When I was a traveling consultant, one of my favorite little games was grabbing the USA Today puzzle section from the Admirals Club and trying to finish the Sudoku before the wheels of the jet left the ground. I got better over time, though I still didn’t finish it before every takeoff.
Eventually, I added crossword puzzles to the mix, largely due to the influence of my friend James. I watched him work through them and wanted to learn the tricks myself. On Sundays—usually parked on the couch watching NFL games—I’d have the Sunday crossword nearby (mostly unfilled). But week by week, I improved. One unexpected perk of getting the Sunday paper delivered (at least on our route) is that you also get a Wednesday edition.
A few years ago I suggested to James that we put these free Wednesday puzzle sections to good use instead of tossing them in the bin. What began as a casual, nerdy side quest quickly turned into a standing weekly ritual religiously observed every Wednesday—or as close to it as schedules allowed. Each session follows the same order: Sudoku first, then the New York Times crossword, and finally the United Media Daily Commuter crossword.
Then I had a silly idea: what if we timed ourselves every week and tracked it? At first it was just for fun. We documented dates, completion times, and a few notes about the puzzle. We ran some basic stats (mean, median, standard deviation) and made a simple graph.
At some point, this stopped being a joke spreadsheet.
Once the dataset got large enough to feel real—157 data points representing three years of Puzzle Wednesdays—I treated it like any other analysis problem: clean the data, normalize what needed normalizing, and then start asking questions without strong prior beliefs about the answers. The core dataset is intentionally simple—date, total completion time, and a handful of notes—which made it easier to experiment.
From there, I started merging in external data sources aligned to the same dates. Weather data came first, followed by broader environmental variables, market data, and a few intentionally absurd dimensions. The goal wasn’t to prove a hypothesis so much as to see whether anything showed a meaningful relationship with how fast we solved puzzles on a given Wednesday.
Most things didn’t.
This is exploratory analysis, not an attempt to publish a paper or claim causality. But it is a real dataset, collected consistently over three years, which makes it a surprisingly rich sandbox for asking questions you wouldn’t normally bother asking. If you think I missed an obvious (or ridiculous) data source to include, I’m open to suggestions.
The funniest result by adding in weather data: higher wind speeds correlate with slower completion times. Not rain. Not temperature. Wind.
I don’t have a theory for this. Maybe it’s noise. Maybe windy days are subtly distracting. Maybe it’s a proxy for something else entirely. Or maybe it’s just a reminder that humans are weird, rituals matter, and patterns will emerge whether we fully understand them or not.
That’s really the point of this whole exercise. This started as two friends doing puzzles to keep their skills sharp. It turned into a weekly ritual. Then it quietly became a three-year dataset. The analysis is fun, but the consistency is the interesting part—the discipline of showing up every Wednesday, starting the clock, and doing the work.
I now present to you the data, the methodology, and the charts—because of course there are charts. Here’s the analysis of our Puzzle Wednesday quests for the last three years (yes we will keep tracking and yes I will post updates!).