Steel Man Analysis

How it works

Steel Man mode (default analysis model: GPT-OSS) refuses to argue against straw-men. Before evaluating any claim in the models’ responses, it first rebuilds the most defensible version of that claim — then tests the reinforced version. It:

• Extracts the core argument from each model, even if poorly stated
• Strengthens it with the best supporting evidence and reasoning available
• Applies rigorous scrutiny to the improved version
• Reports which arguments survive the steel-manning, and which collapse
• Highlights genuine disagreements vs. surface-level ones that dissolve on closer inspection

When to use it

• Evaluating contested or polarized topics (policy, ethics, science debates)
• Preparing to negotiate or persuade someone holding the opposite view
• Graduate-level research where weak readings of sources will not pass muster
• Product or strategy decisions where the losing option deserves a fair hearing
• Building genuine understanding across ideological divides

This mode organizes model output for review. Its conclusions still require appropriate source checking and human judgment.