What the score actually measures.
One conditions number from 0 to 100 per hour, combining cloud, transparency, seeing, moonlight and astronomical darkness. Site darkness is a separate peer measure, because a clear city night and a clear dark-site night are useful for different targets.
Build an hour.
Move any condition and watch it reduce the same 100-point starting sky. Cloud, moonlight and darkness can close the gate; the broader forecast deliberately softens the less-certain transparency and seeing estimates.
Excellent. Go out.
A useful window survives every condition in the chain.
Why it multiplies
The score is multiplicative, not an average. That is a deliberate choice: averaging lets a good component hide a fatal one, and in observing the fatal component is usually decisive. A perfect sky under total cloud is not a middling night, it is no night.
How each factor enters
Cloud cover sets the ceiling. Transparency and seeing then modulate how good that clear sky actually is, moonlight scales it down, and darkness scales it again. The general score deliberately softens the less-certain transparency and seeing estimates: transparency contributes a 0.60–1.00 factor and seeing a 0.75–1.00 factor. Cloud, moonlight and darkness can still close the gate. Target modes reweight the same inputs more strongly when they matter to what you plan to observe.
Inspect it — and know its limits
You can see the arithmetic for any hour. The microscope on every forecast page shows the cascade: it starts at 100 and attributes the points lost at each step, in the same order the code applies them. Nothing is hidden behind a proprietary index.
The score answers one question — are tonight’s conditions better here now? Modelled artificial skyglow sits beside it rather than being multiplied in. It changes target suitability and the displayed visual-limit range; it also conditions moon and haze only where those effects physically interact with the existing sky background. Local lamps, topography, access and recent LED conversions remain outside the model.