Every time the forecast changes, a dated snapshot is kept. When a season is shown and assessed, we compare what we projected with what we then assessed, and publish the result here, good or bad.
For each season that has moved from projection to assessed: the share of details whose assessed stage is the same as the last projected stage, and the share within one stage (for example projected Rising, assessed Peaking).
No back-test result yet. Records start on 2026-10-26. Since then no season has moved from projection to assessed, so there is nothing honest to score. The first test will be AW27, once its runway shows have been reviewed: its last projected stages will be compared, detail by detail, with the assessed ones and published here.
Each row is a published version of the forecast. "Changes" names the seasons in which any detail's stage moved compared with the previous version.
| Date | Projected seasons | Changes since the previous version |
|---|---|---|
| 2026-10-26 | AW27, AW28, SS28 | — |
| 2026-10-31 | AW27, AW28, SS28 | new details added |
| 2026-11-07 | AW27, AW28, SS28 | new details added |
| 2026-11-15 | AW27, AW28, SS28 | no stage changes |
Assessed stages are Forecut's own reading of published coverage, so a back-test measures how stable our projections are against our later assessment, not against sales. Read it alongside the evidence grades.