2.1 Atomicity Bootcamp

Study Time: 3.5 hours Prerequisites: Module 2.0 (Ideal Response Vision) Learning Objectives:

• Define atomicity in the context of evaluation criteria
• Recognize non-atomic criteria instantly
• Decompose multi-aspect criteria into atomic components
• Write atomic criteria from scratch for any task type
• Distinguish between atomic and stacked criteria and when stacking may be acceptable
• Avoid a common cause of evaluation quality failures

---

Introduction

In Module 2.0, you learned to envision the ideal response before writing criteria. You practiced identifying what excellence looks like and deriving criteria from that vision.

Now let's ensure each criterion you derive is properly atomic, the first of the Five Core Properties that make criteria professionally written.

Atomicity is a core skill for professional evaluation work.

Non-atomic criteria cause quality check failures because they make consistent scoring impossible. This module teaches you to recognize these problems and fix them.

What is Atomicity?

Atomic Criterion: A criterion that tests exactly ONE specific aspect or quality.

Non-Atomic Criterion: A criterion that tests TWO or more aspects simultaneously.

Why It Matters

When a criterion tests multiple things at once, it becomes:

• Ambiguous: Unclear which aspect caused a pass/fail
• Inconsistent: Different evaluators focus on different aspects
• Unusable: Impossible to score when one aspect passes and another fails

Example:

Criterion: "Is the response accurate and well-organized?"

Problem: What if the response is accurate but poorly organized?
- Do you mark it as PASS (because it's accurate)?
- Do you mark it as FAIL (because it's disorganized)?
- Do you mark it as PARTIAL (but that's not an option)?

This ambiguity makes the criterion useless for consistent evaluation.

---

2.1.1 The Atomicity Principle

The Core Rule

ONE CRITERION = ONE ASPECT

If you can replace "and" or "or" with separate criteria, split them.

Visual Test: The "AND" Detector

Read your criterion out loud. If you say "and," "or," or list multiple things, it's non-atomic.

Examples:

❌ "Is the audio clear and the content accurate?"

• Tests 2 things → NON-ATOMIC

❌ "Does the response include examples, citations, and proper formatting?"

• Tests 3 things → NON-ATOMIC

❌ "Is the video well-framed or properly lit?"

• Tests 2 things → NON-ATOMIC

✅ "Is the audio free from background noise?"

• Tests 1 thing → ATOMIC

✅ "Does the response include at least 3 specific examples?"

• Tests 1 thing → ATOMIC

The Question Test

Ask yourself: "What specific aspect am I evaluating?"

If your answer includes multiple points, it's non-atomic.

Example:

Criterion: "Is the explanation clear and complete?"

Question: What am I evaluating?
Answer: "Clarity of writing AND completeness of information"

Result: Two answers = NON-ATOMIC

The Failure Scenario Test

Imagine a response that passes one aspect but fails another:

If you can't score it clearly, the criterion is non-atomic.

Example:

Criterion: "Is the response helpful and polite?"

Scenario: Response provides excellent help but is rude.
- Is this a PASS? (helpful)
- Is this a FAIL? (impolite)
- UNCLEAR → Non-atomic criterion

---