Colin Jones (S1 E5): Dark Matter

As early as page 11 of The 21^st-Century Card Counter, Colin Jones mentions the monolithic truth of the universe: “the team’s performance was consistently lower than the math predicted.” Such has been the experience of every team in the history of AP, and every solo card counter, too.

When teams look at their spreadsheets and see the stark gap between AV (Actual Value) and EV (Expected Value), they have a puzzled look like this is some great mystery. The only mystery is why rookie teams ignore the answer that I’m about to explain for the nth time. [PRO TIP FTW: use “nth” the next time you play Hangman.]

There are four answers given for “underperformance.” The first is the scapegoat given by APs in denial (usually the first three years of a career, and extending to an entire career for the degen-cum-fake-AP): variance. I would like to make a word cloud of the online posts made by rookies, and compare it to the word cloud of successful veterans. In the rookie’s word cloud: “EV”, “one spot or two”, “side count”, “cheating”, “optimal bet ramp”, “6:5”, “side bet”, “VARIANCE”, “3 s.d.”, “facial recognition”. In the veteran’s word cloud: “exposure”, “BP”, “chip inventory”, “CTR”, “phone call”, “6:5”, “verbals”, “NERSESIAN”.

By aggregating the performance of many players, teams like CJ’s can get a big enough sample size to see that underperformance isn’t just bad luck. The AV line on the graph is consistently below the EV line, and the gap just widens. I have an announcement, my readers: The time has come …

IT’S TIME TO REJECT THE NULL! The “unlucky” players have an implicit null hypothesis (“the null”) that their hourly EV is a certain amount, say, $100/hour. When “bad luck” inevitably occurs (of course, they call this “negative variance”), they calculate and re-calculate and re-calculate the EV of different scenarios and game conditions to answer the question “How unlucky was I?” At this point I can’t say they are mis-using software, because the software serves its purpose if the AP now draws the right conclusion from what the software is saying. The software says, “If your null hypothesis is true—meaning you really are playing a $100/hour game—then you have apparently suffered a -3 s.d. event. You are 3 standard deviations below EV.”

At this moment, a data scientist with no ego in the game would say, “Hmm, I doubt that I happen to have observed a -3 s.d. event. More likely, my null hypothesis isn’t true.” The in-denial “AP” says, “I’m the unluckiest player who ever played this game. You guys have no idea what it’s like to be this unlucky. The software won’t tell me whether I was born unlucky or whether it’s something I contracted by being around all these losers in the casino.” Will it take a 4-s.d. event for these APs to reject the null? For most purposes, scientists reject the null at 3 s.d. (or 5 s.d. for some applications where life-and-death might be in play), and they look for a hypothesis that better explains the data, such as: “My EV is $50/hour.”

When you cut the EV in half, suddenly the graph looks perfect! Voila! Mystery solved! TML.

So our question evolves from “Why is our AV so far below EV?” to “Why is our real-world EV so far below the on-paper/computer EV?” This question is also not a mystery. I promised four sources of underperformance, and we dispatched the first—variance—as bogus. But the next three are real, though generally unseen (hence “dark matter”).

A real issue facing every team is skimming. Its many forms are rampant in the AP community. I know you may not believe that, and I didn’t either, but when your data sample grows as big as mine, you, too, will accept skimming as an inconvenient truth. (The response, “That’s why I play solo,” is an overreaction taken mainly by rationalizing, arrogant, social misfits.) I’ll have much more to say about skimming in later posts, but as far as CJ’s book, I wish CJ had a chapter about it. Not only do I have a voyeuristic curiosity, but it might benefit all of us to see how a sophisticated AP team deals with the problem.

That said, I understand that CJ wouldn’t want to inflame tensions within the AP community (we all know each other here) by publicly outing ex-teammates who are suspected of skimming. Every author has a vision of what the book should be. If the author’s vision is “uplifting, inspiring docudrama” (is that what the Bible is supposed to be?), then we can forgive the omission of dirty laundry. (But maybe a paragraph or two in the next edition discussing skimming in the abstract? Just throwin’ that out there.)

Anyway, spanning all AP teams, maybe skimming accounts for 10% of real-world underperformance, maybe none if you have a solid crew, maybe more if you’re the West Coast Grinders (who knows? No one talks about WCG.) Let’s get to the bigger causes of underperformance.

Many card counters obsess over bet spreads, finding favorable rules, and playing with cover. And those are all worthwhile. But very few card counters I’ve met consider the massive impact of rounds per hour.

Yeah, I should have put quotation marks around that last paragraph. It’s from p. 124 of CJ’s book, but he nailed it so hard there, that I thought plagiarizing it was the move.

APs run sims assuming 100 rounds/hour for blackjack, and perhaps 50 rounds/hour for carnival games. Where did those numbers come from? They make the arithmetic simpler. That’s like saying, “let’s just use 3 for the value of pi, because it makes the arithmetic simpler. Actually, the value 2 is easier still.” But those benchmark figures could be way off for the game at hand. For carnival games, sometimes only 20 rounds/hour is realistic, with sustained 50-60 rounds per hour possible only under the juiciest conditions—a heads-up game where the dealer is maxing out the machine (the hand is over and the dealer has to wait for the machine to finish shuffling the other deck), with no fills, card changes, or repeated buy-ins from losing. For recurring targets, I like to count the number of hands in an hour, and use that to inform game selection on future trips.

CJ spent the time to do an experiment tallying blackjack game speed under different conditions. The results appear in a chart on page 129 in the section “The Most Overlooked Way to Increase EV as a Card Counter.” Speed is so important that a spotter in a high-edge game might forgo a marginal split if the extra time (dealers can be very slow to re-arrange all the cards and bets on a cramped layout) would sacrifice another round.

So the underperformance is 10% skim and 40% speed, but what about the other 50%? I’ve got bad news for you. Your game needs work, kid. Oh, you’re in the Blackjack Hall of Fame already? Yeah, well, your game needs work, old man.

It’s possible that I’ve seen more APs on a table than anyone, because every time I play there’s another AP at the table! From observing my own teammates over the years, including numerous Hall of Famers, I know how common errors are. Errors are rampant. I’d estimate that a rookie makes some mistake every five minutes, and simple failure of the Raindrop Test would mean a mistake on every hand.

On page 15, CJ notes: “When we re-tested the entire team, more than half the players couldn’t pass the test they’d previously aced.” And that’s on top of the fact that in the wild, there are many ways to screw up that the at-home test won’t pick up. When there’s actual money on the line, a grumpy suit sweating blood, a toke-hustling dealer, and a vigilante “we-don’t-touch-soft-18-or-split-Tens” degen lynch mob, does the counter make the EV-maximizing move?

The 3-s.d. guys online would say they aren’t making mistakes, and sometimes even say that they had a friend check them out. That’s all nonsense, of course. There’s a big difference between a test that someone prepares for, and a pop-quiz. I do pop quizzes. I sneak up on my teammates and watch them from behind. I count down the card counters who sit at my table. I play while other players at the table are trying to HC. I’ve even been at a table playing my game while two card counters (who were wonging out of negatives), oblivious to who I am, were standing behind me discussing the book Beyond Counting (a very amusing conversation!).

I can guarantee that every AP out there is making mistakes they’re not even aware of. We could start with strategy. Does an AP really know the strategy for the game at hand? I recently developed some practice software for my crew, for the very games that we play every day. Without extensive practice on the software, none of us could get a perfect test of merely 33 hands. And I’m quite confident that anyone who doesn’t have access to such software would be a disaster.

For a HC player, we could talk about the weak information. I’ve ranted about Paint blindness for years, but when put to the test, everyone is horrible. A few teammates of mine did better than the average for attendees of the Blackjack Ball, but they’re nowhere close to computer-optimal, and they don’t even know their Paint charts. (I worked hard to make those charts!).

Even for a simple move like counting cards, there are all kinds of possible mistakes, and CJ could talk about it better than I can. I wish his book would go into detail on HOW the players failed the test. I’m sure that misremembering an index is a common mistake. Dropping the count is a real-world mistake no one admits to. Then there’s chickening out. It goes like this: There are two tables. The card counter plops down at the first one he sees, because it’s a new shoe ready to deal. That’s a mistake right there, because the table offers 65-70% pen, while the dealer two tables down offers 75-80% pen, and the sims assume a game-selection standard of 75%. Real-world EV has already taken a hit.

Then it turns out that the dealer is semi-sharp, or at least makes toke-hustling comments when a bunch of small cards come out. So now the counter is afraid to jump his bet from $5 to 2 x $150 (you simmed 1:2×30, right?). So he jumps his bet from $5 to 2x$65 (with the classic rookie badge—red on top of the green!), makes a futile comment about having to change it up (you won the last hand, bozo). Then the dealer makes a snarky comment, at which point the player tosses him a nickel. Now you have an extortionist on the payroll. Sure, you’re not making any mistakes at all, kid.

Thanks to the magic of the Internet, we can hear the rookie’s excuse for parking where he played, the excuse for trying to cash out $3400 (shocked that that would trigger any trouble), the reason for using a player’s card (the “free” buffet!). So, we are to believe that in EVERY aspect where we can audit the decision-making, we see mistakes, but that in every aspect that we are unable to audit (the actual counting, betting, and hand-playing at the table), the execution is flawless? That’s just untenable.

And sometimes we ARE able to audit those other areas. I’ve seen counters making their bets and playing their hands. They’re betting Lucky Ladies too soon. They’re playing too far into negatives. They’re too slow. (When an apathetic dealer is on auto-pilot, there’s no reason to hem-and-haw on an index play. That doesn’t make you look like a gambler; rather, it just wastes time and draws more attention to the deviation. Your default should be: swift, silent.) They’re over-acting. They’re over-tipping. They’re dropping the count after a big multi-way split and double. They’re physically turning their head to see the discard rack. They’re ignoring the phone call. They’re giving ID for no reason. They’re getting age-checked by going to the more dangerous checkpoint. They’re playing in front of the wrong boss. They’re not picking the best table. They’re not picking the best casino. They’re not fully utilizing free online resources. They’re not driving a car that can go up hills (we didn’t think to put that one on the list, but here we are: 2020 was an eye opener! That one’s for you, John Smith!).

If you don’t believe me, start auditing. You can tally results to check skimming, count hands to check game speed, and monitor game execution to check skills. I’d enjoy fine-tuning the 10%/40%/50% breakdown with someone with additional data, like CJ, but I think we’re on the same page. We have the explanations for underperformance. All that talk about God working in mysterious ways? Fake news.