The revenue reforecasts were accurate in total but forecast too high in every version before the last one. The latest version, 9+3, was the most accurate, with a full-year revenue variance of $3K, or 0.02%, and a MAPE of 0.63%. MAPE is the average size of a month's miss.

The accuracy script ran without errors. It confirmed that every closed month matches the P&L actual and that the Budget version matches the P&L budget.

**Revenue**
- **Every version before 9+3 forecast too high.** Bias was 3.2% for Budget, 2.1% for 3+9 and 2.8% for 6+6. Positive bias means the forecast was above the actual on balance.
- **Budget and 3+9 both missed most in 2025-07.** Budget was $132K too high, or 15.0%, and 3+9 was $80K too high, or 9.1%. Since two versions overshot the same month, the July assumption is worth checking.
- **6+6's worst month was 2025-09.** It was $151K too high, or 15.0%, against a MAPE of 2.9%. That is one large miss in one month, not a trend. Someone should check what 6+6 assumed for September.

**Operating expenses**
- **Budget's small full-year gap hides large monthly misses.** It was off by ($649), or (0.03%), for the year, but its MAPE was 4.1% because the monthly misses cancelled out. Its worst month was 2025-05, $32K too high, or 15.7%.
- **The reforecasts came steadily closer.** MAPE was 1.8% for 3+9, 0.77% for 6+6 and 0.30% for 9+3.

**Should you trust the latest version?** 9+3 was the most accurate version for both revenue and opex. However, it had only 3 months left to forecast, and its 9 closed months were already actuals. So its lead mostly reflects timing; it doesn't prove the forecasting got better. Anchor plans on the latest version, but expect earlier revenue forecasts to run high.

Files are in `output/`:
- `accuracy-note.md`
- `accuracy.json`