Temper scores words, not people. It reads only a public figure's own public posts and measures tone: how much of a post is shouting, attack language, threat language, exclamation and length. It says nothing about anyone's health, mind or motives, and it is not a clinical measure of anything.
Per post. Intensity from 0 to 1: 30% capitalised shouting, 30% attack lexicon, 15% threat lexicon, 15% exclamation marks, 10% length. Bare links and very short reposts count half.
Heat (0 to 100). Recent posts, weighted by intensity, fade over a few hours. A long silence cools the number.
Tangle (0 to 100). How crowded and uneven the last two hours were: many posts, wildly different intensities.
The tape. Every number is a pure function of the stored posts and this rubric, so any past minute can be recomputed identically. If the rubric changes, history is re-scored under the new version and the version is shown.
Limits. A word list misses sarcasm and context. Tap the blob to see the last five posts and their scores, and judge for yourself. Corrections: rubric@temperlive.com.