LawGrader
Help Center
Grading and feedback

Scoring a criterion with rules

When a criterion is a checklist rather than a judgment: Binary and Count rules, adding up or deducting, and what the rail shows.

Updated September 12, 2026

Most criteria are judgments: the assistant reads the paper and places it on a level. Some are not. "Every source is cited in the required style", "no more than three missing citations", "a headline, a byline and a pull quote": these are things you check, one by one, and the score follows from what you found. A criterion scored by rules is built for those.

Setting one up

Open Set up criterion from the criterion's row in the rubric builder. Under Scored by, choose Rules instead of Look-fors. The header then shows two more choices:

  • Type: Binary (a rule is met or not) or Count (a rule fires once per occurrence, each occurrence worth a rate, up to a cap).
  • Direction: Add up (the student starts at zero and earns each rule's points) or Deduct (the student starts at full marks and each rule that fires takes points off, never below zero).
Set up criterion on a rules-scored criterion: Scored by Rules, a Type and a Direction chosen, and each rule with its worth.
Set up criterion on a rules-scored criterion: Scored by Rules, a Type and a Direction chosen, and each rule with its worth.

Each rule is a row: the check in plain words, and its worth. On a Count rule the worth is a rate with a cap ("2 each, up to 6"). The right edge totals what the list can move ("up to 10 off"). Type a rule in the box under the list, or pick one from the library of common checks.

Switching Type or Direction is a conversion, and the modal says what it will cost before you confirm: a Binary list turned into a Count list needs rates and caps, and a Deduct list turned into Add up flips every sign. Under a Fixed point, Ranges or Points only criterion, rules are optional and only adjust a score the criterion already has; under Rules scoring they are the score, so the criterion's worth is their sum (or, deducting, the base they take from). See Building a rubric for the rest of the modal.

What the grader does with it

On a rules-scored criterion the assistant does not pick a level. For each rule it returns a verdict, a one-line reason, and the passage it is looking at, quoted word for word. That is all it is asked for, so there is nothing to second-guess beyond "did this fire, and where".

What the rail shows

On the grade page the criterion's card opens a Rules drawer instead of a ladder, and the caption names your setup in the modal's own words: "Binary rules, adding up from 0 · whole points, a partial earns half", or "Count rules, deducting from 10 · charged per occurrence, up to each cap".

The rail card for a rules-scored criterion: one row per rule with its mark, a link to its margin comment, and the arithmetic stated once at the foot.
The rail card for a rules-scored criterion: one row per rule with its mark, a link to its margin comment, and the arithmetic stated once at the foot.
  • Each rule is a row with its mark. A Binary rule has a mark you set to met, partial or missing; a Count rule has a stepper (1 of 3) that turns red at its cap.
  • The ↗ beside a row jumps to the margin comment the grader left for it, so you can see the passage it counted.
  • The foot states the arithmetic once: Deducted 2 of 10 · Criterion 8 / 10.
  • Change any mark and the criterion score follows. Your marks win over the grader's; nothing is recomputed behind your back.

A partial on a Binary rule earns or takes half the rule's worth, snapped to the criterion's rounding (whole points, halves or tenths, set in the modal). It is offered only when half is neither zero nor the whole.

Where rules do not apply

Judgment criteria (argument quality, organization, voice) belong on levels, not rules; a checklist cannot see excellence, only compliance. A good rubric usually has both: a few rules-scored rows for the mechanical requirements, and levels for everything that takes reading.

Was this helpful?

Keep reading

  • Uploading papers — Which files work, how a stack is matched to students, and what each status in the paper list means.
  • How grading works — The assistant proposes; you decide. How scores are made and how you stay in control.
  • Validating the grader against your grading — Run a few papers you have already graded on the Validate step, so the assistant learns how you actually score before it grades the class.
  • Grading a whole class at once — Once validation satisfies you, grade the whole set in one pass, then review the results.
  • Reviewing and adjusting the feedback — Adopt the assistant's observations you agree with, dismiss the rest, and write your own.
  • Reading the grade rail — How much of the AI's read to show, where each score came from, what the look-fors say, and how to filter the margin.
  • Marking up a paper — The grading page: the pen tools, your own comments, suggested rewrites, what students see, and undo.
  • The assistant, and talking to it — The three places you can ask the assistant for help, what each one knows, and how to dictate instead of type.
  • When a grade needs a second try — Stopping a run, re-grading a paper, what a failed or incomplete grade means, and what happens during an AI outage.
  • Your comment library — Every comment you write collects in one library: browse it while grading, insert with a keystroke, and bring in the comments you wrote before.
  • Comment suggestions in your own voice — As you write a margin comment, LawGrader suggests the rest of it, drawn from the comments you have written before.
  • Giving the grader supporting docs — Hand the assistant the documents you grade against: the set reading, a reference text, a style guide.