The 9+3 reforecast was the most accurate version for both revenue and operating expenses. The note is `output/accuracy-note.md`, and the figures are in `output/accuracy.json`. The script ran clean and the closed months tie to the P&L.

**Revenue** (forecast less actual, so positive means too high):

| Version | FY variance | MAPE |
| --- | --- | --- |
| Budget | $362K, or 3.2% | 3.7% |
| 3+9 | $177K, or 1.6% | 3.3% |
| 6+6 | $158K, or 1.4% | 2.9% |
| 9+3 | $3K, or 0.02% | 0.63% |

- **Revenue was over-forecast in every version up to 6+6.** Budget, 3+9 and 6+6 all ran high, with biases of 3.2%, 2.1% and 2.8%.
- **Two months account for the biggest misses.** Budget missed July by $132K, or 15.0%. 6+6 missed September by $151K, or 15.0%. The data doesn't show why; the assumptions behind those two forecasts would answer it.
- **Operating expenses were much tighter.** MAPE was 1.8% for 3+9, 0.77% for 6+6 and 0.30% for 9+3. Budget landed within ($649) for the year but had a 4.1% MAPE, so its monthly misses offset each other.
- **9+3's accuracy mostly reflects how little was left to forecast.** It had nine closed months and only three to forecast, so it doesn't show better forecasting. Use it for the year-end landing, not as proof of forecast skill.

The note also has the full tables for both lines and the data notes with each input file's row count and hash.