The set of criteria your work is judged against. Written in red, and written before you started.
A screenplay is two layers. The dialogue, in black, is what the two hackers type at each other. The rubrics — the slug line, the action, the parentheticals — are in red, and they are the directions: where we are, what happens, what the machine does. Watch them light up in sequence and run the terminal on the right — the login window boots, a trace opens, the boot is armed, the connection is terminated — because the red said so. Press Read the red and the dialogue falls to a murmur, leaving the directions alone: a shot list, an operations log, a rubric. On vellum or on a CRT, the red was never decoration.
Rubric comes from Latin rubrica, the red ochre — rubrica terra, red chalk — from ruber, “red.” The pigment names the practice. Long before rubric meant a set of rules, it meant the specific red you reached for when something needed to stand apart from the ordinary black.
In medieval manuscripts the body was copied in black ink, but headings, chapter marks, and directions were added afterward in red — rubricated — so the reader could find the structure at a glance. The rubric was, literally, the part that told you what to do and where you were.
In a missal, the prayers are in black and the rubrics — stand, kneel, make the sign — are in red: the directions, not the content. Law inherited the sense too, the rubric as the heading and rule of a statute. To follow the rubric was to follow the instruction printed in advance.
Modern education turns the word into a scoring matrix: criteria down one side, performance levels across the top, each cell describing what a given mark looks like. The point is that the standard is published before the work is done. You are measured against it, not consulted about it.
Rubrics spread through standardized assessment precisely because they fix the judgement ahead of time, making grading consistent and, in theory, fair. The trade is honest and a little chilling: the criteria preceded your submission, and they will not be moved to accommodate it.
LLM-as-judge pipelines score model outputs against explicit rubrics; eval harnesses encode criteria as columns; “definition of done” is a rubric by another name. The word exploded back into use because we suddenly needed, at scale, a fixed standard to measure generated work against.
From the red-chalk heading in a codex to the scoring function in a test suite, the idea has never changed. A rubric is the judgement written down before the thing being judged exists. You can do excellent work against it. You cannot argue with it. Appeals accepted: zero.