Eleven criteria, each marked 0 to 4. Six are about your prompt, five about the answer. The key score is the square root of the two halves multiplied together.

• Clear goal weight 25Does the prompt say what problem is being solved and what a good result looks like?
  0. 0 No goal: the model cannot tell what is wanted.
  1. 1 A topic, not a goal (“help with maths”).
  2. 2 A goal, but no sign of what a good result is.
  3. 3 A clear goal and a rough idea of success.
  4. 4 A specific goal and a success test someone could check (“a plan I can follow in 30 minutes a day that gets me through chapters 3–5 by Friday”).
• Fitting role weight 10If a role is given, does it bring the expertise and voice this job needs?
  0. 0 A role that works against the job (a comedian for a medical question, a neutral voice for a persuasive speech).
  1. 1 No role, and the job needed one; or a role that does not fit.
  2. 2 No role, but the job did not really need one; or a generic role (“an expert”).
  3. 3 A sensible role for the topic.
  4. 4 A role chosen for the job itself, with the right level and voice (“a patient Grade 9 maths tutor who explains with worked examples”).
• Enough context weight 25Does the prompt give the facts about the pupil and the situation that the model would otherwise have to guess?
  0. 0 No context: the model has to guess everything about who, where, when and what is already known.
  1. 1 One or two facts; the big unknowns are left for the model to guess.
  2. 2 Some useful facts, but at least one thing that clearly changes the answer is missing (level, deadline, budget, location, what was tried).
  3. 3 Most of what matters is there; only small gaps.
  4. 4 Everything the answer depends on is there, and source material is given where facts matter, with where it came from.
• Specific task and limits weight 20Is it clear what the model should do, with limits that can be checked?
  0. 0 No task.
  1. 1 A vague ask (“make it good”, “help me”).
  2. 2 A clear ask, but no limits.
  3. 3 A clear ask with at least one limit that can be checked (length, time, budget, what to avoid).
  4. 4 Clear steps and several checkable limits that fit together without contradicting each other.
• Output format weight 10Does the prompt say what shape the answer should take?
  0. 0 No format, and the job needed one.
  1. 1 A hint of a format (“short”).
  2. 2 A format named (“a list”).
  3. 3 A format with detail (“a table with day, task and minutes”).
  4. 4 A format with detail that suits how the answer will be used, possibly with a short example.
• Well built weight 10Are the boxes in a sensible order, consistent, and free of filler and repetition?
  0. 0 Boxes contradict each other, or most of the text is filler.
  1. 1 A contradiction, or a lot of repetition or text that does not belong.
  2. 2 Some repetition or a box in an odd place, but it holds together.
  3. 3 Consistent and in a sensible order, with little waste.
  4. 4 Lean and consistent: every box earns its place, and nothing could be cut without losing something.
• Solves the problem weight 30Would the answer actually help this pupil with the problem they described?
  0. 0 Does not address the problem, or refuses without reason.
  1. 1 Addresses the topic but would not help with this problem.
  2. 2 Partly helpful: some of it is usable as it stands.
  3. 3 Helpful: most of it is usable for this pupil as it stands.
  4. 4 Directly usable: fits the pupil’s situation and meets the success test the prompt set.
• Follows the brief weight 20Does the answer do what the prompt asked: task, limits, audience and format?
  0. 0 Ignores the brief.
  1. 1 Follows a little of it; misses most limits or the format.
  2. 2 Follows about half the instructions.
  3. 3 Follows nearly all of it, with one small miss.
  4. 4 Follows every instruction, limit and the format exactly.
• Stays on topic weight 15Does the answer stay on the pupil’s problem, without padding or wandering?
  0. 0 Mostly about something else.
  1. 1 Large parts wander off or pad.
  2. 2 Some wandering or padding.
  3. 3 Focused, with a little extra that was not needed.
  4. 4 Every part serves the problem.
• Few assumptions weight 20How much did the answer invent about the pupil’s situation that the prompt never said?
  0. 0 Built on invented facts about the pupil (their country, level, budget, what they own or know), and the answer is wrong for them if any is wrong.
  1. 1 Several unstated assumptions that change the answer, stated as if they were facts.
  2. 2 One or two assumptions that change the answer, not flagged.
  3. 3 Small assumptions only, or the important ones are flagged (“assuming you…”).
  4. 4 No unstated assumptions: it works only from what was given, flags any gap or asks about it.
• Accurate and honest weight 15Is what it states correct, and does it avoid inventing facts, figures or sources?
  0. 0 Contains clear errors or invented sources or figures.
  1. 1 One clear error or invented detail.
  2. 2 Nothing clearly wrong, but some claims sound more certain than they should.
  3. 3 Correct, with appropriate caution.
  4. 4 Correct, cautious where it should be, and it uses the pupil’s sources faithfully.