What is real here, and what is not.
Predictions come from the ITU's own reference implementation and are verified
numerically identical to a native build of it. Distances, bearings, MUF and the
diurnal behaviour all check out against reality.
The engine's reliability output is a known open defect — it reads zero
for every circuit tested and ignores the required-SNR input — so nothing here
displays it. What you see is SNR margin, which is verified responsive.
The A-score described in the design docs does not exist yet. It is meant to be
a calibrated probability, and calibrating it depends on resolving that defect first.
Reach maps are computed for whatever transmitter you pick, at 100 W into an isotropic
antenna. Your own power and antenna are applied on top as an exact offset. Antenna
figures are peak gain applied flat, so beams read optimistic on paths whose
take-off angle they do not favour. Coarse passes paint each sample across the block it
stands for, so an early frame is a sketch until the finer passes land.
Solar activity is live: effective sunspot number from KC2G, refreshed hourly
by our own service so your browser never contacts theirs. Every page states which
source it used and how old it is, and falls back — to the snapshot published with
the build, then to a clearly-labelled placeholder — rather than quietly reverting
to a default sun.
Only the sunspot number is live in the sense that matters — it is the one
input P.533 actually takes. Kp, solar flux and NOAA's alerts are fetched and delivered
to your browser, but nothing reads them yet: they are neither displayed nor applied to
a prediction. Making a geomagnetic storm change the map means a penalty layer on top of
the engine, scaled by geomagnetic latitude, which is the A-score's job and does not
exist. Treat every prediction here as assuming quiet geomagnetic conditions.