I've written the review to `output/accuracy-note.md`, with the underlying figures in `output/accuracy.json`. Short answer: the reforecasts got more accurate as the year went on, but mostly because each one had fewer months left to predict. Every version before 9+3 forecast revenue too high, and the reforecasts never fixed that. The calculation checks passed: the actual months in each reforecast and the Budget both match the P&L.

**Revenue** (a positive variance means the forecast was too high)

| Version | Months already actual | FY variance | Average monthly miss (MAPE) | Bias |
| --- | --- | --- | --- | --- |
| Budget | 0 | $362K, or 3.2% | 3.7% | 3.2% |
| 3+9 | 3 | $177K, or 1.6% | 3.3% | 2.1% |
| 6+6 | 6 | $158K, or 1.4% | 2.9% | 2.8% |
| 9+3 | 9 | $3K, or 0.02% | 0.63% | 0.10% |

- **9+3 was the most accurate version** for both revenue and opex. It only had 3 months left to forecast, so that doesn't prove the forecasting itself improved.
- **Revenue ran high all year.** The 6+6 bias of 2.8% is almost the same as its 2.9% average miss, so nearly every month it forecast came in too high.
- **Two months account for the biggest revenue misses.** September 2025 was $151K, or 15.0%, too high in the 6+6. July 2025 was the worst month for both the Budget, at $132K or 15.0%, and the 3+9. The data doesn't show why; deal-level timing data would answer it.
- **Budget opex looks right for the year but not month by month.** It finished at ($1K), or (0.03%), yet its average monthly miss was 4.1%. Misses cancelled out, and May 2025 alone came in $32K, or 15.7%, over.

My recommendation is to base opex planning on the latest reforecast; from the 6+6 onwards, opex forecasts stayed close. For revenue, treat early-year forecasts as too optimistic until the July and September misses are explained.

All amounts are rounded to the nearest $1K on their own, so an opex variance won't always equal the forecast minus the actual exactly. The note says so.