I wrote the review to `output/accuracy-note.md`. The figures come from `output/accuracy.json`, which the pack's script produced. The script checked that every closed month ties to the P&L and that the Budget equals the P&L budget.

- **Revenue:** The 9+3 reforecast was the most accurate. Its full-year variance was $3K, or 0.02%, and its MAPE was 0.63%.
- **Earlier versions ran high:** Every earlier version over-forecast revenue, with positive bias of 2.1% to 3.2%. Their revenue MAPEs were 3.7% for Budget, 3.3% for 3+9 and 2.9% for 6+6.
- **Same month missed twice:** 2025-07 was the worst revenue month for both Budget (15.0%) and 3+9 (9.1%). The data doesn't show why, so someone should ask what the forecasts assumed for July.
- **One big miss in 6+6:** Its worst month, 2025-09, was 15.0% off, or $151K. Most of its 2.9% MAPE comes from that one month.
- **Operating expenses:** Accuracy was better in every version. The Budget's full-year variance was only ($649), but its MAPE was 4.1%, so large monthly misses cancelled out. 9+3 had a MAPE of 0.30%.
- **Caveat:** 9+3 had only three months left to forecast, so its accuracy mostly reflects the short horizon, not better forecasting. The earlier reforecasts are the fairer test of the forecasting method.

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