The FY2025 reforecasts got steadily more accurate, but most of that came from having fewer months left to forecast. The write-up is in `output/accuracy-note.md`, and the figures are in `output/accuracy.json`.

**Revenue**
- The 9+3 reforecast was the most accurate version. Its full-year variance was $3K, or 0.02%, and its MAPE was 0.63%. It had nine closed months and forecast only three.
- Budget missed by $362K, or 3.2%. The 3+9 reforecast missed by $177K, or 1.6%. The 6+6 reforecast missed by $158K, or 1.4%.
- Every version forecast revenue too high. Budget's bias was 3.2%, 3+9 was 2.1%, and 6+6 was 2.8%. The mid-year reforecasts did not remove that over-forecast.
- The 6+6 MAPE of 2.9% was only a little below Budget's 3.7%. Its worst month was 2025-09, forecast at $1.2M against an actual of $1.0M, a miss of 15.0%.

**Operating expenses**
- Budget's full-year variance was only ($649), but its MAPE was 4.1% and 2025-05 missed by 15.7%. Misses in opposite directions cancelled, so the full-year figure flatters the monthly forecast.
- MAPE improved with each reforecast, to 1.8% for 3+9, 0.77% for 6+6 and 0.30% for 9+3.

**Caveats**
- I can't separate better forecasting from a shorter horizon with this data. 9+3's accuracy partly reflects only three months left to forecast.
- I can't tell what the forecast assumed for 2025-09 revenue. The forecast owner would need to answer that.

The script confirmed that every closed month ties to the P&L and that the Budget version matches the P&L budget. Inputs were `forecast_versions_fy2025.csv` (96 rows) and `pnl_fy2025.csv` (120 rows).