Done. I've analyzed FY2025 forecast accuracy using four reforecast versions and written it up in `output/accuracy-note.md`.

**Key takeaway**: The 9+3 reforecast (9 months closed, 3 months forecasted) nailed full-year revenue, missing by just $3K or 0.02% with a MAPE of 0.63%. The initial Budget version was off by $362K (3.2%), but accuracy improved dramatically as actuals replaced assumptions. July was the Budget's worst revenue miss (15.0% low), and September tripped up the 6+6 reforecast (15.0% high), both flagging seasonal surprises the mid-year forecasts didn't anticipate. Operating expenses proved far easier to forecast, with the 9+3 version off by less than $1K annually.