The reforecasts got more accurate as the year went on, but the revenue forecasts ran too high until the last one. The 9+3 reforecast was the most accurate: revenue was off by $3K for the full year, or 0.02%, with a typical monthly miss (MAPE) of 0.63%. The script checked that every closed month and the Budget match the P&L, and both checks passed.

| Revenue | Full-year variance | MAPE | Bias |
| --- | --- | --- | --- |
| Budget | $362K, or 3.2% | 3.7% | 3.2% |
| 3+9 | $177K, or 1.6% | 3.3% | 2.1% |
| 6+6 | $158K, or 1.4% | 2.9% | 2.8% |
| 9+3 | $3K, or 0.02% | 0.63% | 0.10% |

- **Revenue was forecast too high through mid-year.** Budget, 3+9 and 6+6 all over-forecast, with bias of 3.2%, 2.1% and 2.8%. The data doesn't show why; customer- or product-level forecasts would answer that.
- **6+6 had one large miss.** September 2025 (2025-09) was over by $151K, or 15.0%, against a MAPE of 2.9%. That is one month, not a trend. Budget and 3+9 had their worst revenue month in July 2025 (2025-07).
- **Budget opex looks exact for the year but wasn't month by month.** The full-year variance was ($649), or (0.03%), but MAPE was 4.1%. May 2025 (2025-05) was over by $32K, or 15.7%. Monthly misses in both directions cancelled out.
- **9+3 had only 3 months left to forecast.** Part of its accuracy comes from that, not just from better forecasting. Even so, it's the right one to anchor FY2026 run-rate planning on. The repeated revenue over-forecast in earlier versions is the process problem to fix.

Files are in `output/`:
- `accuracy-note.md`
- `accuracy.json`