My actual job title has the word “planning” in it, which implies a level of control over outcomes that the outcomes do not, in fact, respect. Every fall I build a demand forecast for winter gear, tents that get repurposed as ice-fishing shelters, headlamps, the specific brand of hand warmers that sells out every single year despite my model insisting we’ve ordered enough. Most years the model is close enough that nobody complains. Two years ago it wasn’t, and I still think about why.

We’d had a mild stretch heading into what should have been peak cold-weather buying season, and my model, reasonably, forecasted lighter demand off the back of that pattern, people don’t buy insulated gear as urgently when it’s fifty degrees in November. Then a chinook wind blew through in early December, the kind that swings the temperature forty degrees in an afternoon, and two days later an actual hard cold front came in right behind it, the way it does out here, warm gust then a slap of real winter, and suddenly everyone who’d been putting off buying a decent coat needed one at once. We sold out of three product lines in a week. My model had been right about the recent trend and completely wrong about what the trend was actually predicting.

What got me wasn’t the stockout itself, stockouts happen, you adjust and reorder and move on. What got me was realizing my forecast had been built entirely on the assumption that weather moves in trends, that mild Novembers predict mild Decembers, when anyone who’s lived here more than a season knows the whole personality of this climate is sudden reversal. I’d modeled the wrong variable. I was tracking temperature when I should have been tracking volatility, how likely a swing is, not which direction the average is currently pointing.

I rebuilt the model that spring with a volatility term baked in, and it’s held up better since, though “better” here means “wrong less catastrophically,” not “right.” I’ve made a version of this mistake in other places too, if I’m honest, treating a comfortable streak as a trend instead of asking how likely the streak is to reverse. It’s the same error, dressed up in different data. A run of easy wins at game night doesn’t mean I’ve gotten better, it usually means I’m about to get complacent right before someone reads my tendencies the way Dana eventually did.

The tent season recovers every year regardless of what my spreadsheet predicted. I’d like to say I’ve fully internalized the lesson, but mostly what I’ve internalized is a healthy suspicion of my own confidence whenever a forecast feels obviously right. That feeling, I’ve learned, is usually the tell that I’ve stopped checking my assumptions and started just trusting the streak.