Crown Melbourne, 2013: a camera scam, a reported A$32 million in eight hands, and the edge one card flips

A$32 million in eight hands

Early in February 2013 an overseas high roller was staying with his family in a villa at Crown Towers, the hotel at Melbourne’s Crown casino; villas like it were reported to cost about A$30,000 a night. In a VIP room, over eight hands of cards played in a short time, he won a reported A$32 million.

According to a 2015 account in The Sunday Age, which quoted a former Crown executive, the size of the wins drew attention, and so did bets the executive called “very, very suspicious”: he “bet against the odds and won.” A surveillance staffer reviewed the video. It showed, the paper reported, that the camera network covering the VIP room had been breached and that someone was signalling the player at the table.

In the middle of the night a security guard knocked on the villa door. The family was turned out, a “withdrawal of licence” notice barred the patron from the casino, and he is believed to have gone back to his home country. A VIP services staff member who had been assigned to him was dismissed.

None of it was public until Friday 15 March 2013, when the Herald Sun broke the story several weeks after the event; The Sunday Age later placed the play the day before a casino publicity event on 7 February. The A$32 million, about US$33 million, was the Herald Sun’s figure, and Crown refused to confirm it. It was reported as the biggest scam in the casino’s 19-year history.

The amount is the Herald Sun’s figure, which Crown refused to confirm. Crown said it expected to recover a significant portion; no recovered figure was ever published, and no charges were ever reported.

The casino’s own cameras

The uncomfortable part is whose cameras they were. The Victorian Law Reform Commission, describing the casino from site visits for its 2010 report on surveillance, called Crown’s video system one of the most advanced, complex and comprehensive in Victoria, at a casino with about 6,000 staff and more than 30,000 visitors a day: cameras that pan, tilt and zoom, often several on a single gaming table, extra surveillance in the premium rooms, live monitoring, and real-time access to footage for the regulator.

On the reported account, nobody had to smuggle a camera to the table; the casino had already pointed several at it. Relays of the Herald Sun’s report said an unauthorised person gained, or was given, remote access to the system, and that information about the cards reached the player as he played. A gambling-security consultant told the ABC the player appeared to receive instructions through a wireless transmission to his ear; that was his reading, not a finding. Crown would not confirm that its camera network had been accessed.

Who watched the feed is unresolved. SecurityWeek read the Herald Sun as pointing to an insider. The Blaze wrote that Crown believed the dismissed employee had used the camera system to advise the player; NBC wrote four days later that it was not clear whether that person was part of the plot. In 2015 The Sunday Age reported that Crown eventually worked out that the VIP services manager who had invited the gambler to the casino was in on the scam. That person was never named, and no charges were ever reported against anyone.

Nor was the game ever named. The reports that can be traced say only cards or card tables; the outlets that said poker gave no source, and the one that said blackjack got the date wrong. HuffPost reported the Herald Sun’s crime reporter telling Melbourne radio 3AW that the player had been told about the other players’ cards, which, if accurate, points to a game in which other players’ hands matter.

What one card is worth

A house edge is a statement about what the player does not know. To see what happens when the player knows more than the price assumes, take a game with no decisions and therefore an exact price, baccarat, as a yardstick, not as a claim about what was played at Crown.

Exact enumeration without replacement. A tie returns the Banker and Player stakes. Excluding ties, Banker takes 50.68% of decided hands and Player 49.32%.

Banker’s 1.06% is among the narrowest edges on a casino floor, and no skill beats it, because after the bet there is nothing to decide. Now let someone tell you cards before you bet, and back whichever side they favour.

Exact enumeration of a fresh eight-deck shoe; ties push. 1 card is the first card dealt, to Player; 2 cards are Player’s two; 4 cards are both two-card hands. With every card known the bet wins 95% on a Banker win and 100% on a Player win, and the 9.52% of ties push.

One card is enough. A 9 as Player’s first card lifts Player’s chance of winning from 44.6% to 55.9%; a ten or a picture card lifts Banker’s to 49.5%. Averaged over every possible first card, the better side is worth about +6.8% a hand. The edge does not just shrink towards zero as information arrives. It changes sign at the first card and grows from there, to +88% with a full view, or +164% if a known tie is backed at 8 to 1.

Neither the stakes nor the bets were ever reported, so the scale is a hypothetical. A$32 million over eight winning hands is A$4 million won per hand: A$4 million staked on Player, which pays even money, or about A$4.21 million on Banker, which pays 0.95.

Stakes were never reported; this assumes flat bets. Honest expectation over eight hands at the eight-deck edges: −A$395,226 on Player and −A$356,347 on Banker. The last row divides A$32 million by the house’s expected take per hand.

The eight hands undid what the house edge would have earned from the same player, at the same stakes, over roughly 650 to 720 honest hands. The expected result was a loss of about A$0.4 million; the reported result was a win of A$32 million. If the reports are right, that swing was bought with information, not luck.

The streak was not the tell

Was the run, by itself, evidence of anything? Less than it looks.

Hands treated as independent, which is very close to exact on an eight-deck shoe. End it: a tie breaks the streak. Skipped: only decided hands count. Any mix of Banker and Player bets made without information lands between 1 in 230 and 1 in 636.

That is the neighbourhood of eight heads in a row. A player who bets Banker for 1,000 hands has about a 65% chance of at least one eight-win streak somewhere in them, and at A$4 million a hand on Player one standard deviation over eight hands is about A$10.8 million, so A$32 million sits about three of them above the average (calculations for this article). Rare, not impossible, and proof of nothing on its own.

That is why the detail in the 2015 account matters. The streak alone did not send anyone to the tape; the size of the wins and bets that looked wrong did. A winning streak is a fact about luck. Bets that only make sense if the bettor already knows the result are a fact about information, and an informed bettor cannot avoid placing them, because information is worth nothing unless it changes the bet. The 1980 Pennsylvania lottery fix was caught the same way.

Nothing face down to see

The official record is thin. Victoria Police said it was involved even though it had received no formal complaint, and later, according to The Sunday Age, described the case as an Ocean’s Eleven-style scam. Crown briefed the state gambling regulator, whose spokeswoman could not say whether it was running an investigation of its own; no public finding has been located. A Deakin University academic told the ABC that casinos were under no obligation to report incidents like this publicly.

Strip the story down and it is about where information lives. A card game keeps its edge only while the cards that decide a hand stay unseen by anyone with a stake in it. A casino keeps that promise with people: a dealer, a pit boss and a surveillance room watching both. At Crown, on the reported account, the surveillance cameras were the leak.

An online table removes the object of the leak instead of adding another guard. On Betkyo’s baccarat the bet and the deal are a single request: no felt, no shoe, no card lying face down before the bet. The cards are drawn as the round settles, from the house’s server seed, whose SHA-256 hash is shown before you bet, and your own client seed, which enters every result.

On the demo deal, which draws every card independently, Banker costs 1.06% and Player 1.23%, within a hundredth of a point of the eight-deck shoe above. What a hash commitment proves sets out the limits: the commitment fixes the house’s seed before your bet, and keeping the house’s secrets from its own insiders is still the house’s job, online as at Crown. What the online deal removes is the thing this case turned on: a card on a table, before the bet, for someone to see.