The reforecasts were accurate on the full year but kept forecasting revenue too high. The last reforecast, 9+3, was the most accurate on both revenue and opex. Its full-year revenue variance was $3K, or 0.02%, and its MAPE (average monthly miss) was 0.63%.

The analysis script's checks passed: every closed month matches the P&L actual, and the Budget version matches the P&L budget.

- **Revenue was forecast too high all year.** Budget, 3+9 and 6+6 all came in above actual, with bias of 3.2%, 2.1% and 2.8%. The data doesn't show which assumption caused this.
- **6+6 had one large miss.** September 2025 came in $151K, or 15.0%, below its forecast, while its average monthly miss was 2.9%. It's worth asking what 6+6 assumed for that month.
- **Budget opex is closer than it looks.** It was off by only ($649), or (0.03%), for the year, but its average monthly miss was 4.1%. Monthly misses in both directions cancelled out, including $32K, or 15.7%, in May 2025.
- **9+3's lead mostly reflects timing.** It already had 9 actual months and only 3 left to forecast, so its accuracy doesn't show the forecasting itself got better. For mid-year reforecasts, 6+6 is the more useful benchmark, after correcting for its upward revenue bias.

The note has revenue and opex tables for all four versions, and the data notes list the input files with their row counts and file hashes.

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