New York dates, whose trend windows reach back into the state's 2022 launch year, read minus 5.94 to minus 3.18. A launch-year climb extended as a trend manufactures a large false effect, so only mature years are used.
All eight states, short windows. With three-month windows and fake launches from April to September 2025, 48 state-level fakes have a standard deviation of 1.79 and a largest move of 4.89; the six pooled fakes run from minus 2.48 to plus 1.67, and the real launch on the same windows reads plus 0.78. With four-month windows, 32 state-level fakes have a standard deviation of 1.64 and a largest move of 4.22; the four pooled fakes run from minus 1.80 to plus 1.43, and the real launch reads plus 0.88.
Other operators at the real date. A related calculation, on the log odds of each operator's share, averaged over the states and converted to points of DraftKings' share, gives BetMGM minus 0.94, Caesars plus 0.80, FanDuel minus 1.03 and DraftKings plus 1.00. Shifts of about a point either way are ordinary for an operator in a year.
The noise used to try the test. The weighted share was modeled as a level that drifts plus month-to-month noise, fitted to the 2025 year-over-year changes: drift of 0.216 points a month, 0.75 a year, and month noise of 0.937. That model is too calm. It implies a spread of 1.15 and 1.06 for the pooled three-month and four-month fakes, against 1.52 and 1.35 observed. So the central case scales it by 1.3, the calm case uses it as fitted, and the stress case doubles it.
Pure noise. A chance move against DraftKings as large as 90 percent of the fake launches, 2.22 points, lands in cannot tell whether or not the slide continues. The largest fake move, 2.83, lands in cannot tell if the old level holds and in falsified only when stacked on a continuing slide. A chance move in DraftKings' favor lands in confirmed, the reading the claim predicts.
Simulation, 5,000 runs a case, with the bite growing as Predictions grew from December 2025 to its October 2026 size. With central noise, a 2.5-point bite is wrongly confirmed in 4.0 percent of runs if the level otherwise holds and in 0.1 percent if the slide continues. With no bite, the test wrongly falsifies in 0.2 percent and 3.9 percent. The test can fail in three places. A share that would otherwise have recovered half a point a year: a 2.5-point bite is confirmed in 13 percent of runs. A smaller bite: 1.5 points with the level holding is confirmed in 14 percent. Stress noise: a 2.5-point bite with the level holding is confirmed in 12 percent, and no bite with the slide continuing is falsified in 13 percent.
How often a real bite is caught. With central noise and the level otherwise holding, a 2.5-point bite is falsified in 14.0 percent of runs and ends in cannot tell in 82.1 percent. A 4-point bite is falsified in 48.7 percent if the level holds and in 87.0 percent if the slide continues. The test is far likelier to say cannot tell about a real bite than to falsify when there is none.
No verdict is a likely outcome, and in the mixture below it is the likeliest. With no bite and central noise, the test says cannot tell in 48 percent of runs if the level holds, 83 percent if the slide continues and 24 percent if the share recovers.
The probability. Our mixture of these cases, weighted by judgment, gives confirmed 37 percent, falsified 8 percent and cannot tell 55 percent, which is 81 percent confirmed if there is a verdict. With the smoothed figure the same mixture gave 52 percent, which the frozen wording states as about 50; the amendment moves the stated figure to about 55 and keeps the 80. The number leans on the design: falsified needs a drop of 2.5 points beyond a continued 2025 slide, so a verdict, when one comes, is more likely to confirm.
The eight states, as each regulator labels DraftKings: New York, "DraftKings", with its own file; Illinois, "Casino Queen, Inc.", by the Gaming Board's brand table; Ohio, the "Hollywood Toledo" row with DraftKings as service provider; Massachusetts, "DraftKings"; Maryland, "Draft Kings", its own license; Michigan, the DraftKings platform of Bay Mills; Indiana, the Ameristar skin Sportsbook.DraftKings.com; Iowa, "Crown IA Gaming, LLC", doing business as DraftKings on the commission's licensing page. The data run from January 2024 to October 2026 in New York, Indiana, Iowa and Maryland, to September 2026 in Michigan, Ohio and Massachusetts, and to August 2026 in Illinois. The weights, each state's share of the eight states' online handle from December 2024 to November 2025, are in the frozen wording.
By state, since the launch, FLAT and then TREND: New York minus 0.84 and minus 0.19; Illinois plus 1.34 and plus 3.20; Ohio minus 0.09 and plus 0.94; Massachusetts plus 0.38 and plus 0.74; Maryland minus 0.62 and minus 0.09; Michigan minus 0.67 and minus 1.40; Indiana minus 0.22 and plus 0.90; Iowa plus 3.15 and plus 4.39. Counting each state once, TREND averages plus 1.06 with a 90 percent interval of minus 0.20 to plus 2.32. The two halves of the period: December 2025 to June 2026, FLAT plus 0.16 and TREND plus 1.01; July 2026 to the latest month, FLAT minus 0.60 and TREND plus 0.25.
Iowa and Illinois. DraftKings' share in both dipped in early 2025 and recovered from August 2025, before the launch. Iowa read 39.5 percent in January to November 2024, 36.8 in January to July 2025, 40.3 in August to November 2025 and 41.0 since the launch. Illinois read 37.2, 34.8, 36.2 and 36.3. Their gains from a year earlier are recoveries already under way. Without the two states, the other six read FLAT minus 0.46 and TREND plus 0.12.
Border states. Illinois, Iowa and Michigan, the three the frozen wording names as border states, read plus 2.26 on TREND against plus 0.27 for the other five. Illinois and Iowa carry the 2025 dips just described. Michigan, the clean case, read minus 1.40. We see no sign of a cross-border pull and no proof against one.
Left out. Pennsylvania's regulator reports handle by casino license and never names DraftKings in its handle workbooks. If one license often reported in the press as DraftKings' is read that way, Pennsylvania shows FLAT plus 4.61 and TREND plus 4.06, which would pull our estimate up; it is in no estimate here. New Jersey and Tennessee publish no handle by operator. Other states were not collected.
The August figure, both ways. New York's file for August 2025 prints Fanatics at $565.8 million, against $137.7 million in July and $219.0 million in September, and the statewide total of $2,044.5 million includes it. As printed, DraftKings' share that month is 28.3 percent. The frozen rule took $387.5 million out of the total, which made it 35.0. Under the amendment the printed figures decide, and the smoothed total is a sensitivity, printed beside the verdict. With the smoothed total, the weighted 2025 trend is minus 0.66 points a year, today's readings are FLAT minus 0.18 and TREND plus 0.48, the newest months read FLAT minus 1.14 and TREND minus 0.48, lower in seven of the eight states, nine in ten of the fake launches fall within 1.97 points, and falsified needs FLAT at or below minus 3.74. Today's readings and the newest months read better as printed because their year-earlier months include that August; the test's matched months do not. Over every FLAT from minus 6.00 to plus 3.00, the two versions of the test give different verdicts only from minus 4.03 to minus 3.69: falsified as frozen, cannot tell as amended.
Two choices are ours and not the regulators': Indiana's Caesars is taken to be the skin its report labels WilliamHill.com, and Maryland, which does not name FanDuel, is left out of the FanDuel comparison, which uses seven states. Both touch only the unscored comparisons.
The other exhibit. DraftKings' second-quarter letter reported "only about 1% customer overlap" between DraftKings Sportsbook and the largest prediction market operator, in sportsbook states, from card deposit data. That counts DraftKings bettors who also use the leader. It does not count DraftKings Predictions customers who were DraftKings customers already.
What was verified. Vista's AI research desk computed in code every figure derived from the regulators' reports and DraftKings' documents, and checked the series against the regulators' own files, every month in six states, DraftKings' figures in Massachusetts and October 2026 in Iowa; the largest difference was $0.00. New York's three weekly figures were read from the Commission's weekly report for Fanatics on November 14, 2026, and their sum is computed in code. A separate fact-check reopened DraftKings' letters, presentations, filings and releases, the slide on page 8 in DraftKings' own PDF among them: the $54 billion of 2025 handle is its finance chief's rounding of the $53.6 billion in its annual report, and the 6.4 percent is the sportsbook margin stated in the November letter. Not verified against an official record: the words spoken on the earnings calls, which are quoted from published transcripts, and my account of 1997, which is from my book.
This hypothesis is research, not investment advice. It states what the filings, the public record and the models show under labeled assumptions; nothing here is a recommendation to buy, sell or hold any security, and the decision belongs to the reader. The lawsuit named near the top is the occasion of this hypothesis and not its subject: nothing here predicts or comments on its outcome or on any other case, says whether any statement was true or false, says whether any product is gambling or lawful, takes a side in any dispute between state and federal authorities, or characterizes any party, lawyer or expert, and the business facts come mainly from DraftKings' own documents and the state regulators' reports.
Russ W. Rosenzweig, Vista's founder, owns no shares of DraftKings, Flutter Entertainment (the owner of FanDuel), MGM Resorts International and Entain (the two half owners of BetMGM), Caesars Entertainment, PENN Entertainment or Restaurant Brands International (the owner of Burger King), as of October 4, 2026, and holds no options on any of them, as of November 12, 2026. Holdings through mutual funds and exchange-traded funds are not counted.
How this hypothesis was made: Written with the help of AI tools. The facts were gathered from the filings, releases, presentations and call transcripts named in the Sources and from the state regulators' own files, the model was built and every derived figure computed in code, the test was frozen, and amended once the same day, before any month it scores had been published, and the draft was fact-checked and read for plainness before I read it. I read it and listened to it before it was published. Why I write with AI. Why I publish these.