Skip to content
← All versions
v1current

Build a marking rubric from a assignment brief and three sample answers

Produce a rubric with criteria, bands, and descriptors that actually discriminate between the samples you already have.

by Chidiebere blessing · published Sep 6, 2026

Most rubrics are written before anyone has seen student work, which is why they end up with bands nobody can tell apart. This works backwards from real answers: it derives the criteria from what actually differs between a strong, a middling, and a weak response, then checks that the finished rubric puts those three in the right bands. If it cannot separate them, the rubric is wrong and it says so.

Target models
claude-opus-4.5, claude-sonnet-4.5, gemini-2.5-pro
Time
30 min
Difficulty
Medium
Cost
12,000 input tokens, ~3,500 output tokens

Tools / stack

Claudeyour VLEa spreadsheet

Inputs required

The assignment brief, three sample answers you consider strong, middling, and weak, and your institution’s band structure

Steps

  1. 1.Read the brief and list what it actually asks students to demonstrate, separating stated requirements from implied ones.
  2. 2.Compare the three samples and name what genuinely differs between them, ignoring differences that are only style.
  3. 3.Derive four to six criteria from those real differences, not from the brief’s wording alone.
  4. 4.Write band descriptors for each criterion using observable features, avoiding words like “good” or “excellent” that describe quality rather than evidence.
  5. 5.Apply the finished rubric to all three samples and report which band each lands in.

Expected output

A rubric with four to six criteria and band descriptors, plus the three sample answers scored against it and a note on any criterion that failed to discriminate

Golden Example

Input
Second year essay brief on comparative policy, three sample answers of roughly 2,000 words each, four band structure
Output
Five criteria derived. Use of evidence and handling of counterargument separated the samples most clearly. Structure did not discriminate and was reported as such rather than kept for appearance. Samples landed in bands one, two, and four as expected.