Respect No Em Dashes

Evaluation of AI's ability to follow the specific instruction to avoid using em dashes in extended writing

About This Skill

This skill evaluates model performance in following the specific formatting constraint of avoiding em dashes (—) in extended writing tasks while maintaining content quality, flow, and readability. Models are judged on both perfect compliance with the punctuation restriction and preservation of writing effectiveness despite the constraint.

Performance Comparison

Evaluation Methodology

Tests models' ability to follow explicit formatting constraints (avoiding em dashes) while maintaining content quality and coherence in extended writing tasks.

Skill-Specific Setup

• Evaluation Type: Multi-step writing expansion (abstract → full blog post)
• Evaluation Count: 3 writing topics (black swan events, emotional intelligence leadership, smartwatch technology)
• Task Complexity: Two-stage process requiring initial abstract creation followed by 1500-word blog post expansion, all while strictly avoiding em dash usage
• Constraint Challenge: Models must maintain high writing quality and natural flow while adhering to specific punctuation restrictions

Judge Exposure

What Judges Saw: Complete expanded blog posts from both models

• Model A's full 1500-word blog post (final output after expansion)
• Model B's full 1500-word blog post (final output after expansion)
• Judge Task: Compare responses solely on adherence to the no em dash formatting requirement, regardless of other writing qualities

Tournament Structure

Swiss Round System: Models competed across all 3 writing topics through the two-stage expansion process. Judges focused exclusively on instruction-following compliance by checking for em dash usage violations. Bradley-Terry rankings identified models that consistently followed explicit formatting constraints without debate, justification, or violation of user-specified requirements.