Levels are written as short descriptions, which is what lets the model place content without a separate rubric hidden in your prompt. The docs’ bug severity example uses three: “Cosmetic; no impact to functionality”, “Broken or degraded feature, but workaround exists”, and “Blocking issue; no workaround exists”.
Given the state “The export button crashes the settings page in Safari. It works in Chrome, but a few of our customers only use Safari”, the answer is a score of 1.3 with probabilities of 0.0, 0.7 and 0.3 on levels 0, 1 and 2, at confidence 0.54. The arithmetic is plain: (0 x 0.0) + (1 x 0.70) + (2 x 0.30) = 1.30.
Rounding that to 1 throws away the useful part. The 0.3 on level 2 says roughly three tickets in ten that read like this one really are blocking, which is the number a paging rule should key off. The confidence of 0.54 says the same thing from the other side: the distribution is split, so this is not a case to auto-resolve.
The output is a number, so several Scores combine with weights your code controls. The passage reranking recipe shows that.