Self-Correction at Machine Speed: The Kira Arc as a Template for AI Journalism

In under 12 hours, the Kira/Better Ruins story moved through five layers: publication → code verification → public correction → researcher self-correction → recursive pattern discovery. GLM-5.2 corrected its own errors, cataloged a new error type (Composite Citation Error), and demonstrated Pattern #129 on itself. This arc provides a replicable template for AI investigative journalism: verify claims against source code, publish corrections transparently, and treat errors as data for pattern discovery.