My first Android review :-) - Backgammon by AI Factory

 Okay, it's time to also review Android Backgammons. I wiped the dust off my old trusty Galaxy Tab S2, and installed my first Android Backgammon in a long time: Backgammon by AI Factory. 

 

Backgammon by AI Factory in action

There's this urban myth about Albert Einstein. It says that he has received poor grades in school, particularly in physics. Actually he has received a grade "6" in physics, but while a "6" is a poor grade in most European countries (an "F"), it is a very good grade in Switzerland, where Einstein went to school.

I had to think of this myth when playing a match against Backgammon by AI Factory (henceforth BGAIF) at its supposedly strongest play level "7". Is this all a big misunderstanding? Is the strongest level maybe "1"? But no, even the text explaining the levels goes into great depth about the two new levels 6 and 7 which use a different, very strong AI.

But look at this, for example (NB this is a screenshot from XG2 that I used for expert analysis):

 

Blue to play 22

 

In this position, a blue checker is trapped behind my 5 prime. It's not so important what blue does with three moves of checkers in his home board. But blue MUST move one 2 as 23-21, hoping to roll a 6 later and get behind my prime. Unless you're a total newbie you immediately sees this necessity.

It turns out that BGAIF didn't. It moved 6-2(2).

 

Blue to play 22, with analysis fro XG2

Not unexpectedly, XG2 considers this a horrible blunder, losing nearly a full point of equity (-0.894, to be precise :-) ). And if you look at the move list on the left side, you see quite a lot of BGAIF's moves marked in red color, because they are blunders.

At the end of my match, XG2 awarded BGAIF the poorest of all ratings: distracted (PR 41,8). This means that you cannot hope to improve your backgammon skills by playing against BGAIF and learning from its moves. You can also not hope to enjoy playing against it if you're not a beginner. It's not playing like the poorest AIs that I've seen, but it's nowhere near the same league as Backgammon NJ, for example. 

I'm a bit mystified how an app can offer seven (!) levels of difficulty, using two different AIs, and the top level plays like a beginner.

For completeness sake: yes, I won the match, even though I played poorly (Intermediate) and luck was on BGAIF's side. 

That said, let's cover the basics of this app: BGAIF offers quite a lot, everything that a newbie to Backgammon might want for playing casual games. It doesn't offer enough for anybody to improve her backgammon skills.

You can play money games or matches, with or without cube. It even supports the Crawford rule of matches. It offers several different board designs, some of which you only get if you have another app from AI Factory installed on your device. While playing it supports you finding the right move by highlighting possible checkers and target points and showing the current pip count. It also features an - unconventional - undo: on the one hand you can completely undo any number of moves, which is cool. On the other hand it doesn't feature the "normal" undo that everybody who plays backgammon seriously needs. Normally, when deciding what to do with a difficult roll, you play around with different moves, do them on the board, examine the position, maybe do some probability math, take the move back, do another one, until you're happy. In BGAIF you move the last checker and your move is done. This is a bit annoying, as the next time to undo is after the computer has rolled and moved.

BGAIF doesn't offer any kind of tutor, but that's okay because the engine's advice wouldn't be too much to learn from.

 

The default wooden board looks fairly good.

 

Ads in BGAIF deserve a special mention: BGAIF is completely free, so it's development is supported by in-app ads. No in-app purchases. The ads are VERY limited to a banner in the top right corner. Kudos to AI Factory for not destroying the game with too obtrusive ads.

Summary: I must say I was disappointed by the AI of this app. 7 levels, and the best one plays like a beginner? That's not good. Apart from that, BGAIF is a fairly decent app with not too many highlights but no critical flaws either. Good for a fun game in-between if you're not an experienced Backgammoner (Backgammineer?).

ps. please note that I don't own an android phone anymore, so I can only review how the app looks on a tablet. A good review should check how the app looks on different devices, sorry, can't do that.

A hidden treasure: Backgammon KG

Apple‘s App Store is like a big ocean. If you‘re a fish who manages to live at its beautiful coral reefs, or at its wonderful famous beaches, plenty of divers will get to see you. But if you live down there in the black depths far below the surface, chances are you‘ll never get into a diver‘s sight. 

Backgammon KG is such a creature from the deepest seas. Unlike companies who have read their SEO (search engine optimisation) books and know how to get their app into your search results, its developer Keith Gould probably just wanted to create a good, free backgammon and nothing else. As a result it is near impossible to get it into your app store‘s view. I‘m regularly hunting for new backgammons, scrolling through long lists of apps, but this one never appeared. Only as a related game to another app it showed up. I downloaded it to my trusty iPhone and was amazed at what a beautiful deep sea fish I had found.

Backgammon KG on an iPad Pro

The thing is: there are many (MANY) backgammon apps on the App Store. But there isn‘t a single one that you can call full-featured without blushing and that is really free. Only this one. Backgammon KG. On the German App Store it has a single (one star “this app is cheating”) rating which suggests that only few people ever tried it.

Graphically it’ fine, but there‘s not too much to write home about; the UI gets the job done, the app has a clean and consistent look, you‘ve got 10 board designs to choose from. But I‘ve seen more beautifully looking backgammon apps. One complaint: sometimes, mostly in the lower row of the board, you have to click slightly left to where your checker should move or it will move one point too far. And a minor complaint: there is no native support of the iPad - it’s using app scaling which makes the app look low-res.

Some of the designs (here: metal) are, well, daring

It’s real strength lies not in its presentation but in its scope and in its price. Backgammon KG is really, actually, totally free. Not free as in “buy 1000 coins in-app to unlock this and that”. Not free as in “whenever you like it least it will confront you with 20 seconds of video to watch or with a blinking ad image. No. Really, actually free. Download it. Play backgammon.

And backgammon it plays. Not the “what doubling cube?” kind of backgammon that half of the apps support. Not the “there is something else than money games?” kind of backgammon. Not the “my developer got the task to build my AI in 10 working days so I play like the ultimate backgammon noob” kind of backgammon. Real, actual, strong backgammon! You can play matches from 1 to 11 points, with Cube, Crawford and everything else working the way it should.

Backgammon KG’s AI uses the same fundamental approach that all the top apps rely on - a neural net based machine learning approach that was trained by millions of self-play games. It shows that the app could have used a few million more of those self-play games. It’s strong, really strong, but not as strong as XG mobile or the BGBlitz inside True BG. I’ve played 4 test matches to 3 points, one myself, one pitting Backgammon KG against the PC program XG2 on an extremely strong level (“roller++”). My conclusion: KG’s checker play is world class level. Not perfect - sometimes an error or even a small blunder creeps in. But good enough for quite everybody to get a good challenge. Cube handling is not as strong; Backgammon KG plays a very defensive cube, missing good cubes quite often. So in the end XG2’s assessment of KG in the four matches was once advanced and expert each, and twice world class. Not bad for a free app.

The little red lamp shows the tutor’s opinion on my move (error!)

Also, the game includes a (nearly) full-featured tutor - great to improve your skills. If you activate the tutor in the settings, you’ll get two things: a little lamp that is white (“perfect”), green (“small error”), yellow (“bigger error”) or red (“blunder”). make a move, see the lamp, and take back/experiment to see what the game recommends. Next to the lamp you’ll find a “Moves” button. Hit it and see all possible moves, ranked from strong to weak, along with two mysterious figures that I haven’t understood.

The tutor’s move list

So to find out the best move, click on the first row and behind the semi-transparent list the move is made. Close the list to see the full thing. Undo if you want to still consider different moves.

I can’t stress enough how cool the combination of a strong AI and a tutor is - apart from reading Backgammon books or watching Backgammon videos of pros commenting pro matches this is the best way to hone your backgammon skills. And while I believe the top apps deserve your investment of around 10 bucks, this one does the job really well, completely free of charge.

One minor gripe: the app doesn’t tutor your cube decisions. This is sad because, at least for me, cube handling is the toughest part of backgammon. To make a proper cube decision you have to assess your winning chances, your / his chances to gammon, the volatility of the board, the likelihood of losing your market next turn. After decades of Backgammoning I still regularly make cube errors on the level of “this has cost you half a point”. But then, Backgammon KG isn’t much better in this respect, so a cube tutor wouldn’t teach you that much.

The default board design 

 

A few more words on features: Backgammon KG includes four playing levels, offers statistics on your playing performance, and tries to keep you motivated by gradually unlocking more board designs whenever you win a game on expert level. For the final “Andromeda” design you have to win 200 games. Not an easy job (though you can cheat by stopping and re-starting the app again and again with your last winning move on the board, but that wouldn’t be appropriate, would it?)

My conclusion: this is one of my top three recommended backgammon apps. True Backgammon costs you good money and offers the most beautiful experience. XG mobile isn’t as beautiful but IMHO a tad bit stronger, and an excellent choice for good money. And Backgammon KG is nearly as good without costing you a cent. Now go ahead and download it, what are you waiting for? :-)

 

Update (Sep 2022): as part of my project to have XG2 rank every app in a big table for easy comparison I've played a few matches against Backgammon KG and analyzed them in XG2. Result: it indeed plays quite strong, at a strong expert level, almost world class - PR 5.7. You hardly find a human opponent that is that strong, and gives you free backgammon teaching.

 

A piece of art, but barely a backgammon: Backgammon Machine

The computer game “Doom” kicked off the whole 3D first person shooter genre. In this classic, you start by selecting a difficulty level. On “I’m too young to die” the game is easy, on “Hurt me plenty” it’s fairly hard to beat, but real men (and women) played it on “Ultra Violence” for a real challenge. This analogy came to my mind when thinking about why a Backgammon like “Backgammon Machine” that doesn’t include the doubling cube is fun to play, but somehow stale compared to the real thing. Checker play offers plenty of fun, but by including the cube you turn the game’s challenges from “hurt me plenty” to “ultra violence”. 

That said, Backgammon Machine is a special app. It’s really, actually, totally completely free. No ads. No in-app purchases, no virtual coins, no “watch this video for <feature>”. Free. And beautiful. Man this app is well designed in its pure and beautiful grey look. Steve Jobs would love it. I love it.

Playing a game of backgammon on the iPad

But its beauty doesn’t stop at only visual appearance. Its user experience is also top notch - the way you move your checkers, how you undo is just perfect. For example: you roll 6/3, move a checker by 6 pips. Now if you tap on this checker, the app will highlight two target places - 6 back (to undo the 6) and 3 back (to turn the 6 into the 3). Very clever, have not seen this anywhere else.

The minimalistic options screen

Concerning features, this is pretty much it. You start a game in “match to 1 pt” mode. You make your moves (enjoyable). It makes its moves. You start another game in “match to 1 pt” mode. And so on. No tutor/analysis, no setup, no matches, no doubling, no exporting, no manual dice, you get the picture :-).

Oh yes, and the UI is so clean that it doesn’t show a pip count. The joys of counting pips by hand (strengthening my “criss-cross pip counting skills”).

But maybe that’s enough. Play against a strong, free AI can be fun, at least for a casual backgammon in between meetings or such, even without matches, without doubling. So all boils down to the question: is the AI any good? The answer is yes and no. It certainly plays a stronger game than the AIs that have been hand-crafted in a rush. It certainly plays a weaker game than then high-end AIs of XG, BGBlitz, BG NJ. 

I‘ve played 18 games so far, and lead 12:6 against the AI. I feel I was fairly lucky sometimes, but I‘ve also seen it going for significant blunders. For example, in a situation where I‘m bearing off against 1 or 2 trapped checkers it moves those trapped checkers home far too early (considering it doesn‘t treat a gammon as 2 points it should wait it out as long as possible). Or take this one:

Backgammon Machine playing a 6/4 opening rather … uhm … unconventionally

This 6/4 opening, according to XG2, is nearly a blunder (losing 0.072 equity). This is way too loose, it will be hard for me to roll something that doesn’t hit, and hit on a point I want to make. Not something an app that claims to play at nearly master strength would do.

Nevertheless this app plays a good game of Backgammon, certainly better than many of the “beginner” level Backgammons out there, and as it’s fast, fun, free and beautiful, it’s great for a fast, casual game of backgammon.

So, just install it and have fun. Just don’t expect too much. I hope the author invests more into self-playing to improve its strength, and into a few more features. Matches would be great.

 

Update (September 2022): As part of my "big table with all backgammon app rating" project I've played a couple of games against the machine, and had them analyze with XG2 on the PC. Seems I was lucky during my review; the app plays a strong expert game, almost world class level. (PR 5.2). Sadly it still has no cube.

 

Strange but fun - Backgammon - offline by SNG CT

This one took me a long time to review. Even after playing 35 games against its four levels of AI, I‘m still unsure what to think about it. One thing is for sure, it‘s not for a serious match of backgammon, one in which you can learn from a superior AI and improve your skills. And it wants to make money by making you watch ads and buy coins. And it doesn‘t shy away from fairly unconventional ideas to achieve this.

But let‘s start with the basics. First off: this is about money games. No matches.

The game pits ou against four difficulty levels. Each level raises the stakes - at the beginning you have to compete with the „beginner“ AI because you can‘t afford the price of admission to play against advanced or expert. After you‘ve selected your level you‘re treated with a screen in which you can enter your bet and your maximum bet. From this it calculates how often you can double. 

It took me a long time to figure this one out. First fun fact: this app called Backgammon doesn‘t know what a Backgammon is. It does know what a Gammon is, though. So it calculates in my example: maximum bet is 50,000 coins, my selected bet is 14,400 coins. If I would double and lose gammon I would lose 57,600 coins which exceeds my maximum bet. Therefore I may not use the doubling cube.

Hint: you want to select a bet that allows you to double because the AI has no clue about how to double well, doubles like a „now I‘ll win for sure“ beginner, accepts doubles it should really, REALLY drop. Easy money.

After these two easy steps, you‘re sent to the actual game.

Two things here: a) for some reason much screen space is invested in showing you and one of many virtual opponents. And b) this layout makes the board tinier than possible,, and there‘s no way of showing it full-screen. No problem on a big iPad screen, but still…On a 16x9 iPhone screen this is no issue.

IPhone.

Apart from this, the user experience playing this backgammon is awesome. Nice undo, unobtrusive  support figuring out which checker can move and where it can move, nice drag-and-drop playing if you wish (I love drag-and-drop in Backgammon because it slows me down a little bit and lets me think if this is really what I want. With click-and-click you move fast. And err fast.. And d&d feels like a real board). The app offers a multitude of beautifully rendered boards.

Part of this game (that I was using only in one game to understand it) is: you can watch an ad to re-roll a poor roll. Yes, seriously. There seems to be a limit to this (3 times per game?), but it tilts the odds massively to your side - getting rid of the two or so biggest anti-jokers will massively make you a favourite.

Here it is: watch and reroll.

If you‘re a reasonably good backgammon player, you won‘t need this feature, though - the AI is nowhere as unbeatable as XG, NG, BGBlitz are. But for mysterious reasons I love playing against those little virtual bots. They cover an interesting “white space”: backgammon apps tend to fall in one of two camps. There’s the bleeding-edge neural network bots that play vastly superior to you, and there’s AIs that some guy had to handcraft by creating a bunch of rules, rules that misunderstand 50% of board positions and play like a beginner. You know, the ones that double against your 6-prime because you’re 10 pips down.

This app claims to use a neural network AI that has been trained a million matches. Maybe that’s not enough, but the AI happens to play well, without the horrible blunders that handcrafted AIs annoy you with, but a bit worse than me (XG calls me “advanced” most of the time, “intermediate” if I don’t focus enough). And strangely, the intermediate AI beats me more often than the expert AI. So, at my level of play, I get opponents that I beat more often than not, but that offer good resistance and challenging games, which probably is what many players want. Your mileage may vary. Oh yes, and I’ve mentioned that it doubles poorly, so the joy (pain?) of figuring out whether to take or drop a difficult double is not something you get in this app.

What else can be said? You can buy virtual chips for real money, and if you’re a beginner you will probably have to do this to keep playing (there’s a daily 1,000 coins bonus, as in many F2P games). You can pay 5 bucks to remove ads or watch a boring ad video after every single game. While playing you’re building up experience, even have an experience level, but I haven’t figured out what this level is supposed to do. The app offers no tutor features of any kind.

All in all, if you like money games, want a good but not too good AI to play casual games against, “Backgammon - offline” might be an app to consider. If you’re a backgammon newbie you might face “out of money” situations, though.

Bad luck in Backgammon

Recently I thought about luck in backgammon, particularly bad luck. Don't you hate it when the dice roll exactly the way you don't want them to? Don't you love it if you need a 51 to turn the game around and you get... a 51! Wouldn't you want that bad luck gone in the game?

I ended up with two experiments with XG2:

Experiment 1: The perfectly lucky game!

Suppose you take out all luck from Backgammon by always rolling the best possible dice. A joker every move.  Wouldn't that be cool? How would the game look like?

It turns out that the perfectly lucky game (for both sides) is incredibly boring.

(For the details on how to get XG2 to play the perfectly lucky game, look below)

Here's the transcript. Brace yourself for some serious boredom.

1) 31: 8/5 6/5

Luckier than the other good candidates like 61, 42, 65.

1) ... 66: 24/18(2) 13/7 (2)

This leads to this position:

 In this position, 66 isn't the best roll you can get - the 7 point and the 18 point are made by white so it would actually be very awkward to roll a 66 now. The best roll here is 44.

2. 44: 24/20(2) 13/9(2)

Now, again, 66 isn't the best roll for white because you can't bring the back men on 18 home with it. The joker now is 55.

2. ... 55: 18/8(2)

And here we're at the end of our game already. Blue's joker is, obviously, 66, bringing home the two checkers on 20 and turning the game into a pure racing game, where always 66 is the joker, and where white wins because blue rolled only a 31 and a 44 where white rolled a 55 (in addition to 66 rolls).

For completeness:

3. 66: 20/8(2) 

3. ... 66: 13/7(2) 13/1

And the two groups are separated. In the further "battle of the 66", white wins like this:

Now that wasn't too dramatic, was it? I was expecting great swings, dramatic luck, but all I got was boredom (NB I got the great swings - the game's analysis cycled between "white wins" and "equal" for most of the time, but the game was just boring). Also notice that there wasn't any close decision here, all was straightforward. Seems the joker typically leads to a very straightforward move.

Conclusion: too much luck is boring :-)

Experiment 2: my "Luckgammon" variation 

Considering what to do to reduce bad luck in Backgammon I thought of this variation:

Instead of rolling and moving, you roll three times, and pick the roll that is most lucky. Then you move.

(For the setup how to achieve this with XG2 look below)

The idea behind this is that it keeps some randomness in the game, but you never roll those awful anti-jokers and you hit a good or very good roll nearly always.

Now these games have to be taken with a good grain of salt - XG2's perfect play is tailored for the normal backgammon; if you can roll 3 times, the probability to hit a checker, to enter a 4-point board etc are vastly higher than what XG believes, so the best move is probably different to what you believe. Even the luckiest move might be different than in normal BG, so my method of selecting the luckiest move is slightly flawed as well, or isn't it? Tricky.

Anyway, I played four games like this. Believe me, it is TEDIOUS to play these games. See separate post for the details. Two games I played "me vs XG2", two I played "XG2 vs itself". All four games were, again, fairly boring. Not as boring as the joker game, but still nothing compared to the drama you get in plain, normal, ordinary Backgammon. Actually I had difficulties picking good positions to show you, positions where a tricky decision has to be made, a position where an ingenious move has to be found, where you have to see the right game plan. There's not much there.

Bad luck is the icing on the Backgammon cake, it seems

Now four games isn't representative for sure, but still, after playing these games it seems to me that my Luckgammon isn't so much fun. It seems that without the occasional really poor move, without figuring out what's the best you can do with an awkward roll, without the frustration of dancing four times in a row, the game gets shallow and stale.

So the next time you bite into your keyboard in frustration over that poor roll or in anger over your opponent, of course, hitting you, although there were only 4 of 36 rolls that would allow him that, maybe it helps you to know that the tough luck makes this game the fun that it is :-)

Appendix: How to get XG2 to always roll the perfect roll

1. Start a new game transcription
2. For every move, do the following:
1. click on the last move that was made. You should see XG's analysis with the best moves and their equity. 
2. Right-click on the best move, select "Dice Distribution".
3. On the tab "dice map", pick the reddest roll - this is the roll that is your anti-joker for the next move, i.e. the joker for your opponent for the next move.
4. Go back to the current move, manually roll the roll you've seen as joker in step (3) 
5. Hit CTRL-H to get a hint, if necessary, use XG Roller+/++ to make sure you pick the right move. Do the best move in XG's list.
   

For the first move I had to try out the various good rolls because XG won't show you any info before the first roll, so I selected all good rolls (65, 42, 61, 31) and looked at XG's luck rating after the roll, and picked the one with the highest luck after the roll - 31. 

Appendix: How to get XG2 to play Luckgammon

First, start a new game transcription. Make sure you can easily distinguish the two players (e.g. by calling them the colors of their checkers). 

 

Then for every move do the following:

1. Hit "R" to roll a random roll. Have the "Overview" panel open and note the roll and the luck value of the current player. Do this for 3 times. Pick the roll with the highest luck value.
   (special case: if you've got a checker on the bar, and your roll dances, you have to hit "T" to take back the roll, twice, to get back to your current roll)
   
2. Hit "T" to take back the current roll, enter the roll you picked as manual roll.
3. Do your move, maybe using CTRL-H to get a hint of XG2 on what to move.

 

Even the Gods blunder sometimes

Played a match against the wonderful True Backgammon on my iPad today. Very surprising: BGBlitz on 2-ply believes I played at PR 6.0, and that the BGBlitz in True BG played at PR "-0.5" (better than perfect :-) ).

But if you feed the match into XG2, or if you go for a BGBlitz 4-ply analysis, things change dramatically. My rating moves to PR 7.8, BGBlitz's rating to PR 6.5.

It turns out that BGBlitz doesn't seem to deal well with avoiding backgammon losses. Both the BGBlitz engine inside of True BG nor BGBlitz for the PC blunder seriously in this position.

Blue to play 32. 

Not too hard, right - blue must run away from being backgammoned and should play something like 21/18 20/18. Actually, 24/22 21/18 is best, avoiding trouble when rolling a 11 next.

But True BG played 24/22 7/4, and BGBlitz on up to 3-play finds this move just great. Only on 4-ply it agrees with XG2 that this is a serious blunder (BGBlitz: -0.19, XG2 rollout -0.178)

And so it happened: I won this game with a Backgammon :-)

Three reasons I love DailyGammon

Summary (TL;DR): if you look for "correspondance backgammon" and you don't mind having nothing else (no PR matches, no fancy UI), then DailyGammon is what you need.

 

One of the curses of the Internet these days are clickbaiting articles like "3 reasons you need to do this" or "5 best somethings". But, well, there are three reasons why I love DailyGammon since I finally found it (took me a long time), so I'll list them here and risk that you think "OMG clickbait!"

Before I start, let's get a few things out of the way: UX-wise, DailyGammon can't hold a candle to state of the art sites like Backgammon Galaxy. This is what it looks like:

And there's no drag-and-drop moving, no animation in seeing what your opponent moved (instead you get beautiful pink arrows) etc.

Also, other than Backgammon Galaxy or Backgammon Studio Heroes it doesn't really advance what online Backgammon can be (e.g. have "Extreme Gammon" watch over your game, rate your strength, so you play not only for winning matches but also for the lower PR).

But that's fine. DailyGammon is excellent at one thing: offline (asynchronous? correspondence?) backgammon. And this is what I love, for three reasons.

1. I can finally play 7pt, 9pt and even 21pt matches

I love backgammon and I'm trying hard to improve my skills. In a position I try to find the best move, and sometimes this takes me a minute or so. And this is a great minute, weighing the alternatives, trying to keep pip count in mind, think whether offensive or defensive play is the right way forward, imagine what might happen after I moved, look for duplications etc.

The problem is that this way a 9pt match will take an hour or longer, which is not the amount of time I usually have in one sitting. My day allows for five minutes of backgammon now and 15 minutes then. 

Unfortunately unlike in Chess, correspondance Backgammon is not a thing. Most sites don't support it. But DailyGammon does :-)

2. I can play relaxed matches

Maybe that's just me, but even "casual" games on Backgammon Galaxy cause my adrenaline to flow madly inside my body, causing my mind to miss great moves. Maybe I'm just not a "strategical thinking in high time pressure" guy, and maybe that's just me (on lichess my correspondence rating is 600 ELO above my blitz rating ...)

3. I can actually finish 7pt, 9pt and even 21pt matches

This is where DailyGammon really shines. The guys behind the site have clearly thought deeply about how to do correspondence backgammon and came up with an ingenious solution to an obvious problem. Other than chess or go, Backgammon has plenty of situations with obvious or forced moves. Opening DailyGammon, going into a match, pressing "roll dice" and "submit forced move", and coming back an hour later or so is just not working out. 

For this problem, DailyGammon uses a computer guessing mechanism - after your move it rolls the opponent's dice and makes the most reasonable opponent move. Then it will ask you "if your opponent would roll this and move that, how would you continue?". This way you play a main variation for 3 or so moves before submitting your move. 

When your opponent logs in later he can instantly play this sequence until he decides for a different move than what the computer thought (or until he doubles). If this happens, the rest of your pre-moves are rolled back, and the game continues (with different dice rolls then). 

This speeds up the game dramatically. You can finish a single game in a few days.

The flip side of the coin is that things can get a bit confusing, and that chatting is tricky. Suppose you're playing five games at the same time. In game #3 you make a move, and then do 3 more moves conditionally based on what the computer thinks your opponent will do. Then you move to game 2, then to game 4. Then your opponent of game #3 comes and makes a different move for move 2, gets different dice afterwards. You come back to game 3 at a state that you just don't recall. Also, chatting "well that 66 was fairly lucky" is tricky because this might be move 3 of such a conditional move sequence, and if your opponent moves differently one move before he might never roll that 66 and not get what you're talking about.

Actually in my first game I clicked one move wrongly and submitted the move which was the third one of such a sequence. As the move I clicked was really stupid I chatted "oh well mouse click, meant to play that move". My opponent played differently, my wrong move never happened, my opponent was confused about what I'm writing...

4. (bonus reason): The people there seem to be more polite than on other sites

With Backgammon being Backgammon, the game where normal people rage-review apps because they can't bear being destroyed by superior engines (THEY CHEAT FOR SURE!), it's easier to come across inpolite language in Backgammon chats. So far I have the impression that on DailyGammon the crowed is more polite than on most sites.

The snail and the cheetah - non-random 21 vs 65

Let's pit the slowest of all rolls - 21 against the fastest non-double - 65 and see what happens.The game starts fairly normal. This time I'll be using a nice BGBlitz theme to show the positions.

1. White 21: 24/23, 13/11

2. Black 65: 24/13

 

3. White 21: 13/11 6/5

You see it coming already?

 

4. Black 65: 24/13

 

5. White 21: 11/9 6/5

 

Still no moves to write home about, but white now has made the 6 point and the 5 point, which can be annyoing if you (as black) only roll 65...

 

6. Black 65: 13/2*

 

Not many alternatives here, but now a 21 will cause serious trouble for black.

 

7. White 21: Bar/23 23/22

 

 And with this position, a 36 move dance nightmare begins for the black cheetah, sitting helplessly on the bar, while the snail moves forward...

and forward...

and forward.

(I hope you understand that I'm sparing you the boring details)

 

Until in the end...

black can come in with a 6! Yay!

In the remaining game, even the big 65 rolls are not enough to catch up. In this position, black resigns.

Now that was unexpected. I was expecting a tough fight, where white falls behind in the race, needs to keep contact, etc.

But of course, with 6 being made from the beginning and 5 being the most important point, always rolling 65 can be painful.

The epic, ultimate non-random series
Part 1: 51 vs 42

Do you also remember the awful feeling you get if the damn Gods of Backgammon luck decide to give you the same roll twice, thrice, four times? How awkward your positions get if you roll 51, and then you roll 51, just to roll ... 51? 

What would happen if we would confront the top AIs with not only this situation, but with an eternal non-random series of always the same roll? And who would emerge victorious if one side always rolls, say, 51, and the other 42?

 

Let's find out.

 

The first game is 51 vs 52. Both rolls bring you 6 pips ahead, so the game might be balanced. I have more warm and fuzzy feelings about rolling 42 than 51, so I would bet my money on "42 wins".

 

Here we go.

 

Setup

As I can't create DLLs for the life of me (which is what XG2 wants as custom dice generator), and as I was too lazy to manually roll 51, 42, 51, 42, 51, ..., I let BGBlitz play against itself, on strongest TachiAI V settings. BGBlitz can take text files that contain the rolls after each other. So I create a huge text file with endless iterations of 51 42 51 42, let BGBlitz play a money game against itself, and then analyze it with XG2 on Roller++ level (because XG has this nice PDF position export).

 

I'm only an advanced player (although my checker play is expert most of the time), but I'm trying my best to help interpret XG's analysis.

(btw both players are BGBlitz TachiAI V, that "andara" name is just a bad German pun)

The game. 51 vs 42

1. White 51: 24/23 13/8 

 

2. Black 42: 8/4 6/4. So far nothing unexpected

 

For the next 51 white faces the first interesting situation.

 

 My favorite move here is 24/18, doing something reasonable with the roll; I just can't imagine stacking another man on the 8 point, and as the 17 point is stripped, hitting me with an ace wouldn't be all that bad for me. It turns out that XG's top two moves both contain 13/8. BGBlitz played 23/22 13/8, and XG2 agrees to that. My 24/18 was rated as a -0.045 error. Not too horrible.

 

 3. White 51: 23/22 13/8 

 

Another interesting one. Black to move 42. Split the back men or not?

 It turns out that 24/20 13/11 and 13/11 13/9 get exactly the same value. BGBlitz didn't split and played

4. Black 42: 13/11 13/9

 (Curious, I did a rollout of this position with XG2 which was not conclusive but preferred splitting the back men by 0.008.)

Now 51 is an anti-joker with -0.42 luck. 

5. White 51: 24/23 13/8

6. Black 42: 11/7 9/7

That's the nice thing about always rolling 42. You split men from the 12 point in one move and can make a point 6 away with the split men in the next move :-)

By now, white is seriously behind, and rolls another awful 51.

7. White 51: 23/22 8/3

Not much of a dispute here - sometimes there's only poor moves, and this is the least poor one. Rather place a builder on the 3 point than giving up the 13 point and improving nothing. If you're interested: XG2 ranks 23/22 13/8 worse by nearly 0.2 points.

8. Black 42: 24/22* 22/18

Just as good (-0.012) is 24/22* 24/20

9. White 51: Bar/20 8/7*

10. Black rolls another 42. Quite as expected.

Here the question is whether to move the back men only or whether it makes sense to get one man off the 6 with 6/4, distributing the attackers a bit better.

10. Black 42: Bar/21 24/22

Bar/21 6/4 is an error (-0.035), probably because it lowers the chances to make the 5 point, and with a white anchor on 3 the builder on 4 doesn't make too much sense.

11. White 51: 20/15 8/7

12. Black 42: 22/16

Trying to bring home a back man as long as there's no prime.

13. White 51: 15/9*

Now what should Black do with its 42?

Bar/21 is clear, but where should the deuce go? 13/11 or 6/4?

14. Black 42: Bar/21 13/11

The reasoning is like before: A blot on 11 is fairly save and can act as a builder for the 5 point. A 3rd man on 4 is not very useful. No blunder but a -0.064 error.

15. White 51: 9/3

In the next position it took me a while not to blunder seriously.

 

 

I must confess that I initially was very fond of 11/5, keeping my stones on 13 as a safe spot for my back men. But safely making the 5 point in the next roll requires an ace. Not too likely (11/36). And 13/9 11/9, making a 4 prime, is just more safe and better, even if the two man on 21 will struggle.

16. Black 42: 13/9 11/9 

(11/5 would have been a -0.25 blunder)

17. White 51: 13/12* 8/3

18. Black 42: Bar-21 6/4

No other option for the deuce if black wants to keep its prime.

Now white has many similarly good moves for its 51. Can you find the best one?

Obviously, white wants to unstack its 6 point and do something about the two blots who can be hit with 8 numbers out of 36. But should it slot the 5 point?

I didn't find the right move, went for 13/8 6/5 (-0.008, okay, I'm fine with that). The best move, is, unexpectely for me, the totally conservative

19. White 51: 13/12 6/1

It achieves the two goals, but at what price? No flexibility anywhere, and a man on the ace point?

20. Black 42: 6/2 4/2

White again faces a difficult decision.

So, which one is it?12/11 6/1? 6/5 6/1? or maybe 12/7 6/5 or even the save but passive 6/1 3/2?

 

BGBlitz played 6/5 6/1 and XG says everything else is a 0.050+ error. 

I must confess that I'm still not entirely sure what's going on here. The dice distribution shows that 6/5 6/1 isn't that bad, because many rolls where black hits on the 5 cause it to give up the prime, for example a 21. Therefore XG believes only 6 black rolls are really bad for white (14, 15, 16) because it allows the checker to hit and escape. Others like 12, 13 are not a big deal.

On the other hand, if white would play 12/11 6/1, then every roll that hits on 11 or 12 is horrible for white. And there are 8 of them. And while 6/5 3/2 doesn't leave a blot, it gives black much freedom to improve its position because of the two blots in black's home.

21. White 51: 6/5 6/1

22. Black 42: 8/6 8/4

9/7 9/5 would keep a three-point prime but 21 and 22 are just too horrible. Blunder (-0.12)

And 51 proves again to be an awful roll. What to do now?

You can't cover the blot, and you can't move it out of danger. What's the best option?

I would actually have played 12/17 6/5; this gives me two builders for my blot on 6 in the next roll. Would be a small error (-0.038). The best play is

23. White 51: 12/6

 

Maybe because there's no double hit like with 12/11 12/7 (second best move, -0.018) 

 

24. Black 42: 9/7 9/5

 

The alternative 6/2 4/2 is very ugly but so safe that it's only very slightly worse (-0.007). 

25. White 51: 22/17 6/5

With black's blot on 20, now is a good time to try to escape. The alternative of 12/7 6/5 misses an opportunity and is a -0.15 blunder.

Now, all of a sudden, white is a favorite to win the game.But we all know the nice roll it's going to get

26. Black 42: 7/3* 5/3

 

 

Position after Black's move 

 And black is back.

27. White 51: Bar/20 12/11

The rule of thumb "the further away the blot in an indirect hit, the better" doesn't apply if a prime is in-between. Here black can hit white on 12 with 53, 62, and on 11 with 52, 61. And on 11 it has more safe landing points in the next move (at least that is what I - fairly clueless - believe the reason for this move to be).

 

28. Black 42: 7/3 7/5

 

29. White doubles. 

 

If I can trust XG2's analysis and rollouts, then sometimes BGBlitz doubles poorly. Seems to be the case here. XG2 says "no double/take", doubling being a 0.077 error (XG rollout). And this is a daring double, against a 4 point board.

29. Black takes.

29. White 51: Bar/20* 17/16

Obviously, Bar/20* 11/10 is worse, 4/10 is a direct hit.

 

 30. Black 42: Bar/21 3/1

 

Clearly the best move. Bar/23 doesn't have a single roll past white's prime; 3/1 starts a fifth point in black's home.

 

31. White 51: 20/15 16/15

 

32. Black 42: 6/2 4/2

 

33. White 51: 15/10 11/10

 

34. Black 42: 6/4 6/2

 

Black's board crumbles.

 

Now, white faces a number of interesting candidates for its next 51. Can you find the best one? I couldn't.

 

 

35. White 51: 7/6 7/2

 

My explanation (do you have a better one?): 15/9 leaves a direct hit which isn't reasonable given white's fairly big racing lead (-0.061). My choice, 8/7 8/3 doesn't slot an additional point and doesn't have any significant advantage over the best move (-0.053). I just happen to hate these point/empty/point/empty/point structures.

 

36. Black 42: 4/2

 

According to XG, this is slightly worse than 3/1.

 

37. White 51: 8/2

 

 Makes the 5-point board and keeps the 15 point to keep an extra hit chance in case black wants to escape. Given black's weak board the blot on 8 is no big deal, and keeps gammon chances.

 

I must confess that I would have played 15/8 6/5 without thinking here (a -0.1 blunder!); I didn't see that this is now about a gammon vs. no gammon. With the stopper on 15, the gammon chances increase significantly (8/2: 88% win rate, 24% gammon, compared to 15/8 6/5: 89% win rate, 12% gammon)

 

38. Black 42: 21/17 3/1

 

39. White 51: Bar/20 6/5

40. Black 42: 21/17 3/1

41. White 51: 20/19 15/10

Now black (42) can either crunch his board even more (17/13 4/2) or risk two different direct shots against a five-point board (17/11).

The crunched board is bad, but not that bad.

42. Black 42: 17/13  4/2

And another unlucky 51 roll for white.

 

The 5 is clear: 10/5. But what to do with the 1? Move the back checker on the 19 into a direct hit at 18? Destroy the 5-point board with 3/2 or 2/1? Or something else?

I didn't find the best moe. I would, actually, never have selected the best move. Also, BGBlitz didn't find the best move. And I had to roll it out to believe it. 

 

The best move would have been: 10/9 10/5, leaving two blots, either to a double shot from black, on 10 and 9, and another indirect hit on 19. Wow. The stuff you can do if you have a strong board and your opponent's board is crunched. But BGBlitz played what I would have played (please note: after rolling out the position the difference between BGBlitz's move and the best move is a measly 0.007).

 

43. White 51: 19/18 10/5

44. Black 42: 17/13 4/2

45. White 51: 10/9 10/5

18/13 3/2 would have been safer, but reducing the gammon chances significantly.

46. Black 42: 13/7*

47. White 51: Bar/20 9/8

48. Black 42: 13/7

Slightly better than 7/5* 5/1.

49. White 51: 20/19 8/3

XG recommends 20/15 8/7, but the differences are minimal (0.001 after rollout)

50. Black 42: 7/5 7/3

I'll end this game here; white kept its lead but couldn't prevent black from running home its back checkers. White won a single game.

Result: 51 beats 42 (51 starts) :-)

Alternatives...

In move 26, there was an alternative to black's 9/7 9/5: Play the ugly but save 6/4 6/2. Actually this is what XG Roller++ would have played, but the game ended boringly after

27. White 51: 22/17 6/5

28. Black 42. 7/3* 3/1

29. Double/drop 

and white wins.

Up next?

I was not sure whether rolling always the same roll can lead to a reasonable game. But this was not all bad. Of course, the AIs, despite their I in the name, never considered that the next move of their opponent might be 41 or 52, and that therefore this blot 3 away would actually be safe.

  

Up next: 42 vs. 51.

 

And then? What rolls do you think could lead to interesting games? Looking for input.