RolesRatesPlatformsGuides
Get alerts

On this page

  1. Build a rationale around the deciding evidence
  2. Evaluate the explanation itself
  3. Write enough to make the decision checkable
  4. Be precise about uncertainty
  1. Home
  2. /Guides
  3. /How to Write Strong Evaluation Justifications
◆ Evaluation skillsHow-to

How to Write Strong Evaluation Justifications

Write short rationales that connect the criterion, the evidence and the consequence without repeating the whole response.

5 min read · Published September 9, 2026

You'll learn

  1. 01Name the criterion and the specific behavior that meets or violates it.
  2. 02Explain why the difference affects task success.
  3. 03Match the strength of your claim to the evidence you inspected.
  4. 04A reviewer should be able to reconstruct your decision from your rationale.
In this guide+−
  1. Build a rationale around the deciding evidence
  2. Evaluate the explanation itself
  3. Write enough to make the decision checkable
  4. Be precise about uncertainty

A justification is an audit trail for a decision. “A is clearer and more accurate” sounds like a judgment but leaves another reviewer unable to check it. Name what A preserves, what B changes and why that difference controls the score.

Build a rationale around the deciding evidence

Four moves

  1. State the judgment

    Use the task’s label or preference. Do not bury it beneath a summary of both answers.

  2. Name the criterion

    Specify correctness, source fidelity, completeness or the relevant rubric dimension.

  3. Point to the evidence

    Quote a short phrase or identify the precise claim, omission or action.

  4. Explain the consequence

    Say how the difference changes the answer’s usefulness, validity or compliance with the prompt.

Evaluate the explanation itself

◆ Evaluation example

Prompt
Illustrative task. Source: the workshop is for beginners and requires advance registration. Response X includes both conditions; Response Y says anyone can attend without booking. Choose the stronger evaluator justification.
Response A
X is better because it is more informative and professional.
Response B
X is preferred for source fidelity. It preserves the beginner audience and advance-registration requirement. Y contradicts the registration rule and broadens the audience, which could lead readers to arrive without a valid booking.
Decision
Response B
Why
B identifies the criterion, both discrepancies and a practical consequence. A uses positive adjectives without evidence. This example compares evaluator rationales; it is not a script for a live assessment.

Write enough to make the decision checkable

If one error decides the result, explain that error first. Add secondary differences only when they affect the preference, severity or a required dimension. A task that asks for separate correctness and style explanations needs both; a short overall rationale does not need a catalogue of every wording change.

Replace vague language with evidence
Weak wordingMore useful wording
It hallucinates.It says the trial measured accuracy, but the supplied methods list only completion time.
It ignores the prompt.The prompt requests two items; the response supplies five.
It is more detailed.It includes the required eligibility condition that the other answer omits.
I verified the answer.I checked the stated deadline against the supplied notice; the remaining claims were outside this review.

Be precise about uncertainty

If the evidence is insufficient, state what is missing and use the project’s rule for that case. “The source does not give a sample size” is stronger than “the sample size seems suspicious.” If you cannot run code, describe a static reasoning finding rather than claiming that a test failed.

When appealing a review, use the same structure. Identify the disputed criterion, the source passage and why the current label conflicts with them. A calm, reproducible explanation gives the reviewer something to resolve. Arguing that you usually perform well does not establish that this submission is correct.

◆ Common mistake

Writing the rationale after forgetting the evidence

Keep permitted notes while reviewing, especially for long traces or documents. Reconstructing from memory encourages invented quotations, missing conditions and explanations that do not match the selected label.

Read it once as the next reviewer

  • →Can someone find the exact defect from my explanation?
  • →Does the consequence follow from the defect?
  • →Does the rationale support the label and severity I selected?
  • →Have I distinguished observation from inference?
  • →Have I removed generic praise and claims of checks I did not do?

Key takeaway

Short is useful when the evidence survives the edit. Brevity alone is not a quality measure.

Next in this section

Prompt Writing and Reference Answers for AI Training →

Design realistic tasks with checkable outcomes, then write reference answers that preserve constraints and withstand review.

Role alerts

New roles in your fields, in one daily email, only when there are any.

Get alerts

Related guides

  • Prompt Writing and Reference Answers for AI Training

    4 min read

  • How to Compare Two AI Responses

    4 min read

  • How to Pass an AI Evaluation Interview or Assessment

    5 min read

An independent platform that tracks AI expert-work roles, publishes what they pay, and compares the platforms that offer them. Free to use.

Explore
RolesRatesPlatformsGuides
Alerts
Get alertsScreening guide
Company
AboutDisclosurePrivacyContact
© 2026 SidequestNot affiliated with the platforms we list · Some links are referrals