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

- **Revenue:** The 9+3 reforecast was the most accurate, with a full-year variance of $3K, or 0.02%, and a MAPE of 0.63%. It had only 3 forecast months left, so that doesn't show the forecasting itself improved.
- **Earlier revenue versions:** Budget, 3+9 and 6+6 all over-forecast revenue. Their full-year variances were $362K, $177K and $158K, with biases of 3.2%, 2.1% and 2.8%. The reforecasts didn't fix the direction of the Budget error.
- **Largest revenue misses:** 6+6 missed 2025-09 by $151K, or 15.0%, against a MAPE of 2.9%. Budget missed 2025-07 by 15.0%. The data doesn't show what those months assumed.
- **Operating expenses:** Budget landed within ($649) for the year, or (0.03%). Its MAPE was still 4.1%, with a worst month of 2025-05 at 15.7%. Opposite misses cancelled, so the full-year figure hides the monthly error.
- **Later opex versions:** 6+6 and 9+3 were tight, with MAPEs of 0.77% and 0.30%.

The note also lists both input files with row counts and hashes.