2.6 Platform Rubric Patterns

Study Time: 2 hours Prerequisites: Module 2.5 Learning Objectives:

• Recognize common rubric patterns across platforms
• Adapt evaluation approach to different platform styles
• Build platform-agnostic transferable skills
• Understand platform-specific terminology
• Create systematic evaluation checklists

---

Introduction

While specific platform requirements vary, common patterns emerge across all major AI evaluation platforms. This module teaches you to recognize and adapt to these patterns.

---

2.6.1 Universal Patterns

Pattern 1: Quality Tiers

Most platforms use 3-5 tier scoring:

3-Tier (Simple/Hard/Abstention):

• Pass/Fail with complexity rating
• Common in error-fixing tasks

5-Tier (1-5 scale):

• 5 = Exceptional
• 4 = Good (meets requirements)
• 3 = Acceptable (minor issues)
• 2 = Poor (major issues)
• 1 = Unusable

Application: Structure criteria to distinguish between tiers

---

Pattern 2: Common Components

Most platforms require some combination of:

• Factual accuracy verification
• Completeness checks
• Format/constraint adherence
• Safety/appropriateness (no harmful content)

Check your specific project guidelines to confirm which components are required.

---

Pattern 3: Modality Requirements

Common expectation: Audio/video tasks typically require modality-specific criteria.

Platforms may call it:

• "Audio quality assessment"
• "Technical evaluation"
• "Production standards"
• "Delivery assessment"

Same principle, different names.

---

2.6.2 Platform Terminology Translation

Concept: Rubric/Criteria

Called:

• "Evaluation criteria"
• "Assessment dimensions"
• "Quality factors"
• "Review rubric"
• "Scoring guidelines"

All mean: Standards for evaluating content

---

Concept: Atomicity

Called:

• "Single-factor assessment"
• "One criterion per dimension"
• "Focused evaluation"
• "Discrete scoring elements"

All mean: One aspect per criterion

---

Concept: Instance-Specific

Called:

• "Task-relevant criteria"
• "Content-specific evaluation"
• "Contextual assessment"
• "Prompt-aligned rubric"

All mean: Reference actual task content

---