Forecasts and ranges

A forecast projects where your citation rate is heading, as a range. The range is the honest part, and the record of previous runs is the useful part.

What you'll learn

• What a forecast is built from, and why it is a range
• How past runs are kept and scored
• What you can and cannot promise from one

What it is built from

Your own citation history: the checks already made, per surface, over time. It is not an industry model and not a benchmark. A workspace with two weeks of history gets a wider range than one with six months, which is correct rather than unfortunate.

Forecasts re-run regularly, so the range moves as evidence accumulates.

Why a range and not a number

Because a single number would be false precision. Engines change how they answer, without notice and without a changelog. A range that contains the truth is more useful than a point estimate that does not, and a range that narrows over time is a signal in itself: it means your behaviour has become more predictable.

The record of past runs

Past forecast runs are kept and can be opened individually, which lets you do the only thing that makes a forecast trustworthy: read what it said against what happened.

Two rules govern that scoring:

1. A projection is only scored where a baseline and an outcome exist on the same metric. A difference between two different measurements is not a difference.
2. Where there is no record to score against, the surface says so rather than reporting an accuracy figure nothing supports.

What you can say, and what you cannot

Fair:

• "Our current trajectory projects a range of X to Y over the next month, on our own history."
• "The range has narrowed since we started tracking, which means our performance is steadier."
• "Here is what the forecast said last month and here is what happened."

Not fair:

• "We will hit X percent." The range exists for a reason.
• "The forecast is 85 percent accurate." Only if your own record says so, on the same metric.
• Mixing a citation-rate forecast with the projected lift on a change in This week. They are different quantities with different bases, and the platform keeps them apart deliberately.

When a forecast is worth acting on

Rarely on its own. Use it as a tiebreaker:

• A widening range on a surface you were confident about is a reason to look at that surface.
• A projection that keeps being wrong in the same direction usually means something changed that you have not found yet.
• A projection is never a reason to skip the readiness work. The forecast models your behaviour, it does not fix it.

Try this in Outercite

Open Intelligence and find the forecast record. Read the oldest run against what your rate actually did.

Related