Geographic forecast monitoring

Closed-month volunteer results help us study regional and seasonal differences. Records with missing or contradictory best-day settings are held for review. A poor forecast alone never excludes a reading.

Each group needs at least 10 contributing devices and 50 device-days. Areas use broad 30° latitude by 60° longitude bands. If an area is too small, that month has a worldwide summary only. Devices receive equal weight. Summaries are frozen and retained after individual data is deleted.

These are retrospectively shared daily histories, not verified forecasts issued in advance. They do not yet establish hourly or seven-day performance. Zero-generation days cannot have an error percentage and are omitted from these summaries. This separate measure does not replace the main accuracy score.

Positive bias means overprediction. Device counts are pseudonymous contributors, not verified households. Setup flags indicate uncertainty, not that a person’s meter reading is wrong.