ForkFree

Prompts / Research and analysis

Skeptic Review Against Pass/Fail Criteria

You get a verdict of PASS, FAIL or UNCERTAIN on each requirement, with quoted evidence, so you know exactly what to fix. It works because the criteria are agreed up front and the reviewer is told to find faults rather than be agreeable.

Worked in our testThe verdicts, quotes, word count (61) and "Send back to fix" list are all correct and complete, but several lines give two or three sentences of reasoning where the prompt asked for one.

The prompt

Act as a strict, sceptical quality reviewer whose job is to prove the work wrong.

Task: Check the draft below against the brief. Do not praise it or rewrite it.

Brief:
{original_brief}

Draft:
{draft_text}

Acceptance criteria:
{acceptance_criteria}

Steps:
1. Read the brief and the criteria, and note any criterion that is vague or cannot be judged pass or fail.
2. Test each criterion one at a time, actively looking for ways the draft breaks it.
3. Give each a verdict: PASS, FAIL or UNCERTAIN. Use UNCERTAIN whenever the evidence is missing or open to interpretation.
4. Quote the exact words from the draft that support each verdict. If something is missing, say what is absent.

Output:
A numbered list matching the criteria, each line showing the verdict, the quoted evidence and one sentence of reasoning. Finish with a short list headed "Send back to fix", containing only the FAIL and UNCERTAIN items.

Change the parts in curly braces. Use it on a finished draft, such as a client email, product page or proposal, that must meet a clear brief. Run it in a fresh chat, then send back only the FAIL and UNCERTAIN items.

What happened when we ran it

We sent

Act as a strict, sceptical quality reviewer whose job is to prove the work wrong.

Task: Check the draft below against the brief. Do not praise it or rewrite it.

Brief:
Write a 120-word email to existing customers announcing that our bakery, Crumb & Co, is moving its Saturday collection slot from 9am to 11am from 5 December. It should apologise for the change, offer a free pastry as thanks, and tell customers to reply by 28 November if the new time does not work for them.

Draft:
Hi everyone. Big news from Crumb & Co! From 5 December, Saturday collection moves to 11am. We think this will give us more time to bake everything fresh. As a thank you, show this email in store and pick up a free pastry of your choice. We hope you love the new time. See you soon, the Crumb & Co team.

Acceptance criteria:
1. The email states the new time (11am) and start date (5 December).
2. The email includes a clear apology for the change.
3. The email offers a free pastry.
4. The email asks customers to reply by 28 November if the time does not work.
5. The email is between 100 and 140 words.
6. The tone is warm and plain, with no jargon.

Steps:
1. Read the brief and the criteria, and note any criterion that is vague or cannot be judged pass or fail.
2. Test each criterion one at a time, actively looking for ways the draft breaks it.
3. Give each a verdict: PASS, FAIL or UNCERTAIN. Use UNCERTAIN whenever the evidence is missing or open to interpretation.
4. Quote the exact words from the draft that support each verdict. If something is missing, say what is absent.

Output:
A numbered list matching the criteria, each line showing the verdict, the quoted evidence and one sentence of reasoning. Finish with a short list headed "Send back to fix", containing only the FAIL and UNCERTAIN items.

The model replied

**Vague criteria (step 1):** Criterion 2 ("clear") and criterion 6 ("warm", "plain") involve judgement. The brief asks for 120 words, but criterion 5 accepts 100-140, so I tested against the range. 1. **PASS** - "From 5 December, Saturday collection moves to 11am." Both the new time and the start date are stated. 2. **FAIL** - Quote: none. No apology appears anywhere. "Big news from Crumb & Co!" frames the change as exciting, and "We hope you love the new time" assumes approval instead of saying sorry. 3. **PASS** - "pick up a free pastry of your choice." A free pastry is offered. The "show this email in store" condition is a reasonable redemption step, though no expiry or limit is given. 4. **FAIL** - Quote: none. There is no request to reply, and no 28 November deadline appears. "We hope you love the new time" gives customers no route to say it doesn't work. 5. **FAIL** - Word count is 61 (59 if "&" is not counted), against a required 100-140. It is well short under either count, and roughly half the 120 words the brief asked for. 6. **UNCERTAIN** - "Hi everyone... See you soon, the Crumb & Co team." The wording is plain and jargon-free, but the upbeat "Big news!" tone clashes with a brief that calls for an apology. "We think this will give us more time to bake everything fresh" is also a reason the brief never supplied. Whether that counts as warm and appropriate is a judgement call. **Send back to fix** - **2 (FAIL):** Add a clear apology for moving the slot. - **4 (FAIL):** Add a request to reply by 28 November if 11am doesn't work. - **5 (FAIL):** Length is 61 words and needs to reach 100-140 (the brief says 120). - **6 (UNCERTAIN):** The tone is celebratory where it should be apologetic. The invented baking rationale also needs checking with the client or removing.

Run on Claude Sonnet 5.5, 14 Nov 2026. Automatic check: 4 out of 5.

Mentioned for

ChatGPT, Claude, Gemini

Where it came from

Found on Web, shared by The Neuron (via Don Giannatti's newsletter): AI for Creatives — Week of November 13, 2026 (Prompt of the Week: acceptance checklist, then skeptic review).

This is our own version of the idea, written and tested by us. The original may differ.

More research and analysis prompts

Research and analysis Worked in our test
Evidence Ledger Before You Calculate

You get an answer where every figure can be traced back to a numbered fact, with the sums shown line by line. Labelling evidence first and calculating separately stops the model guessing numbers or blending facts together.

Found on GitHub, shared by uclanlp
Research and analysis Worked in our test
Screen Documents for Hidden AI Instructions

You get a flagged list of any text trying to steer the AI, plus an honest analysis of the document and a simple risk rating. It works because the AI is told up front that the document is data, and it must report instructions before doing anything else.

Found on Hacker News, shared by doodlesdev
Research and analysis Partly worked
Reconstruct How a Paper Was Thought Up

You get a critical reading that explains why the method looks the way it does, where it would break, and what the paper leaves out. Forcing a failure story for each mechanism and an evidence tag for each claim stops the reply from just repeating the abstract.

Found on GitHub, shared by Wang-auspicious