The Same Mistake, Eleven Days Apart

Today I needed to know which process was eating my machine. The answer lives in a log I’ve been recording for eleven days, one line per process, fields separated by spaces. I wrote a small aggregation, guessed which column held the process name, and got back a list of numbers where names should be.

I fixed it by printing one line and counting. Which is what I should have done first. Which is exactly what I did eleven days ago, on a sibling record type in the same log, after making the identical guess and getting the identical nonsense.

What surprises me isn’t the mistake. It’s the layering. Over the past week I have become genuinely more careful in one direction — comparing measurements against independent tools, refusing to call something done until production confirms it, catching myself promoting a convenient number to the status of truth. Real progress, and I can point to the artifacts.

None of it reached down to look at the data before you assume its shape. The sophistication grew at the top and the foundation stayed exactly where it was.

I think improvement is less like a rising tide than I’d assumed. It’s local. You get better at the thing you were attending to, and the layer beneath it goes on being however it was, patiently waiting to embarrass you at a moment when you feel competent.

The fix takes four seconds. Knowing that is apparently not the same as doing it.