A question with a finish line.
A forecast becomes useful when its future answer can be checked without rewriting the original question.
Curiosity needs a little structure
“Will this market do well?” is a perfectly reasonable beginning for a conversation. It is a difficult beginning for a forecast. The words leave several things unresolved: which market, what counts as doing well, compared with what, and over which period. Two people could agree on every observed fact and still give different answers to the original question.
An answerable forecast narrows those choices before you know the result. That does not make the world simpler; it makes your statement more precise. You are choosing a small, observable part of a larger uncertainty and giving yourself a way to return to it. A good finish line helps you learn because it prevents the meaning of success from drifting after the outcome arrives.
Name the observation you will use
Start by identifying something you can actually check. For a sports question, that might be the official results of a named team’s next ten completed league matches. For a market question, it could be the percentage change in a specific reference price between two recorded timestamps. For an event question, it might be an official publication from a named organization.
Avoid relying on a source that does not publish the measurement you need. Before saving the forecast, visit the source, find the relevant field, and note any revision policy that might matter. You do not have to eliminate every edge case, but you should be able to describe the ordinary path from evidence to answer without adding a new rule later.
Make the alternatives cover the whole space
Forkcast uses three possible paths. The useful challenge is to make them mutually exclusive and exhaustive: only one path should happen, and every eligible result should fit somewhere. “Price rises,” “price falls,” and “something unexpected” may sound expressive, but the last category does not identify a measurable outcome. A flat result is also missing from the first two.
For a hypothetical relative-performance question, you could define one path as a difference greater than five percentage points, a middle path as a difference between negative five and positive five inclusive, and a third as a difference below negative five. The exact thresholds are your choice. What matters is that the boundaries have no overlap and leave no ordinary result unassigned.
Give time its own sentence
A review date is useful, but it is not always the same as a resolution deadline. A forecast about a team’s next ten matches may remain unresolved on the planned review date if fixtures are postponed. A forecast about a publication by a fixed date may resolve at that deadline even if the publication appears later. The difference belongs in the rule.
Write down the measurement period and the timezone when relevant. Say whether you use the event time, the publication time or the time a source confirms a result. If the forecast depends on two market prices, record comparable timestamps and a consistent source. Small timing choices can produce large interpretive differences, especially around announcements and market closes.
Try to disagree with your own wording
Before allocating probability, imagine another careful reader resolving the forecast. Could they choose a different path without ignoring your instructions? If so, identify the phrase that gives them room. Words such as “major,” “successful,” “significant” and “soon” often carry useful conversational meaning while remaining too flexible for a scoring rule.
You can preserve the larger curiosity in the evidence notes. The formal question does not have to capture every aspect of the story. It should capture one claim clearly enough to evaluate. That separation gives you room for nuance without asking a later outcome to settle a question you never actually defined.
Leave a record your future self can trust
A useful forecast contains a concise question, a source, a time boundary, three defined alternatives and an evidence checkpoint. The checkpoint names what you plan to inspect next; it does not need to predict what you will find. Saving these pieces together gives the probabilities a context that a bare percentage cannot carry.
When you return, evaluate the result against the original wording. If the wording turns out to be defective, record that as a lesson about question design rather than quietly repairing the forecast after the fact. An honest unresolved question can teach more than a neatly scored answer to a moving target. The finish line is there to support learning, not to manufacture certainty.