I've written the review to `output/accuracy-note.md`. The numbers behind it are in `output/accuracy.json`. Every closed month matched the P&L actuals, and the Budget version matched the P&L budget.

**The short answer:** the 9+3 reforecast was the most accurate version. It missed full-year revenue by $3K, or 0.02%, with a MAPE of 0.63%. That accuracy comes mostly from the calendar, not from better forecasting: 9+3 had 9 months already closed and only 3 left to forecast.

**Revenue was too high in every version before 9+3:**

| Version | 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% |

- **Persistent overstatement:** 6+6 still forecast revenue 2.8% too high with six months to go, so revenue forecasting didn't get meaningfully better. The data doesn't show why. The pipeline or bookings assumptions behind each version would answer that.
- **Biggest single misses:** Budget and 3+9 both missed most in July 2025, by $132K and $80K. 6+6's worst month was September 2025, at $151K or 15.0%, against a MAPE of 2.9%. That's one large miss, not a trend, and it's worth asking what each version assumed for those months.

**Operating expense forecasting did improve:**

| Version | MAPE |
| --- | --- |
| Budget | 4.1% |
| 3+9 | 1.8% |
| 6+6 | 0.77% |
| 9+3 | 0.30% |

The Budget's near-zero full-year opex miss of ($649) is misleading. Its monthly misses ran as high as $32K, or 15.7%, in May 2025, and they cancelled out over the year.

**What to rely on:** the opex reforecasts can be trusted. For revenue, plan from the current run-rate rather than the forward months of early reforecasts, which ran high.