What we do
Three steps, in this order, every time. The order is the whole point: once a forecast is registered, there is nowhere left to hide.
First
Pick a question with a date on it
Questions where a decision turns on a number that does not exist yet, and where an official body has already put its own number on the record. Those are the ones worth forecasting, because there is something to be measured against.
Then
Register it before the outcome
The claim, the benchmark and its value on the day, the data used, the dates it will be marked and the rule it will be scored by — committed to a public repository while the answer is still unknown.
Afterwards
Publish the mark either way
Scored on the fixed date against the number it set out to beat. Hits and misses alike. Nothing is edited, nothing is quietly dropped, and the misses stay on the record.
The standard
A forecast you cannot check is an opinion
- The benchmark is named, with its vintage, before the forecast is made
- The marking dates are fixed in advance, so a miss cannot become a longer horizon
- Every claim on this site links to a registered entry, or it is not made
The series
Four lines of work, one method. Anything may enter if it can be forecast in advance and scored on a fixed date; anything may leave once it has been scored honestly.
Which are next?
Distress screens of public bodies and regulated companies, published as a pre-registered shortlist and scored on what happens.
Breaking points
Capacity against a threshold, with a dated forecast of when the threshold is crossed.
Against consensus
National series forecast ahead of the official and market numbers, marked on release.
A generation from now
Town-level demography, housing and cohort forecasts to 2046 and 2066, with the backtest published alongside.
What has been registered so far, and how each entry stands, is on the ledger. The rules it runs on are set out in the method.