Keep an error log for your judgment
Pilots, surgeons, and forecasters all keep records of when they were wrong. Most of us never do — and our judgment quietly stops improving. Here's a lightweight version anyone can run.
The best forecasters in the world share one unglamorous habit: they write down what they predicted, and later they write down what actually happened. Not the dramatic calls — all of them. The boring ones, the near-misses, the times they were almost right in a way that turned out to matter.
The rest of us mostly don't. We make dozens of small judgment calls a day — this meeting will run long, that feature will take two days, she'll say yes — and then we never check. The outcome arrives, we update our mood, and the prediction itself evaporates. We never find out how calibrated we actually are.
Why memory won't do this for you
You might think you don't need a log, because you'll remember when you were wrong. You won't, and not because your memory is bad. It's because memory is helpful. After the fact, it quietly edits your prior estimate to sit closer to what happened. You end up feeling like you "kind of knew," which is the precise feeling that prevents you from learning anything.
This is hindsight bias, and you cannot introspect your way around it. The only defense is a record written before the outcome was known — a timestamp your memory can't reach back and revise.
A format that takes thirty seconds
An error log doesn't need to be elaborate. The entire value is in capturing the prediction before reality lands. A single entry can be this small:
{
"date": "2026-05-28",
"claim": "This refactor takes one afternoon.",
"confidence": 0.8,
"outcome": null
}You set outcome later — true or false — and you do nothing else clever. The magic isn't in the analysis. It's that, a month in, you can scroll back and see something you could never see from the inside: the shape of how you're wrong.
What the shape tells you
Patterns surface fast once the data exists. Maybe everything you mark at 90 percent confidence comes true only 60 percent of the time — you're systematically overconfident, and now you know the discount to apply. Maybe you're sharp about technical estimates and consistently wrong about how people will react. That's not a flaw to feel bad about; it's a calibration you can correct, but only if you can see it.
This is the whole idea behind the error log inside Dragon Slayer. Log a claim and your confidence in the moment, resolve it when the outcome is known, and let the gap between the two become visible over time. Avoidance is about the tasks you won't face; calibration is about the judgments you can't check. Both fail silently. Both get better the instant you write them down.
Start a log this week. Five entries is enough to feel ridiculous about the first one you resolve — and that flinch is the learning.