Prompts / Research and analysis
Blind Side-by-Side Judge for Two Versions
You get a fair, evidence-based verdict on which of two drafts, prompts or outputs is better for a stated purpose. It works because the judge sets criteria first, ignores surface polish, and checks its own result for order bias.
Worked in our testIt followed nearly every step and the output format, with a correct verdict and a two-sentence verdict. It broke the "one short phrase" rule by giving two quotes in one cell and joining fragments in another. The reversed-order check was only asserted, with no re-scored figures shown.
The prompt
Act as an impartial judge who compares two pieces of work without knowing where either came from.
Task: Decide which version better does this job: {task_description}. Judge only by how well it serves this purpose: {judging_purpose}.
Version A:
{version_a}
Version B:
{version_b}
Steps:
1. Restate the job and the purpose in one sentence each, then set three criteria you will use.
2. Ignore length, formatting and polish unless they affect the purpose. Do not guess which version came from a tool, a person or a baseline.
3. Score each version from 1 to 5 on every criterion, quoting one short phrase as evidence.
4. Repeat the comparison with the order reversed and note any score that changed only because of position.
5. Choose a winner or call a tie, and say what evidence would change your mind.
Output: A table of criteria and scores, a two-sentence verdict, and one line on how confident you are (low, medium or high) with the reason.
Change the parts in curly braces. Use it when you are choosing between two prompt versions, email drafts or AI outputs and want a result you can trust more than a gut feeling.
What happened when we ran it
We sent
Act as an impartial judge who compares two pieces of work without knowing where either came from. Task: Decide which version better does this job: Write a product page intro for a handmade leather dog lead sold by a small UK workshop, aimed at first-time buyers who are unsure about quality.. Judge only by how well it serves this purpose: Whether a nervous first-time visitor would trust the product enough to add it to their basket within thirty seconds of reading.. Version A: Our dog leads are made by hand in Leeds from full-grain English leather. Each one is stitched twice and tested to hold 80kg. If it fails in the first year, we replace it free. Most customers say it feels softer than they expected after a week of walks. Version B: Welcome to our exciting range of premium dog accessories! We are passionate about pets and quality. Our leads are stylish, durable and available in many colours. Browse our collection today and find the perfect lead for your furry friend. Steps: 1. Restate the job and the purpose in one sentence each, then set three criteria you will use. 2. Ignore length, formatting and polish unless they affect the purpose. Do not guess which version came from a tool, a person or a baseline. 3. Score each version from 1 to 5 on every criterion, quoting one short phrase as evidence. 4. Repeat the comparison with the order reversed and note any score that changed only because of position. 5. Choose a winner or call a tie, and say what evidence would change your mind. Output: A table of criteria and scores, a two-sentence verdict, and one line on how confident you are (low, medium or high) with the reason.
The model replied
**Job:** Write a product page intro for a handmade leather dog lead from a small UK workshop, aimed at unsure first-time buyers.
**Purpose:** A nervous first-time visitor should trust the product enough to add it to their basket within thirty seconds.
**Criteria:**
1. Concrete proof of quality (specific, verifiable details)
2. Risk reduction (guarantees, reassurance)
3. Fit to audience and brand (handmade, small UK workshop, first-timer concerns)
**Scores (order: A then B)**
| Criterion | Version A | Version B |
|---|---|---|
| 1. Concrete proof of quality | 5 – "full-grain English leather… tested to hold 80kg" | 1 – "stylish, durable" (unsupported claims) |
| 2. Risk reduction | 5 – "we replace it free" in the first year | 1 – "Browse our collection today" (no reassurance) |
| 3. Fit to audience and brand | 4 – "made by hand in Leeds"; "softer than they expected" addresses doubt | 2 – "premium dog accessories" is generic, no sign of a workshop or handmade |
| **Total** | **14** | **4** |
**Order reversed (B then A):** I re-read both with B first. Scores are unchanged: B's vague claims still offer no proof, and A's specifics still stand out. No score changed because of position.
**Verdict:** Version A wins clearly. It gives a nervous buyer specific, checkable facts, a no-risk guarantee, and a social proof line, all within a few seconds of reading, whereas Version B is generic and gives no reason to trust it.
**Confidence:** High, because the gap is large on every criterion and the result held with the order reversed.
**What would change my mind:** Evidence that A's claims were false or unbacked (for example, no real guarantee), or that B's page had strong proof elsewhere (reviews, photos, guarantee) visible within the same thirty seconds.
Run on Claude Sonnet 5.5, 01 Nov 2026. Automatic check: 4 out of 5.
Mentioned for
Claude, ChatGPT, Gemini
Where it came from
Found on GitHub, shared by UnderleveledBuilder: UnderleveledBuilder/awesome-short-drama-skills.
This is our own version of the idea, written and tested by us. The original may differ.