I used to hunt birds. I had reasons, and they were the ordinary ones. It was something I’d grown up around, and I had a tidy justification for why it was fine.
My sister ended it with a division problem.
The equation
The question she asked was how many animals die per meal.
Per meal, which is the unit at which a person actually makes the decision. Nobody chooses their annual consumption. They choose dinner.
Run that and something uncomfortable falls out immediately. Small animals are catastrophically inefficient by this measure. A bird feeds one person once, maybe. So does the next one, and the next. Ten birds, ten deaths, a handful of meals.
A cow sits at the other end. One animal produces an enormous number of meals, feeding a large number of people. Per meal, the arithmetic isn’t close. It isn’t within an order of magnitude.
Which meant the thing I was doing (the one I considered the more skilled, more traditional version) was by this measure by far the worst option available to me.
The part I didn’t expect
What got me wasn’t that I’d been wrong. It was that I had never run the number.
I had a whole position. I could defend it in conversation, fluently, and had. It was internally consistent and it had never once been checked against the simplest possible measurement. There was no moment where I considered the arithmetic and rejected it. I’d simply never done the division, and the absence had felt exactly like having thought about it.
That’s the part I still think about, years later, and the reason this is worth writing down rather than being a story I tell at dinner.
The feeling of having a considered position and the fact of having one are not the same, and from the inside they’re indistinguishable.
I’ve since found this everywhere. Architecture decisions defended fluently by people who never measured the thing they were optimising. Technology choices with a whole narrative attached and no benchmark underneath. In every case, including my own, the fluency of the defence was unrelated to whether the check had ever been run.
If anything, fluency is a warning sign. Positions that have survived a real measurement tend to get stated with more hedging, not less, because the person knows exactly which conditions they tested and which they didn’t; certainty is cheapest for the people who never looked.
What “run the division” means at work
The crude, obvious, slightly embarrassing measurement is different in every domain, which is why people skip it: there’s no standard form and nobody asks you for it. Some that have paid off for me:
- Per unit, not in total. Almost every misleading number is a total. Cost per request, or minutes per person per week. The total flatters; the per-unit version is the one that decides anything.
- Count the thing you’re actually claiming. If the claim is “this saves the team time”, count hours rather than tickets closed or lines changed. The proxy is where the self-deception lives.
- Compare against doing nothing. Half of all improvements are improvements over a baseline nobody measured. Sometimes the baseline was already fine.
- Time the loop. For anything about developer productivity, the honest measurement is how long it takes to find out whether a change was correct. It’s the number that predicts almost everything else, and almost nobody has it.
None of these take more than minutes, and they occasionally demolish something you’ve believed for years, which is the best return on minutes I know of.
The unfairness that falls out
There’s a second thing in the arithmetic worth sitting with.
The cow, the animal producing by far the most value per death, receives the worst treatment of any of them. The efficiency doesn’t translate into anything better for the animal. It translates into industrial scale, which translates into conditions that wouldn’t survive a minute of honest description.
So the measure that says “this is the least bad option per meal” is silent on the thing that would actually matter to the animal. The arithmetic answered the question I asked. It had nothing to say about the question I hadn’t thought to ask.
I don’t have a resolution to that. I’m not writing a position piece on food ethics and I’m not qualified to. What I have is the observation that a metric which correctly ranks your options can still be measuring the wrong thing entirely, and that noticing this is a separate skill from being able to compute it.
Anyone who has watched a team optimise a number into the ground will recognise the shape. Response times that improved because the slow requests started timing out. Ticket counts that fell because people stopped filing them.
What I actually took from it
Three things, in descending order of how often they turn out to matter.
Run the division. Whatever the crude, obvious, slightly embarrassing measurement is, do it before constructing the argument. The order matters more than the measurement does: an argument built first will find a number that supports it.
Notice when the defence is too smooth. If I can explain why I do something without hesitating, and I can’t immediately name the evidence, I have a rehearsed position. The hesitation is the tell: real positions have edges the person knows about.
Let people who aren’t invested do the checking. My sister could run that number because she had nothing at stake in the answer. I couldn’t, because I did, and not consciously, which is precisely the problem. Everyone needs somebody with no stake who’s willing to do the arithmetic out loud, and the more senior you get the harder that person is to keep around. It’s the same reason judging your decisions on process rather than outcome requires writing things down before you find out.
A moral argument would have failed; I’d probably have argued back and enjoyed it. I stopped hunting because someone did a division I could have done at any point in ten years and never had, and once the number existed I couldn’t go back to not knowing it.