Rubric, defined

A rubric is the list of things a tool claims to judge, and how much each one counts toward the number you get back. Every rating tool has one, even the ones that never say so - a model cannot produce a score without some internal notion of what it is looking for.

The difference between tools is not whether a rubric exists, it is whether you ever get to see it.

A small example

Say a tool judges three things: proportion, symmetry, presentation. A published rubric would tell you the weight on each - proportion 50%, symmetry 30%, presentation 20% - and let you work out, roughly, why a submission landed where it did. An unpublished rubric produces the identical number with none of that visible. You get the output of the weighting without the weighting itself, which means you cannot tell whether a change in your result came from the thing you think changed or from something else entirely.

Rubric versus total

A total is what a rubric produces once it has been applied and collapsed. The rubric is the recipe; the total is the dish. You can taste a dish and guess at the recipe, but you cannot audit it - you cannot change one ingredient, retaste, and confirm the change did what you expected, because you never knew the proportions to begin with.

That is the practical cost of an unpublished rubric. It is not that the tool is lying about what it measures. It is that you have no way to attribute a change in your score to a change in your submission, because the function connecting the two was never shown to you.

What this is not

This is not a claim that a published rubric is a fair one, or that its weights are the right weights - who decided what a 9 looks like is a different question, and every rubric encodes somebody's judgement call rather than a fact discovered in nature - the OECD and European Commission Joint Research Centre handbook on composite indicators (2008) says plainly that, whatever the method, "weights are essentially value judgements." It is also not a claim that a rubric, once published, stays put: tools revise them, and a rubric that changes silently makes every earlier score uninterpretable. Published or not is only the first fact worth establishing about any tool's rubric. It happens to be the fact that decides whether the other facts are even checkable, which is why it comes first.

Where the word stops applying

A rubric is a property of an automated tool, weighing axes it computes itself. It is not the right word for what a human reviewer does with a submission - a person forms an impression rather than running a weighted list, and calling that a rubric borrows precision it does not have. It is also not the word for a tape measurement: a length in centimetres is not weighed against anything, it is read off directly. And a rubric sits above the model, not inside it - what the underlying vision model is actually doing is a lower layer, and a tool can publish its rubric without ever touching that.

For the fuller comparison of how rubrics actually differ across tools in practice, rubrics compared end to end covers axis count, versioning, and whether a breakdown is real or decorative. Rate Cock is one instance of a tool that publishes its rubric outright, with the weights visible on the same page as the score - the exception worth noticing precisely because it is rare.