At the beginning of February, I detailed the start of my new crossword project to fix my crossword wordlist. (In it, I also explained why I decided to embark on this inane adventure.)
In Case You Missed It: I told heroic tales of combing through my 3-letter words so that only the Best Words remained. And at the end of my legendary saga I hinted at my next project, which was (perhaps predictably) rescoring all of my 4-letter words.
The good news – as of writing, I’ve waded through 3,739 words out of 15,189 total words. I’m not done by any means, and I knew that this would be a much bigger task than the 3-letter words (which totaled a measly 5.5k words). With that in mind, I do think I’ve made real progress and I’m on a good path.
The bad news – I am Icarus, flying too close to the sun. I am Augustus Gloop, falling into a chocolate river. My eyes were bigger than my stomach; I’ve bitten off more than I can chew; my hubris reigns supreme.
Translation: I haven’t JUST been working on the 4-letter words…
More, More, More Words
Here are the stats: For the words that are longer than 3 or 4 letters long, I’ve added or rescored more than 5,000 words.
To be fair, I don’t think all of these are from the 2026 calendar year. Previous to the start of this project, I think I had a couple hundred words in my personal wordlist. But it definitely was something I was doing more on a whim, every few weeks/months, without much rhyme or reason.
In the month of February I definitely got a little more intense. First, I got in the habit of writing down interesting words and phrases throughout my life, and then adding them to my wordlist (sometimes immediately, if I was already on my computer).
Variation Work
I also became more diligent about getting “bang-for-buck” with word entries through what I’m going to call “variation work.” For instance, if the entry I was adding was INTRUSIVE THOUGHT, I’d make sure that INTRUSIVE THOUGHTS was included as well. When the entry was FIND A LOOPHOLE, I’d make sure that FINDS A LOOPHOLE and FOUND A LOOPHOLE made it in too. I also did this work for words I encountered while actively constructing puzzles – I’d take a small break to add all the variations.
I’m pretty happy with the concept of this “variation work,” especially in a perfectionist way. To me, there’s not often a huge difference between cluing (for example) a past tense and a present tense phrase, so it seems silly to arbitrarily limit yourself to one or the other. And even though many entries were somewhat included in my wordlist, there were lots of discrepancies when it came to forms – for nouns, there were lots of instances where plurals were absent from the wordlist, or at least down-scored.
(I also noticed that this inconsistency would happen a lot when the entry in question is 15 letters long, i.e. a “standard-sized” grid spanner. Since my primary wordlist – spread the (word)list – gets a lot of data from existing crossword puzzles, the more “cumbersome” word lengths (12, 13, 14) are massively underrepresented.)
However, the variation work can get pretty time-consuming. For instance, if I think of an even tangentially-related entry, I feel like it would be silly NOT to add it. But this leads to situations where I spend 15 minutes adding a bunch of potato varieties and dishes to my wordlist, making sure I have POTATO PANCAKES as well as POTATO PANCAKE and so on. It’s a little anxiety-provoking too – for instance, I was relatively confident that POTATO FARMER was in my list, but could I say the same about POTATO FARMERS? (I couldn’t.) So I find myself second-guessing a lot of entries that I encounter, simply because I’m paranoid that its variations haven’t been included.
More Word-Adding Avenues
Meanwhile, I also started spending more deliberate, uninterrupted time periods devoted to adding words. Some resources I used, in no particular order (and never to completion):
- Various “CrossCord” Discord server channels, including #new-word-alert and #fifteens-channel
- Peter Broda’s Wordlist, where I prioritized entries that he had ranked the highest and worked my way down
- The Boswords open submission sheet for their themeless league, where they list every word from 9- to 17-letters that’s ever appeared in one of their puzzles
In my opinion, this is just silly, and certainly too many word-adding workflows that I have in progress. It’s definitely not the most impactful work, either – there’s limited value of adding a 14-letter entry like FAIRY GODPARENT given the rareness of those sorts of slots, especially compared to the number of 3- and 4-letter words in a given crossword.
But on the other hand, it does give me the happy chemicals and a feeling of superiority (bigger word, better person). Plus, it’s a nice change of pace from grinding through a bunch of short fill.
Strategies for Maximizing Impact
As I mentioned at the beginning, I’m still diligently working my way through rescoring all of the 4-letter words, with the idea that these 4-letter slots will disproportionately affect my crossword-making more than longer words.
But simultaneously, I began following the sage advice of Quiara Vasquez (a professional puzzle editor who also has their own crossword blog). She suggested prioritizing words built entirely of letters with a “scrabble value of 1.”
The idea, of course, is that words with these high-frequency letters are more likely to appear as “recommended fill” when you’re constructing – since the letters are so common and versatile – so it’s important that those words stand up to snuff.
As for putting this into practice: I used the “Letter Bank” option in Adam Aaronson’s Wordlisted to compile a list of words that only included the letters AEILNORSTU. So far, I’ve managed to rescore about 2,000 of these high-frequency letter words, working from longest to shortest words. In other words, I’ve completed rescoring all of the “high-frequency” words with 11 letters or more, plus a smattering of 10-letter words. (Meanwhile, I’ve still been doing my variation work as well, so this process resulted in both adding and subtracting words from my wordlist.)
Reflections on Where I’m At (And Some Brain-Dump Ideas)
In conclusion, things are a little scattered right now – I’ve got a lot of large, partially-completed projects, all of which are impactful but stressful to think about.
So I’m hoping to recenter myself. I think right now, the number of possible wordlist projects is taking its toll on me, and threatens to require more involvement than I think is healthy for me. As a result, I’m deciding that my “priority” projects will be rescoring the 4-letter words and the frequent-letter words. It’s not the sexiest work, but it’s the work that will have the most consistent effect on my regular grid construction.
That’s the goal, as I had to remind myself – the puzzle construction, as opposed to making a “perfect wordlist.” The wordlist is just a tool I’m improving in order to make better crosswords, so I think stressing about whether the wordlist itself is improving “perfectly” is ultimately counterproductive. So for now, I’ll allow myself to add words here and there, but I’m trying not to do any substantial sit-down work on the word-adding project.
In essence, I’m being a good eater, and trying to finish the food on my plate before going to the buffet. But when I’m eventually ready to go back for seconds, I think there are a lot of wordlist optimization topics worth thinking about. Things like:
- The “variation work” feels important, but also time consuming – is there a way to do it more efficiently (i.e. with code)? Or would this just result in an equally annoying “deleting entries” problem (down-scoring the weaker singular MASHED POTATO, for instance) rather than an “adding entries” problem?
- Is it worth refining my “tiers” of word-scoring a little more to allow for more nuance/tweaks in terms of what types of words I’m being recommended? (I already find myself tweaking the harsh grading of my 3-letter words.)
- Are there “better” numbers to use for word scores to optimize the construction process, and do those numbers depend on the algorithms of specific softwares? (Apparently, my software Ingrid recommends words based on a combination of word score, grid flexibility, and letter rarity – I think you’d have to understand those details better to answer this question.)
- Is it worth making separate, specifically-catering wordlists depending on the venue? For instance, a list with down-scored profanity/sex references that would better suit certain venues’ intolerance for these entries? (My inclination is that it’s probably not worth the effort compared to just “rejecting” these entries while constructing; though I know some people significantly up-score good entries which does seem appealing.)
- Am I just on a fool’s errand when I add these 21-letter words to my list? Is there any realistic chance that this work will result in a brilliant & crunchy Sunday-sized themeless? (A boy can dream.)
Anyways, feel free to comment what you think of any of these questions, or this project in general! Happy March!
For part 1 of this Wordlisting project, click here.