Judgment rules and Vale rules

A Vale rule is a pattern. It can find a banned word, a filler frame, a sentence over twenty words, a paragraph over six sentences, a hedge stacked on a hedge. It cannot tell whether a reply led with the answer, whether a term was defined before it was used, or whether the last sentence of a paragraph is a fact or a flourish. Those are judgments.

The plugin keeps both kinds in one list of bullets. A bullet with Vale rules behind it is enforced twice: the text tells the model, and the linter checks. A bullet with none, a judgment rule, is enforced once, by the model’s reading of it. The vale field on each bullet is the line between the two, and it is why the field matters beyond documentation: the reminder drops a bullet when its rules are off, and the summary knows whether a rejection tagged with a bullet’s id was something Vale could have caught.

The two kinds fail differently. When a Vale-backed rule is broken, the hooks see it: the file check, the publish check, the reply report. When a judgment rule is broken, nothing sees it but you. That is what /just-say-so:restate is for. It records the misses the hooks cannot, and over time the records show which judgment rules need rewording, which need a short text so they reach the reminder, and which could be turned into Vale rules after all, because a package in the registry has one.

The shipped rules come from the ASD-STE100 standard for Simplified Technical English, minus its 900-word dictionary. Four of them ban things a pattern can match and are generated as the Vale rules of the JustSaySo package. The rest are judgment rules, and the plugin’s value is that they stay in force across a long session instead of fading after the first prompt.