Score Blind

Brand preference is loud. Blind ranks are quieter and more honest. Lock 4–6 scoring criteria with Copilot before you paste any outputs: accuracy, structure, actionability, tone, calibrated confidence, usefulness-as-is — whatever this job actually needs.

You will otherwise favor the option you already pay for.

Judge in a separate Copilot chat. Paste A and B only. Do not tell it which option wrote which. If you ran variants, paste them under each letter.

If solo logistics are hard: wait overnight so you forget which tab was which, or ask a colleague to paste the letters.

Core principles

  1. Criteria before outputs. Scoring after you have seen the winner is fan fiction.
  2. Score 1–5 per criterion with a one-line why. Rank overall.
  3. Note where rankings flip across variants of the same task type — that is the interesting result.
  4. Describe house style per letter: what that author seems built for, and what to watch.
  5. Reveal the mapping only after the verdict.

Check yourself

When should you write scoring criteria?

Do this in Copilot

Lock criteria, paste A and B into a fresh Copilot chat, score, then reveal your mapping on paper.

New Copilot chat. Lock criteria first. Paste letter-labeled outputs second. Do not reveal the mapping in this chat until after the rank.

Blind judging

I ran the same Copilot task on more than one option and want a fair blind score. Step 1: Interview me on the task type, who the output is for, and what “great” looks like. Propose 4–6 scoring criteria for THIS job. Lock them with me BEFORE I paste outputs. Step 2: I will paste outputs labeled A, B (and C if needed). Do not guess which Copilot option wrote which. Step 3: Score each 1–5 per criterion with a one-line justification. Rank overall. Note if rankings would flip across variants. Step 4: Describe the house style of each letter — what work it seems built for, and what I should watch. Step 5: Stop. I will reveal the mapping myself after your verdict.

Techniques: Claim spot-check · Constraints

Open Copilot

Did you run this in Copilot? Mark complete when you have tried it.

Next lesson: Write a Routing Card →

AI

Part of the MillerKnoll collective

Studio · 2026