The reforecasts got the full year closer each time, but every version before 9+3 forecast revenue too high, by roughly 2–3%. All four versions tie to the P&L.

**Revenue**
- **9+3 was the most accurate.** It finished $3K, or 0.02%, off the full year, with a typical monthly miss (MAPE) of 0.63%. But it only had three months left to forecast. That shows there was less left to get wrong, not that the forecasting got better.
- **Budget, 3+9 and 6+6 all ran high on revenue,** with bias of 3.2%, 2.1% and 2.8%. The over-forecast survived two reforecasts, so the revenue assumptions weren't reset after the early misses. The data doesn't show which assumption caused it; customer- or volume-level forecasts would answer that.
- **6+6's biggest miss was September 2025:** $151K too high, or 15.0%, against a MAPE of 2.9%. That one month is worth asking the planners about. The Budget had a miss of the same size, 15.0% or $132K, in July 2025.

**Operating expenses**
- **The Budget got the year right but the months wrong.** It finished within ($649), or (0.03%), of the full year, but its MAPE was 4.1%: monthly misses in both directions cancelled out. Its worst month, May 2025, was $32K too high, or 15.7%.
- **The reforecasts tracked opex closely.** MAPE fell to 1.8% at 3+9, 0.77% at 6+6 and 0.30% at 9+3.

**Recommendation:** use 9+3 for the revenue run-rate. Treat earlier-in-year versions as likely to run high on revenue until the bias is explained.

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