::To be expanded with links and theory and stuff::
This tool is designed to help you calibrate your estimation biases. Your goal is to be about 90% accurate, which means that you want to get exactly 9 of each 10 questions correct. Don't try to game the system, as this process can provide you valuable feedback about how you make educated guesses.
You'll be shown ten questions requiring a specific numeric answer. You probably won't know the exact answer, but in some cases you'll have some domain knowledge that can help you narrow it down, and in some cases you won't. That's totally fine, because that's how it'll work when you estimate costs for engineering tasks.
To answer, guess a minimum and maximum for the correct answer in which you have 90% confidence. Generally speaking, the broader your range, the less confident you are. I can state with 100% confidence that each answer lies in the range [-∞, ∞]. I have about 0% confidence that every answer will be in the range [0, 1].
We're trained to go straight for a single number answer, but don't start there and just add some arbitrary error bars to get to the minimum and maximum, as that's not really a proper measure of your confidence. You may find that for some questions, the difference between your minimum and maximim is a couple of orders of magnitude, while for others it's less than a couple percentage points.
You may find it helpful to consider the minimum and maximum guesses separately, trying to avoid having one guess influence the other. You can do that by, for example, answering all the minimums first and then going back and answering all the maximums. Experiment and come up with a style that works for you.
Dang, you've run out of questions to answer. You might want to consider contributing some questions of your own over at the GitHub project.
You can refresh this page to start over with the same questions in a different order.