Best Practices For Writing Questions

Writing questions can be surprisingly challenging. Many people have a hard time with it at first.

Cover all likely possibilities

Make sure to cover all of the conditions that could happen.

Bad: "How much will project X cost?" Better: "How much will project X cost? If it does not happen or is cancelled, it counts as 0."

Use clear terminology or use standardized dictionaries

Lots of colloquial language is not precise enough for great forecasts. For example, the question "When will project X be complete?", may leave the definition of both "project X" and "complete" vague. People may have different impressions of what things are and aren't part of "Project X", and different impressions of what "complete" means.

Bad: "When will project X be complete?" Better: "When will Project X be complete? This is according to our Asana schematic of Project X, and our Kanban definition of complete."

Make sure you have a way of resolving a question

When you write a question on Foretold, you are responsible for eventually resolving that question. If it's not crystal clear how you would actually get or determine the number, either say that you will judge it yourself, or better yet, reference a specific authority you will use for question resolution.

Bad: "How much will the current trade deal hurt the economy?" Better: "Authority X routinely estimates the net benefit of trade deals. If they do so on this deal, what number will they give?" Better(alternative): "In Jan 2025, I will personally estimate the benefit of trade deal X. What number will I give it?"

Further tips

There are subtle ways in which your question can break even if you get the above right.

For example, measuring the productivity of your team in terms of number of lines of code written written will fail to capture things like code readability and performance, which might be what you really care about.

Just because a question measures something does not mean that it measures what you care about.

You can find a more in-depth discussion of how to avoid these failure modes in this page from the AI Forecasting Dictionary.

Last updated