The short answer for the CFO: the opex outlooks can be trusted, but the revenue outlooks ran too high until the 9+3 reforecast.

The scoring script ran cleanly. Each version's actual months match the P&L, and the Budget version matches the P&L budget. Below, "MAPE" is the average monthly miss over the months a version had left to forecast. "Bias" says which way it missed on balance; positive means too high.

**Revenue** (actual for the year: $11,238K)

| Version | 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 three times in a row.** Budget, 3+9 and 6+6 all had positive bias, so this is a steady lean, not random noise. The data doesn't show why; the revenue assumptions behind each version would.
- **Single big misses.** 6+6 missed September by $151K, or 15.0%, and Budget missed July by $132K, or 15.0%. It's worth asking what each version assumed for those months.
- **The 9+3 accuracy is partly flattering.** 9+3 scored best, but it only had 3 months left to forecast, all close to when it was made. Planners should use the latest reforecast. That doesn't prove the forecasting method itself got better.
- **Opex accuracy was good.** Budget's opex total was only ($1K) off for the year. Month by month, though, its MAPE was 4.1% and it missed May by $32K, or 15.7%; the misses cancelled out over the year. From 3+9 onward, opex MAPE was 1.8% or lower.

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