All axes
Every entry across seven axes and four stages. Narrow it further from the sidebar.
Workbench
A Practical Opus 5.2 Research Workflow for Consultants
This Workbench article provides a practical, end-to-end research workflow for consultants using ...
Field Tests
Five Repeated Tests That Reveal Whether an AI Workflow Is Stable
This Field Tests article presents five repeated-test methods that expose instability in Opus 5.2...
Failure Reports
What Opus 5.2 Missed in a Document Review With Conflicting Evidence
This Failure Reports article examines a controlled document-review task containing deliberate co...
Under the Hood
How to Build a Small Evaluation Set for Your Own Opus 5.2 Use Case
This Under the Hood article explains how to construct a small, reusable evaluation set tailored ...
Workbench
Opus 5.2 for Customer Support: Drafting Answers Without Losing Escalation Judgment
This Workbench article presents a controlled workflow for drafting customer-support replies with...
The Human Check
The Difference Between Delegation and Abdication
This Human Check essay distinguishes responsible delegation of work to Opus 5.2 from abdication ...
Field Tests
My 30-Day Verdict: Where Opus 5.2 Earned a Permanent Place in the Workflow
This Field Tests article delivers a 30-day verdict on Opus 5.2. It identifies the workflow categ...
Field Tests
My Opus 5.2 Testing Protocol: Tasks, Scores, Costs, and Human Review
This Field Tests article lays out a reusable protocol for evaluating Opus 5.2 on realistic profe...
The Human Check
Why This Blog Exists: Useful Is a Higher Bar Than Impressive
This opening essay explains the purpose of The Model Bench: rigorous, reproducible testing of Op...