You give 2 to 10 ordered levels. You get a probability-weighted mean of the level numbers.
Each level is judged against the state on its own. It never sees its own number or its neighbours. So levels have to describe recognisable situations — “Broken or degraded feature, but workaround exists” — never degrees like “moderately severe”, and never bare numbers.
The returned score is arithmetic over that distribution, so a 1.0 on a three-level Score can mean a confident middle or an even split between the ends. Read the probabilities, not just the mean.
