Patent draft
The forecasting method at claim level: a synthesised wave fitted to measured readings, damped across a held window, collapsed to one peak with a confidence figure.
Raw observation arrays are read for the target location and fitted, by least squares, to the trend component of a synthesised wave, so the projection is anchored to what actually happened rather than extrapolated from nothing.
Additional measured signals — surface pressure, dew-point spread and wind — are normalised and folded in as phased contributions with fixed relative weight.
Variance is damped in proportion to how far the projection has stretched from the last real observation, so confidence falls with distance instead of staying flat.
A held window is applied and the projection is collapsed to a single peak date, peak value and confidence figure.
Matched analog years are selected on strength and timing and cross-weighted against the projection.
The coefficients, the phase constants, the weighting ratios and the damping constant are the protected part of the method. They appear in no interface response, no client bundle, no log line, no document and no test fixture. The shape of the method is disclosed; the numbers that make it accurate are not.
The method is not a re-presentation of a public forecast. Public data is consumed as raw observation input only. The output is produced by the local engine and remains correct when every external model host is unreachable.
Published summary of an attorney-review draft or engineering record — not a filing, not a patentability opinion, not a safety certification. Engine parameters, signing keys and customer records never appear in any published paper.