The window is fixed in advance so that nobody chooses when to look based on how things are going.
Why it matters when the plan changes
A window chosen after the fact is chosen with knowledge of the outcome, which destroys the comparison. Setting it in advance is what makes resolution a test rather than a retrospective, and it removes the most common way a track record quietly improves itself: looking at each call at the moment it looks best. This is why a forecast is only meaningful once compared against outcomes at a defined point, the condition a scoring rule such as the Brier score depends on.
The tension is between certainty and usefulness. A long window almost guarantees a clean resolution and arrives too late to matter; a short one is decision-relevant and may resolve ambiguously because the effect has not fully appeared. The window should be set by when the answer would change a decision, not by when it will be tidiest, which is what makes it part of a forecast's definition rather than an afterthought fixed once the outcome is known.
In practice
A forecast about an unowned dependency is given a window of eight weeks, shortly after the point at which an owner would have had to be in place. At eight weeks the answer is clear and still actionable. The same forecast with a six-month window would have resolved more cleanly and told the leadership team nothing they could use.
Evidence
Forecasts are compared with actual outcomes, which requires a defined point at which the comparison is made.
Forecasting, Wikipedia (2026)Scoring rules measure how well stated probabilities matched outcomes, which presupposes the resolution point was fixed beforehand.
Brier score, Wikipedia (2026)
What it cannot tell you
A resolution window tells you when a forecast will be checked, not whether the check will be informative. A short window can end before an effect has fully appeared, producing an ambiguous result rather than a wrong one. The window says nothing about the quality of the evidence available at that point, only that a point was fixed in advance.
Questions
The horizon is how far ahead the judgement claims to apply. The window is when you check. A forecast can have a two-year horizon and a three-month window if the thing that would falsify it should become visible early, which is often the case.
From when the evidence that would settle the question becomes available, and from when the answer would still be early enough to change a decision. Setting it by the reporting calendar instead produces resolutions that are tidy, well attended and far too late to inform anything.
Only with the extension recorded and a reason given at the time, because extending a window while watching an outcome develop is how a call quietly avoids ever being wrong. An honest extension is itself a finding about how the original forecast was specified.
That is recorded as unresolved with what is still unknown, which is a legitimate result and better than waiting for clarity. An unresolved call at the window usually means the expectation was specified in terms the situation could not produce an answer to.
Yes, including the ones nobody expects to check. A window on every judgement is what makes the record complete, and a complete record including the dull calls is what allows calibration to be measured across a series rather than on the memorable ones.