The reforecasts only became accurate late in the year, and revenue was the weak line. I've written the full note to `output/accuracy-note.md` and the script figures to `output/accuracy.json`.

- **Best revenue version:** 9+3 (nine closed months, three forecast) finished $3K, or 0.02%, from actual, with a MAPE of 0.63%.
- **Earlier revenue versions ran high.** Budget was off by $362K, or 3.2%. 3+9 and 6+6 were still off by $177K and $158K, or 1.6% and 1.4%. Their biases of 2.1% and 2.8% show they kept overstating revenue. 6+6 missed September by $151K, or 15.0%.
- **Operating expenses were closer.** Budget's full-year total was within ($649) of actual. Its MAPE of 4.1% shows large monthly misses that cancelled out. MAPE improved with each reforecast, to 0.30% by 9+3.
- **Caveat:** 9+3 had only three months left to forecast. Its accuracy mostly reflects the short horizon, not better forecasting. I'd anchor on it for the year-end landing but not treat the earlier revenue versions as reliable.
- **Open question:** The data doesn't show why revenue stayed optimistic. The July and September assumptions would answer that.

Every figure in the note is copied from the script's output. The script confirmed the closed months tie to the P&L.