The reforecasts were accurate on the full year but leaned high on revenue every time. The best version was 9+3, which missed full-year revenue by $3K, or 0.02%, with a MAPE of 0.63%. MAPE is the average size of the monthly misses. The analysis script ran cleanly, and its checks confirmed that all the input data matched the P&L.

| Revenue | 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 always forecast too high.** Bias was 3.2% in the Budget, 2.1% in 3+9 and 2.8% in 6+6. For 6+6 the bias is almost as large as its MAPE of 2.9%, so nearly every monthly miss went the same way. This is a steady upward lean, not misses that cancel out. The data doesn't show why; the monthly revenue assumptions behind each version would.
- **There were two single large misses.**
  - 6+6 overcalled 2025-09 revenue by $151K, or 15.0%.
  - Budget and 3+9 both had their worst month in 2025-07, at 15.0% and 9.1%.
- **9+3's accuracy is partly built in.** It had the lowest MAPE on both revenue and operating expenses, but it only had three months left to forecast. That alone doesn't show the forecasting got better.
- **The Budget got total operating expenses right but not the monthly timing.** The full-year miss was ($649), or (0.03%), but the MAPE was 4.1%. Monthly misses in both directions cancelled out, and the worst, 2025-05, was over by $32K, or 15.7%.

My recommendation is that planners anchor on the latest reforecast and push back on early-year revenue forecasts that look high.

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