Skip to content

Define a rubric

A rubric is the standard a piece of work is scored against before a person is asked to look at it. Governance → Rubrics in the settings sidebar is where they live. It needs the manage_org_roles permission.

Three things:

  • Rows, each a criterion in one sentence, with a weight. A weight of 3 matters more than a weight of 2.
  • A passing score, the total a piece of work has to reach.
  • A version, which is how a rubric changes without rewriting history.

Alpenglow’s content rubric, Content bookmarkability, asks whether a draft is worth a reader saving. Its five rows carry weights of 3, 3, 2, 2, and 2, and the bar is 8 out of a possible 12. So a draft can miss one thing that matters and still pass, and cannot miss two.

Writing the rows is the whole job. “Well written” is not a criterion anyone can score; “carries one idea a reader could repeat afterwards” is.

  1. Click + New above the list.
  2. Name it after the judgement, not the artifact: “Route builder beta gate” says what it decides.
  3. Add a row per criterion, each with an id, a description, and a weight.
  4. Set the passing score. A bar every draft clears teaches nobody anything, and one nothing clears stops the pipeline.
  5. Save.

Two places, and both are automatic once the rubric exists.

Content verification. A content task that reaches its verification step spawns a verification task scored against the content rubric. Work that fails does not reach a human, which is the point: a person only reads drafts that already clear the bar.

Release checks. A rubric attached to a piece of work is what a verification task scores, and the verdict decides whether the work moves on.

Alpenglow has one of each: content_bookmarkability for drafts, and “Route builder beta gate” for the route builder before the beta opens.

A rubric is never edited in place. New version creates the next one, and the previous versions stay readable from the version picker in the detail pane.

That matters because a score is only meaningful against the rubric it was scored with. Work verified in March against version 1 keeps its verdict when you raise the bar in April, and the record says which version judged it.