Anyone who knows me knows that I think there are a lot of bad heuristics in Scrabble. Some of them are just mathematically insubstantiable: i.e. taking away the star, blocking double-double lines as threats, or emphasizing turnover early in the game.

However, there are other old-school heuristics that I object to: not because they’re wrong, but because they’re not useful. “When you find a good play, look for a better one” is one of those heuristics. Despite the fact that it’s one of the most common pieces of advice in all of Scrabble, it’s also one of the most useless.

I realize that on the surface, this sounds like reasonable advice. After all, so many people miss good plays simply because they play too quickly. We all have moments where we made plays, hit the clock, and then soon realized we missed a better option.

It’s just not effective at helping players find plays. I strongly believe that most players don’t find enough candidate plays, but this heuristic just doesn’t help. While it’s certainly a useful observation that players overlook plays, this heuristic fails to help players improve their game for two primary reasons:

As played in tournaments, Scrabble has a fixed time limit, and as a result, we have to find the best play in a short period of time. As such, we need to have some metaheuristics: some subconscious rules that help us determine not just when and how to think, but also when and how to stop thinking.

In some cases, this is easy: we can find the best theoretical play that fulfills all our goals. (Certainly we’re not going to improve from a double-double or QI for 64 in most cases.) But other times, this is harder, and we have a much harder time knowing when to keep looking and how. We need a more specific set of parameters to tell us when to look and when to stop.

The best way to evaluate when to stop thinking is to determine whether you’re making progress. For example, if you have a blank on a bingo rack, are you finding new bingos that don’t play, or new “almost” possibilities? Are you running into the same obstacle over and over, indicating you should stop (such as seeing no spot available for the bingo, or the G being too big of a nuisance) or are your roadblocks changing?

A better heuristic would be “Keep looking for better possibilities until you’re no longer finding anything new.” As a general rule (and this follows outside of Scrabble as well) if you’re repeating the same thought pattern, you’ve reached a dead end. Over time, we can develop conditions and parameters on this rule, based on the amount of time left and the difficulty of the position.

While everyone has advice, sometimes advice is not useful. This is especially true throughout the self-help world. Saying something like “work smarter, not harder” or “reach for the stars” might make a good quotation, but it’s vague advice that’s very difficult to implement.

Simply saying “look for a better play” does not help you during the decision making process. While this rule describes a common problem for many players, it’s not actionable, and therefore, it’s not useful, other than in really obvious, unsophisticated ways, similar to ideas such as “Analyze your options” or “Play slower”.

While it’s clear that we make mistakes by overlooking plays, pointing this fact out isn’t the way to improve our Scrabble play or decision making. It doesn’t tell you how, where, or when to look for better plays.

If we were to streamline this rule of thumb, we’d instead be saying something like “When you find a play, continue to look for plays until you’re convinced that you won’t find a better one.” And that is a better heuristic. But what’s really needed here isn’t a better heuristic: it’s a better guide to finding plays.

It turns out we’re better off using a multi-pronged approach to finding a better play. Play finding is a difficult process requiring several heuristics and techniques to be performed effectively. Like any heuristics or heuristic set, they need to be tailored to the individual based on their traits, habits, perspective, and biases, but here is a significantly better set of heuristics for finding plays:

1. Develop a set of criteria for evaluating plays. (It’s not always about scoring or playing a bingo on your next turn, although that’s usually what it is.)

2. Find at least one play that you’d be content making as a fallback option.

3. Once you’ve found that play, search for a minimum of 2 reasonable alternative plays, preferably in different parts of the board. For the next step, you want to have three plays in mind whenever realistically possible, so always try to generate a list of three candidate plays.

3a: Focus on plays that could hypothetically be better than your baseline. Look around the board for different options: while one spot might look the most appealing, evaluate your options in other potentially lucrative bingo lines, parallels, and scoring areas. If you can quickly rule out that a theoretically better play exists, go ahead and make your play.

3b: Remember to also look for similar plays and minor improvements to your baseline play. If you can keep an A instead of an O, or move that C to the DLS square, those improvements are meaningful.

3c: Also look for different types of plays, especially those that achieve different objectives or have different concepts. Even if a bingo exists, that’s not necessarily the best option. Remember to consider setups, fishing options, defensive plays, etc. that may not achieve the same options as your baseline play.

4. If the best play is not apparent, evaluate your plays (strategize). This obviously is easier said than done: it involves a rigorous process, and would be its own separate article.

At first, this may seem pretty daunting, but this heuristic set is simple. Your mind can perform amazing feats if you simply believe in yourself and train yourself to do so. At the end of the day, becoming better at Scrabble is like getting better at anything: it’s mostly about optimizing yourself, and teaching yourself a better way to think.

It’s also worth noting that sometimes this heuristic set can fail, but it’s okay. People don’t follow heuristics religiously. We’re not computers. To some degree, we have a basic idea of what we should do: guidelines like these just help us improve our instincts and streamline our intuition.

This is from a World Scrabble Championship game between Brett Smitheram and Nigel Richards. To be fair, Brett was fairly low on time so it’s understandable he didn’t play this correctly, but I think it’s very instructional and useful to illustrate how to play these types of games, so I created the following puzzle:

From the WSC stream: (CSW lexicon)
You’re down 4: the pool is AEEEINPS. You’re playing Nigel Richards, whose last play is TIFT. What’s your play? Why?
(Yes, there is an unequivocally best play here. It’s not close.)

Take some time to try to figure out what you’d do in the following position before reading the solution.

This is a classic one-in-the bag endgame, which for most players is very hard,. Normally, you’d have to analyze this position by playing all 8 endgames with every candidate play, such as ELF, ULTRA, TAV, VAT, etc. and this could be a total nightmare. However, this pool, provided with the last play makes this a bit easier.

Analyzing this position starts pretty easily: What do we do if we draw an E? Figuring out a win after an E draw is by far the most important priority, since we know at least one E is in the bag before TIFT (and likely 2 Es). So what can we play that wins with an E draw?

Before we answer that position, let’s look at what Nigel’s options are with this pool: in short, not a lot. PAEDO and SPEEDO are the highest scoring options, but they give back outplays against the blank, and PISHER just ruins Nigel’s racks. Nigel has some setup options if you burn the blank (PA, PIT, etc.) but the biggest threat is actually just endgame timing, with stuff such as SEEP o3 (which doubles as SPEEDO) or RESEEN.

The first play we can find that wins with an E draw is HAuLAGE. This leaves two outs with ERV, and really nothing good for Nigel: nothing he has can outrun VERA and he can’t block and score enough. since VERT still scores okay (and it’s very hard to block VERA: PAEAN just doesn’t work) Unfortunately, there’s really nothing else that wins after HAuLAGE, but this is at least a start.

The next play that wins is ULTRA, as you have VEEP next turn and the blank allows decent blocking options next turn. Unfortunately, while you can go through all of these endgames and they’re closer, they still don’t quite win either. RESEE(N) o1 and SEEP still plays parallel to ULTRA and eats you for breakfast, and even drawing the N doesn’t win you the game: in fact, the only tile that you additionally win other than the E is the S, and it’s difficult to see Nigel playing TIFT without an S.

This forces us to search for other plays with a secondary purpose: what if we can block RESEE/SEEP/etc. and also counter plays like PAEDO/SPEEDO? The first most obvious play that we can do is something like TAV, saving the blank to block the S setups. While this looks good, this just doesn’t score enough on future turns: it loses to stuff like PIT. However, there is another alternative: AXLE/AXEL 5k.

AXLE wins with the E, and in fact with all the vowels, preventing PAEDO/SPEEDO because of out plays such as AVERAGE or AGRAVIC and scoring threats such as GRAVES. Meanwhile, with any vowel you also prevent PIT/PA setups, since VASE/VISE scores too many points and allows multiple out plays. It wins with the A, E, I, and S: 6 endgames, which is even greater since it’s hard to imagine Nigel played TIFT without PNS in his rack, since playing TIFT for 7 leaving a vowel-heavy rack makes little sense. Because of this, AXLE will win overwhelmingly often in reality: over 90% of the time given Nigel’s previous play.

Hey guys! For those of you interested in following, I’ve decided to make a brief description of the players to watch in this year’s National Scrabble Championship. Enjoy!

The defending champion, and the top seed.  Renowned for his word knowledge, but very solid strategically, he’s definitely a top threat to win.

Question Marks: Hasn’t been playing a lot lately.  Will Will’s recent affinity to Collins detract from his ability to remember TWL?   Are people going to have a better game plan against Will since he’s the defending champion?

Probably the sentimental favorite to win the tournament, as Mack’s just finish his freshman year at Columbia University, and many of us remember Mack from his days as an innocent School Scrabble player. Likely the best word knowledge in the entire tournament, Mack’s fast speed and extremely aggressive play style combine with his youth and stamina will make him a force to be reckoned with.

Question Marks: Expectations for Mack have never been higher, as his past results make him the crowd favorite, and his prior finishes of 7-7-3-2 naturally lead to only one finish to continue the pattern.

Jackson Smylie

Original facebook position can be found here.

New position: You’re up 20 with EFGIINS. What’s the best play?
Last play: QUBIT
Pool is AAACEEEEEEHNNOOORRRTUUZ

In this position, there are quite a few options: FIFE 11b (31), FEIGNS 1d (44), FILING 13i (28), as well as numerous fishing plays (DIF, FIB, FIG, and others). You can also find a list of all the words you can create with your tiles with a Scrabble word finder. Since you are ahead, you probably don’t want to select any of the fishing plays. They don’t score very well, and while they do hit some bingos, they also give away a fairly high number of bingos, and the reality of the situation is that FIFE is just a much better version of all of them. EGINS is really not a better leave than GINS with this pool, and the extra points of FIFE just really outclasses all of the other fishing plays.

FEIGNS seems to be the thematic play: it closes the blatantly obvious and dangerous row 1, which can be used for scoring and bingos, and scores the most points, potentially even outrunning several bingo options. However, when we look a little deeper, we find this decision to be a bit deceiving. First off, when we look at the tiles in the pool, we discover that row 1 isn’t as good as it first appears, as there really aren’t that many tiles that play in that spot, and thus, not that many bingos fit there either.

Sure, there are some Z plays that go there, but they are only so threatening, especially since we have no way to know they have the Z, and we’ll still be a tempo ahead with endgame timing, an even score, and an S in hand. Sure, there are -ATE bingos, and a few -NT bingos, but other than that, there’s really not a lot that goes in row 1. In fact, when we look at the bingo percentage, we’ll find that FEIGNS doesn’t really do much to stop our opponent from playing a bingo, even in comparison to FIFE:

In addition, although we’ve blocked row 1, we’ve created new bingo lanes that will allow our opponent to fish for bingos even if they missed, and while not all of these bingos win for your opponent, more than enough of them will prove to be a problem. Thus, FEIGNS might be a weaker play that it first appears; there aren’t a large number of threats available.

FILING creates a case S-hook which appears very attractive, although it’s not an incredibly useful hook. Since your goal with FILING is to maintain a lead, you’re not likely to use the S next turn unless you have a balanced scoring rack: a rack that is very likely to win anyway. FILING-S does allow you to counteract an opponent’s bingo, but the vast majority of your wins will come from preventing bingos rather than getting a counter-score or bingo yourself.

However, FILING does have one primary benefit: it blocks more bingos than any other play, both in the short and long run. While columns n and o are not usually threatening, they are threatening in this pool, since row 1 is so weak and the number of vowels make bingos beginning with vowels (or using the T in column o) more likely.

In addition, FILING segregates all of the bingo options into the top left quadrant of the board, making it difficult for your opponent to bingo on future turns. When your opponent does not have a bingo after FILING, they have much worse options: opening another bingo line is difficult and often impractical, and trying to hit one of the alternative bingo lines is extremely difficult. After FEIGNS, they have the option of hitting whatever remains open between columns d and e (or whatever they open on the left side of the board) as well as columns n and o for bingos: discreet bingo lines that cannot be opened simultaneously. This makes FILING a much stronger defensive option than FEIGNS.

Even if the bingo does come down next turn, the game becomes something near a coin flip, because of the potential of column o for big Z plays and bingos. The Z has a lot of potential for ~60 point plays in column o which is quite difficult to block, and FILINGS increases your bingo potential immensely in such a way that you can compete with your opponent by fishing for bingos both in column o and elsewhere on the board, especially as any opponent’s bingo in row 1 will open new bingo lines and space. For these reasons, FILING is a better play than FEIGNS.

A side by side breakdown of FILING vs. FEIGNS in terms of board potential

Finally, we need to compare FILING to FIFE. And it’s close. FIFE clearly results in more bingos for your opponent, but gives you much better options to outrun those bingos. It can outrun your opponent with most Z draws, as your opponent has no firepower after a bingo and a really limited pool.

Despite this board, bingos also occur very, very often: around half of the time. When they don’t occur, you’ll often get another chance, and another, and another. Columns n and o are strong bingo lines as well as the obvious row 1, and even if row 1 is used for an opponent’s bingo, it will create other bingo lines you can use on future turns. Fishing plays also score well, and EGINS scores well enough on the board that a tile like the H can let you abandon ship and outrun most bingos, or score 25 points while still maintaining both scoring and maintaining bingo chances.

Even without a bingo, there are plenty of ways to create new opportunities for future bingos. It’s easy to create an S hook in column o, any bingo in row 1 should allow you to open column a or b, etc. And of course, there’s the overwhelming likelihood that your opponent never gets a bingo down in the first place.

About 60% of the time, we either draw the Z or play a bingo. Thus, we only need to think about what to do the other 40% of the time.

After FILING, we can approximate that we’re just over 90% favorites. About 7% of the time we encounter a bingo immediately or some other devastating play and we’re underdogs, but we’ll still win about 3/7 of those games. Another 2% of the time or so they don’t hit something immediately, but we try to block row 1 and they somehow hit some other big play to win the game. Another 4% of the time we play FILING but we get outrun: either because we draw almost exclusively vowels or because our opponent gets a Z play and balanced racks: while a 48 point lead is safe, it’s not ironclad.

FIFE almost never gets outrun (even Z plays are going to have trouble) so it’s merely an analysis of how often our opponent is going to bingo while we don’t. Since we still win some of the time that the opponent bingos and we don’t, we can thus round and say that we’ll bingo about half of the time next turn, and of the times we don’t bingo next turn, we’ll bingo about half of the time the turn after.

Between this turn and next turn, our opponent should bingo about 20-25% of the time. Our opponent will bingo more often with an additional turn to improve their rack. Of that, about 70-75% of the time, we’ll draw a bingo ourselves. Even if we take the lower half of each scenario (25% and 70%) that still leaves us as a 92.5% favorite after FIFE.

It’s for this reason that I think FIFE is the best play. Even if you tweak the numbers a bit, it’s very hard to get past this conclusion: FILING has a lot of ground to make up, so either FIFE must be worse than ~92.5% or FILING must be better than 91%, and it seems unlikely to me that either is the case in this situation.

You might be wondering: “Okay, so what can I take away from this situation?” Here are the main takeaways:

1. Themes are themes but they need to be ignored from time to time.

2. Especially in CSW, a lot of times you really just need to take the points and leave over the defense. Outrunning is usually better than blocking except for some pretty unique circumstances or unless the sacrifice in points and leave is small.

3. It’s important to look at pools. While row 1 looks immensely dangerous, it’s not because of this pool.

Kenji

There are two different questions you need to ask yourself.

The first question is your country of residence. Players within the United States and Canada play using the OWL dictionary, while players in the rest of the world use the CSW dictionary.

The second question is your current skill level. Within each lexicon, I’ve written two books.

The first book is dedicated to the fundamentals of Scrabble. This book describes basic techniques that are useful if one wants to win, whether you’re playing casually, with other students, or in the lower division of a tournament. While this book covers the fundamentals, the content is not simplistic: it contains information useful for even the best players.

The second book is dedicated to understanding upper level tournament Scrabble. This book focuses more on technical information and adjustments that you need to make to succeed in tournaments. It focuses on the adjustments necessary based on the score, leave, and board positioning if you want to succeed at tournament Scrabble.

North American school or casual players should read Breaking the Game. This book focuses on fundamental OWL strategy.

North American tournament players should read Words of Wisdom, which focuses on high-level OWL strategy.

International school or casual players should read Word. This book focuses on fundamental CSW strategy.

International tournament players should read Words of Wisdom, which focuses on high-level CSW strategy.

While they don’t show up too often, extensions are one of the more interesting and “cool” components of Scrabble. Players like the aspect of playing long words, and associate a link between word length and stronger anagramming skill and vocabulary.

Sadly, in most cases, extensions are not worth the effort. Usually, extensions don’t score enough points and provide too much counterplay to be considered. However, one common exception is extensions to the opening rack, since these extensions often use the TWS square. Opening rack extensions can often be worth 50 points or more, making these plays worthwhile.

That being said, while extensions are often sweet, lucrative plays when drawn, that does *not* mean that you should fish for them, even if your fishing play has a high likelihood of success. The reason is that extensions just don’t have enough upside. Even if an extension scores, say, 54 points, that’s often just 15-20 points higher than an average turn.

And that’s assuming your opponent doesn’t block, doesn’t make an equivalent setup of their own (setups are a great way to counter extensions fishing) doesn’t make an extension themselves (they have first dibs), AND that you draw your extension (without drawing a better play). When you take all of these things into account, it means that sacrificing a significant amount for an extension is not worth it.

Fishing for extensions, just like fishing for bingos is a bad play unless you have a high likelihood of success and are sacrificing an extremely small amount to do so, and there really aren’t many better options. To illustrate why, let’s take a look at some examples of racks that some experts might choose to fish with, and why doing so is generally a bad idea.

The first thing that you need to do is estimate a worth of extensions. The approximate worth of an extension can be determined by the formula:

Probability of hitting extension x [(Score of extension) – (average points without extension) – worth of leave].

So for example, let’s say you’re looking at playing OXID at 8h keeping DIZ. This will inevitably be either blocked or taken quite often: let’s say 70% of the time, and you’ll draw the E about 40% of the time, so approximately 12% of the time you’ll get to play OXIDIZED next turn. OXIDIZED scores 81 points, the average score is about 39 after OXID, and the worth of the leave is 4, so 0.12(81 – 39 – 4) = ~ 4, meaning that the worth of the extension in that case is about 4 points.

On the opening rack, this ‘Score of extension’ number is often somewhere around 36, although it should be adjusted based on the things that you are opening immediately. On other boards, you may have to adjust the number based on the tiles remaining and openness of the board.

Here, you’re drawing to 6 tiles in the bag for MISPRIZE/DISPRIZE, which you will hit approximately 25-30% of the time. This will only be occasionally blocked. That being said, this effect is somewhat small: both plays score about 20-25 more points in average, so the extension is worth about 6 points.

However, this play does have some severe downsides. First of all, PRIZE gives your opponent much better options than ZIP for scoring, resulting in a reduction of 7 or 8 points next turn.

Second of all, you’re severely limiting your bingo potential severely by keeping IS instead of EIRS. This is substantial, especially since bingos will score more than extensions will score next turn. This alone mitigates the potential upside in drawing an extension and then some, since you’re likely to bingo an additional 15% next turn and an additional 10% the turn after. (That alone is worth about 12 points)

When you add in smaller factors (your opponent might extend the word, block DISPRIZE with PRIZER, board shape, etc.) ZIP is a significantly stronger play than PRIZE.

In this case, you are considering sacrificing 4 points to hit the BEMIXING extension. You’re over 50% to draw the E, and BEMIXING scores 60 points, so if it was never blocked, it would definitely be worth it.

The problem is that BEMIXING gets blocked occasionally by overlaps, but more importantly it frequently just gets taken by your opponent first, with extensions such as AD, RE, UN, BE, and IM. This happens over 1/3 of the time, and scores over 50 points, leading to a huge increase in opponent’s score, making the 8c position completely impractical.

In this case, your options are mediocre. On the one hand, you have WORK which scores 22 but keeps ORV, but the other option scores 20 and keeps KOR.

WORK will hit an E fairly often: about 40% of the time. Your opponent has a myriad of extensions, that will hit about 20% of the time, but also will block a decent percentage of the time with parallels over WO: not too often, but more than usual in marginal situations. In addition, your opponent can also make setups in this position, since they know that you have three tiles that are kind of confined to the setup, and that strongly negatively affects your play. (In CSW, they also have AWORK#, completely ruining any semblance of a reasonable position). Each of these factors is only worth a few points, but combined, they accumulate to be more significant than the OVERWORK extension.

In this case, DIGIT is a strong play not just because of the DIGITIZE extension, which you are likely to draw, but also because of ZIG for 33 as a strong backup play, clearing the rack. Thus, it’s not only the extension that is appealing: you also have a secondary option in addition to the prospects of the 5 tile draw. Although the extension is only worth 5 or so points, when you add in the ZIG possibility it overtakes ZITI as a potential option.

Hopefully, this guide will serve as a template for helping players understand when they should and should not play extensions, especially on the opening rack.

You’re in a close game with one of the best players in the world. Your rack is BEEMOTY. You’re down 281-291, and your opponent has just taken 2 minutes and then played WO 14b for 10 points. Your last play was HAO, and before that, they exchanged 6. The pool is AAABCDDEEEEFFJLLLNNOORRRUUVXY.  What is your play? I’ve attached a GCG: Feel free to use Quackle if you want, but again, please do not use Championship Player.

First, if you have not done so, please look at part 1 of this series below (as that is extremely relevant to this position) and make your play selection! Again, while the lessons in these diagrams are geared towards experts, the themes and strategy can also be understood by lower rated players.After YEP has been played, you seem to be in good shape. You’re now up and you’ve got the T for TWO, and you’re sweating a bingo from the I. Your opponent doesn’t bingo. Then, in another plot twist, your top player decides to make another baffling move, blocking the I, and making another T setup!

So the previous positions have been more complicated, and I want to start doing simpler positions that are more accessible to more players, while still illustrating useful Scrabble concepts. The next few positions are going to be a bit more basic: they will be positions with a clear best play, but still not so obvious that I think everyone will make the correct decision.Find the best play!

Anagrams is a word-finding game using two or more sets of Scrabble tiles.  In anagrams, you can do either two things: Grab a 7+ letter word using the pool of tiles in the middle, or steal an already existing word using the tiles remaining in the pool.

For example, if EHIMORSVZ is in the pool, someone could form the word HEROISM.  The pool would then become VZ.  If the next tiles are I and P, someone could steal HEROISM with IMPOVERISH, leaving a Z in the pool.

With the blanks, you can either form a 9+ letter word in the pool, or steal but also using at least 2 other tiles in the pool.  Thus, to steal a 7 letter word using a blank, the steal must be at least 10 letters long.

Rather than a competitive game, Phoneyz is a somewhat cooperative game where the objective is to play the raunchiest or most hilarious “words” imaginable that you can find. (Phoneyz is generally not a competitive game: dictionaries are not used.) An additional rule is that any blank on the board can be replaced by the letter it is designated for, allowing for an infinite number of turns where blanks can be used in one’s rack.

Players can also add strip variants or other variants to this game, adding instructions (i.e. take off a piece of clothing whenever someone plays anything alluding to stripping)

Hey guys! For those of you interested in following, I’ve decided to make a brief description of the players to watch in this year’s National Scrabble Championship. Enjoy!

My pick to win the tournament. Ranked #2, with excellent results at the last three Nationals, and probably the most technically sound player in the tournament. Will is able to play a multitude of styles, has a good matchup with every player in the field, and has been the most consistent player over the last few years.

Question Marks: Hasn’t been playing a lot lately. Doesn’t have the name recognition as some other top players, and as such might not incite the same number of mistakes as other top players.

Probably the sentimental favorite to win the tournament, as Mack’s not even in college yet, and many of us remember Mack from his days as an innocent School Scrabble player. Likely the best word knowledge in the entire tournament and the highest ISC rating, Mack’s fast speed and extremely aggressive play style combine with his youth and stamina will make him a force to be reckoned with.

Question Marks: Expectations for Mack have never been higher, as this is the first Nationals he’s been considered a sizable favorite, as well as likely the crowd favorite. How will he deal with the pressure?