How a rubric works

The structure

Each area breaks into subareas, and each subarea into leaves — the individual criteria actually assessed. A typical area carries at least forty leaves; across sixteen areas there are 781.

Each leaf is scaled 1 to 10, with the levels written out. Not adjectives — descriptions of what a person at that level actually says and does.

The mechanism test

The single rule that keeps a rubric usable across the political spectrum:

A high level requires naming a mechanism and saying who pays. It never requires reaching a particular conclusion.

A candidate who says "we should improve education" scores low regardless of whether the assessor agrees. A candidate who says "here is the intervention, here is the evidence it moves the outcome, here is what it costs, and here is who bears that cost" scores high — whichever intervention they named.

What was removed, and why

Early drafts encoded positions as levels. "More funding is the education 10." "Repealing this statute is the permitting 10." A pricing mechanism presented once as an obvious win and once as an obvious harm.

All of it was stripped and replaced with mechanism-based criteria: outcomes measured against dollars, statutory clocks, and the specific instruments involved.

The test applied afterward: can a market Democrat, an industrial-policy Republican and a deregulatory libertarian each reach a 9 here? If not, the rubric is broken.

Written before anybody is assessed

Rubrics are authored and published first. Only then is anybody read against them.

This ordering is not a courtesy — it is what makes the result a finding rather than a rationalisation. A standard written after seeing the candidate is a standard shaped by the candidate.

What the AI does with it

Reads long-form material — interviews, writing, public commitments — and marks it against the leaves. Its value is consistency: the same rubric applied identically to everyone, including to people the movement likes.

It does not decide whether the rubric is right. That is a human argument, held in public, against a published file.

Percent known

Every assessment carries how much of what we would want to know we actually have. A rubric applied to thin material produces a thin result, and the number says so rather than letting confident-looking output pass for a strong finding.