4. All the Other Words

 The question, then - where to start? We have an entire grid to fill out - how do we decide where to put our first words?

Our starting grid.

My software has a 'Best Location' button, which highlights the slot with the fewest options to pick from. For 4-Down, it warns that only IRAQI makes for a workable grid - let's put it in.

After that, it suggests that I look at 13-Across - which crosses the last I of IRAQI followed by the A of IQALUIT. We have a few decent options: GENIAL, MEDIAN, DENIAL, INDIAN. I'm unlikely to pick INDIAN, because that would mean crossing a 'geography' word with another - and that's another layer that we should be mindful of, as we pick our words.

Our biggest concerns, in rough order of priority, are:

• Each word in the grid must be a legitimate word. 
• This means that we should exclude random strings of letters; but also that we should avoid entries like REPLAYERS or LAURAS, which could be grammatically-correct ways to refer to people who hit replay or a group of people named Laura. That's technically correct, but we're aiming for better than that.
• Each word in the grid must be unique.
• ICE HOCKEY and ICY are too close semantically to have in the same grid. With so many words that we could use, there's no reason to not diversify.
• While ICE HOCKEY and BOBSLED are similar in being wintery sports, they don't repeat a singular word.
• Likewise, WORKING-CLASS and WFH should not be included in the same grid, because - even if the grid itself has no repetitions - someone reading out the answers would see WORKING-CLASS and W(ork)F(rom)H(ome). 
• There should be a minimum of unfair crossings. 
• The definition of 'unfair' is completely subjective: some parts of your audience are going to be less familiar with geography, some with politicians, some with sports. Every one of them should be able to enjoy your puzzle. 
• Our goal, then, is going to be to avoid crossing entries from a certain type (sports players, demonyms, acronyms) with each other. If a solver doesn't know the first, they might not know the second either, and they would then be left completely stuck.
• The ideal puzzle, then, would avoid any cases of crossing proper nouns/acronyms with the like. Use your judgement in terms of what counts as common knowledge.

In this case, because I have a few good options and only one I would want to avoid, I don't want to lock myself in too early. Luckily, I can thread the needle: by marking the second letter to be an E, the software will continue considering GENIAL, MEDIAN, and DENIAL as options (because they follow our ?E?IA? pattern), while eliminating INDIAN. 

At a crossroads.

With that knowledge, the constructing software now warns me that there are only two options for the selected slot, one of which is TRIED ON. We can already see our options narrowing: the black letters are the ones I've typed in myself, locking them in; the pale grey letters are just one of the options proposed, but we're by no means forced to accept them; but the dark grey letters are those that, the software warns, will be forced should I select TRIED ON as my option. 

The SCIENCE option.

The other option I have for this slot is SCIENCE. I'll take that between the package-deal of TRIED ON and TERRA, so SCIENCE it is. Making that choice means that I have to accept BASIC, ACCRA, and either BAA/ALIAS or BAE/ELIAS to fill out that corner - I have no flexibility. It does mean crossing ACCRA with IRAQI; to compensate for that difficulty, I'll pick BAA/ALIAS to ensure that the A of IRAQI crosses the evocative ALIAS rather than the proper name ELIAS.

The rest of the grid is filled with this sort of trade-offs, sometimes conceding that I have to restart an entire section of the grid to avoid certain combos. 

The grid is now segmented.

My personal strategy, with a strategic focus on the longest (and therefore least flexible) entries, is to segment the grid as much as possible. By filling in the top-left-bottom-right diagonal, this means that I can play around in the top-right or bottom-left corners to my heart's desire without affecting my options on the other side. 

An important note for this strategy is that I'm following the symmetry of my puzzle. By following that axis, it means that I can react in real-time to any helper squares. If I had gone around the grid in a clockwise manner, any work done on the top-right corner would be dependent on my choices in the top-left corner; and if I decided that I had to place a helper square in the bottom-right corner, that might mean having to redo the symmetrically-placed top-left - while also rejigging the top-right corner impacted by both. Therefore, in a grid with horizontal/vertical symmetry, you may want to fill it in in rows/columns rather than in a diagonal.

A grid that didn't make it.

The middle of the grid is one of the most sensitive parts of a grid (at least, an open one), because changes anywhere else in the puzzle will have ripple effects going through it. That unfortunately locks me into either LLM or LLC - LLM can be clued in relation to the Large Language Models powering various AI tools, and so I can live with it. However, as much as I love the two Zs contained in ABUZZ, I don't want to subject the solver to back-to-back abbreviations: 23-Across INC followed by 24-Across LLM. Alas! 

Of my options for 23-Across, ORC is one of the few common words ending with C, so I'll take that path. I do have thoughts on how to clue ORC, but this is a delicate time - I need to think of the good of the greater crossword before committing to a word, as much as I might love to reference Dungeons & Dragons or World of Warcraft with ORC. Especially for such a short word, there will be other opportunities.

A complete grid.

ORC does seem like a good fit, though, and so I continue on. PINKY becomes another word that strikes my fancy, both for its letters and just for having a fun 'feel' to it - thankfully it fits in nicely in the bottom-left, between SPAIN and ANNEX! Once the top-right falls together, I can rest for a bit, double-check my grid for any words that are a little too similar to each other, and tinker with different corners.

I'm frankly quite happy with this grid, though - these are all words that are in reasonable usage, and there is good potential for a variety of cluing angles. 

Even better, I've met a personal goal: there are no 'S's in any bottom-right corners! This is very much a rule of thumb, but a lot of words finishing in S are either plural nouns or conjugated verbs - which makes them longer than the base word, but not any more interesting. In my view, those slots are often better by 'plain' , unmodified entries, which is what I've done here.

Time for the cluing.

Crosswords in the Age of Chat-GPT

You'll notice that all of the actual work to put together our grid, while assisted by technology, was done by a human. Software is not yet ready to put together a proper crossword grid: as you can see further down this section, software struggles mightily even when a workable grid is provided (as in the right two cases).

That's not to say that you can't use AI in your crossword construction, but it will primarily serve as a second pair of eyes rather than a breakthrough assistant. Two prompts you can use are:

• Can you screen this list against your general audience checklist?

• Can you ensure that this list respects Canadian English conventions, such as spellings. Also, please indicate any direct connections to Canada or the Canadian experience.

While you should be prepared to assess your words yourself, these prompts can provide an additional pair of eyes to find what may lie within your cultural blindspots. In this case, ChatGPT urged caution around terms that can feature in geopolitical discourse (IRAQI, RPG (as a rocket-propelled grenade, rather than a role-playing game), ORC) and flagged potential implications around the words TABOO and DELETED. It also noted the Canadian spelling of LITRE.

In this case, enlisting AI review did not uncover any major issues, either positive or negative. 

Grid put together by ChatGPT.

Software with my wordlist.

Software with Internet wordlist.

To give credit where credit is due, there is some beauty in randomness - my software placing 11-Across TERRA across from 12-Across COTTA , in the middle grid, has a certain charm to it. That is unfortunately undone by the slew of acronyms across the middle, and the heavy geographical focus in the bottom-left.