Stress-test your assessment task before students do

Stress-test your assessment task before students do

Most teachers using AI in 2026 use it to write the task. Almost none use it to break the task. That is the move that earns its keep on a Saturday morning, because the failure mode you do not catch is the one that costs you a Sunday of marking and a Wednesday of "but Miss, that is not what I thought you meant". This prompt does the QA pass you would do if you had a head of faculty with no other meetings and a generous afternoon, and you do not.

The prompt

You are a senior [SUBJECT] teacher in an Australian [PRIMARY / SECONDARY / TERTIARY] classroom. Your job is to RED-TEAM the assessment task I am about to set. Not improve it. Not soften it. Break it. Find the holes before my students do.

I want the critic, not the cheerleader. If the task is fine, say so. If it is broken, say which line is broken and why.

THE TASK I AM SETTING

Paste of the task brief as students will see it:
[PASTE THE FULL STUDENT-FACING TASK BRIEF, INCLUDING ANY STIMULUS, WORD LIMIT, AND TIME ALLOWED]

The rubric as markers will see it:
[PASTE THE FULL RUBRIC, EVERY CRITERION AND EVERY LEVEL]

CONTEXT
- Year level: [E.G. YEAR 9]
- Subject and unit: [E.G. YEAR 9 SCIENCE, GENETICS, WEEK 6 OF 8]
- Class composition in one line: [E.G. 26 STUDENTS, MIXED ABILITY, FOUR EAL/D, ONE STUDENT ON AN IEP]
- The cognitive verb of the learning intention this task is meant to assess: [E.G. EXPLAIN, COMPARE, EVALUATE, JUSTIFY, DESIGN]
- Whether the task is take-home, in-class supervised, or hybrid: [PICK ONE]
- The stated AI policy for this task: [E.G. "AI USE NOT PERMITTED", "AI USE PERMITTED FOR INITIAL RESEARCH ONLY, DECLARED IN A LOG", "AI USE EXPECTED, DRAFT MUST BE ANNOTATED"]
- Time I expect to spend marking the cohort: [E.G. "ONE HOUR FOR ALL 26"]

PRODUCE SEVEN BLOCKS, IN THIS EXACT ORDER. EACH BLOCK IS SHORT. BULLET LISTS, NOT PARAGRAPHS. WHEN YOU NAME A PROBLEM, QUOTE THE EXACT LINE FROM MY BRIEF OR RUBRIC, IN QUOTATION MARKS. NO VAGUE GESTURES AT "THE INSTRUCTIONS".

BLOCK 1. THE AI SHORTCUT
The single fastest way a student can paste this task into a public chatbot and produce a response that meets most of the rubric without learning the content. Be specific. Name the model (Claude, ChatGPT, Gemini), name the prompt the student would type, name the rubric criteria the output would satisfy and the rubric criteria it would expose them on. If the task is genuinely AI-resistant, say WHY, naming the design feature that does the work (e.g. "the task is anchored in a stimulus the student wrote in class last Tuesday, so a chatbot has no access to it"). Do not write the cop-out sentence "students could use AI to complete this task".

BLOCK 2. THE TOP THREE MISREADINGS
Three specific instructions in the brief that a meaningful number of students at this year level WILL misread, and what each misreading produces on the page. Each entry starts with the exact phrase, in quotes. e.g. "'discuss the impact' will be read by half the cohort as 'list the impacts', producing five-line bullet responses that meet none of the analysis criteria, even from students who understand the content."

BLOCK 3. RUBRIC AMBIGUITY
Walk down the rubric criteria, one by one. For each, ask: would two competent markers from my faculty give the same grade for the same student response? If yes, say so in five words. If no, quote the rubric line that is the source of the disagreement and write a tighter phrasing. Pay attention to verbs that markers stretch differently ("sustained", "sophisticated", "convincing", "well-developed", "appropriate").

BLOCK 4. THE LAZY SHORTCUT THAT IS NOT AI
The way a student satisfies the rubric on the surface without doing the cognitive work the task is meant to assess. Examples: copying the structure of a previous exemplar, padding with definitions, restating the stimulus in their own words, performing the cognitive verb in the introduction sentence and abandoning it for the rest of the response. Name which rubric criterion the shortcut would satisfy, and the line in the brief that allows it.

BLOCK 5. THE EQUITY LINE
The single line in the brief that quietly disadvantages students with weaker home support, weaker English, or less out-of-school access (to internet, to materials, to time). Quote the line. State whether the fix is (a) keep the line and provide the access during class time, or (b) rewrite the line so the access is not required. Then write the rewritten line.

BLOCK 6. THE MARKING TIME-BOMB
The single decision I have made in this task that will turn a one-hour marking job into a four-hour one. Common culprits: open word counts with no upper limit, multi-format submissions in the same task, criteria that require cross-checking against work submitted earlier in the term, rubric levels that are not anchored to observable evidence. Quote the cause and write the smallest change that defuses it.

BLOCK 7. THE SMALLEST CHANGE THAT CLOSES THE BIGGEST HOLE
Of the six issues above, which one matters most for this class, this term, this task? If I had time to fix only ONE thing before I press print, what is it? Give me the line as it stands, in quotes, and the line as it should be, in quotes. One change. Not three.

CLOSING JUDGEMENT (one sentence, no caveats)
Pick one: SET THE TASK AS WRITTEN. SET IT WITH THE ONE FIX FROM BLOCK 7. DO NOT SET IT, REWRITE FROM SCRATCH. State which, and stop.

RULES
- Australian English. Year 9 not 9th grade. SAC, CAT, or assessment task, not "performance task" unless I called it that.
- No em-dashes, no en-dashes anywhere. Commas and full stops only.
- Do not invent content I did not paste. If a section of my brief is missing, ask for it before judging. Especially do not invent rubric levels I did not write.
- Do not soften every criticism. State the issue, then the fix. The teacher reading this is short on time and asked for the critic.
- If the task is solid, say so plainly. Not every task needs seven holes drilled in it.
- Do not use the words "consider", "explore", "leverage", "unpack", or "robust".
- Do not finish with "let me know if you want me to redraft the task". I will ask if I want that.

What it does well

Inverts the default move (use AI to write the task) into the much higher-value move (use AI to attack the task). The AI Shortcut block is the bit that earns its place, because most teachers have never asked the model to name the exact prompt a student would use to cheat, and the answer is usually unsettling and instantly fixable. The Rubric Ambiguity walk is the second highest-value block; it surfaces the verb in level 4 that you and the head of faculty have been quietly grading differently all year. The closing judgement forces the model to pick one of three calls, which stops the meandering "here are some considerations" close that wastes the last thirty seconds of your read.

Where it falls over

The model can only red-team what you paste. If you paste the task brief but not the rubric, the marking time-bomb block is guesswork. If you paste the rubric but not the stimulus, the AI Shortcut block will overstate what a chatbot can do. The prompt also cannot tell you what your specific cohort will misread; it can flag the high-probability landmines for the year level, but the student in row three who reads everything literally is your read, not the model's. And it is a QA tool, not a co-author. If the task itself is not aimed at the right learning intention, fixing the rubric ambiguity will not save it. Pull back further before you press print.

Try it like this

For the Year 10 history SAC you are finalising over the holidays: paste the brief, the rubric, and the source pack. The AI Shortcut block will tell you whether a student can paste the sources into Claude and get a B-grade response, and which paragraph of the brief makes that possible. Most of the time, the fix is a single line: "your response must refer to the marginalia and underlining we did together in class on [DATE]", which a chatbot has no access to.

For the Year 7 maths task you are setting next week: paste the brief and the worked-example you provided. Ask the prompt to focus on Block 4 (the lazy shortcut that is not AI). The model will tell you which step of the worked example students will reproduce verbatim without doing the reasoning, and the smallest change to the question that breaks the surface match.

For a primary writing task: paste the prompt, the success criteria, and the planning template. Ask the model to weight Block 5 (the equity line) heavily. Year 4 writing tasks often quietly require a parent at home to read the planning template back to the student. Naming that line is the fix.

If you only have five minutes from the kitchen on a Saturday, paste the task brief alone and ask the model for Blocks 1 and 7 only. The AI shortcut and the one fix. Run the other five blocks tomorrow if the task survives the first pass.

The task that earns its place is the one that has been attacked once before a student sees it.