How the Forecast Is Synthesised

Live measured readings for your location carry first weight. Matched historical years carry second weight. The synthesised projection carries third weight and only where measurement cannot reach.

That order is the whole design. A forecast that leads with a model and checks reality afterwards drifts; this one is pinned to observation and lets the model fill the gap.

History back to 1950 is searched for years whose pattern matches the current one on two axes: strength and timing. Matching on strength alone finds hot years that peaked at the wrong time; matching on timing alone finds years that peaked on cue but nowhere near the same magnitude. Both axes must agree before a year is allowed to influence the projection.

The peak date is the single day the projection collapses onto. The held window is the span the method was told to consider — it is shown so you can see the answer was not cherry-picked from a wider range.

The confidence figure falls as the projection stretches away from the last real reading. A high figure near the observed record means the answer is anchored; the same figure far out would not be, which is why the damping exists.

The engine forecasts a temperature peak within a window. It is not a severe-weather warning service and must not be relied on for one. Where the external model host is unavailable, the local engine answers alone and the page says so rather than pretending to a precision it has not got.