I wrote the review to `output/accuracy-note.md`. The script's figures are in `output/accuracy.json`. The closed months tie to the P&L, and the Budget version matches the P&L budget.

- **Revenue:** 9+3 was the most accurate version. Its full-year variance was $3K, or 0.02%, and its MAPE was 0.63%.
- **Revenue bias:** Every version forecast revenue too high. Budget missed by $362K, or 3.2%. 3+9 missed by $177K and 6+6 by $158K.
- **Worst revenue months:** Budget's worst month was 2025-07, a variance of $132K, or 15.0%. 6+6's was 2025-09, a variance of $151K, or 15.0%. The data doesn't show what the forecasts assumed for those months.
- **Operating expenses:** Budget ended within ($649), or (0.03%), of actual. Its monthly MAPE was still 4.1% because misses offset each other. 9+3 was the tightest, at 0.30%.
- **9+3 caveat:** It had only 3 forecast months left, so its low MAPE shows less runway rather than better forecasting. Use it for the year-end landing, not as proof the process improved.

The note has the full tables for both lines.