Snapshots recorded
—
Months with a settled outcome
—
A month can only be scored once it has closed
Median revision between runs
—
How much a forecast moves day to day
Forecast evolution
How the prediction for a given month changed as that month approached.
Accuracy once the month closed
For each completed month, the forecast that was on the site at the time, and what was actually published.
No month has both a recorded forecast and a settled outcome yet. This table fills in as months close — the first entries appear once a month forecast here has ended.
| Month | Published | Forecast at the time | Error | Lead time | Convergence |
|---|
Model stability
How much each model revises its own prediction between runs, on data that has not changed much. A model that swings wildly day to day is telling you something about its variance, not about the world.
| Model | Mean revision | Largest revision | Stability |
|---|
Data revision
CVE records keep arriving after a month ends, so the count for a month depends on when you looked. Each run records what every recent month looked like at that moment, which is the only way to measure the effect — it cannot be reconstructed after the fact.