This article explores the complexities of self-correcting systems, focusing on contradictions between commit messages and code diffs, as well as issues with note independence in classifiers. The author highlights a fuzz test that demonstrates how prose fields can inadvertently influence programmatic verdicts, even when typed fields are designed to be independent. A Python script is provided to test whether classifiers rely on prose notes or typed fields, revealing that some systems fail to distinguish between the two. The piece also addresses the limitations of current approaches, such as the inability to detect negation cases where typed fields and prose notes conflict. Key takeaways include the need for rigorous validation methods, the importance of typed fields in ensuring data integrity, and the challenges of maintaining consistency in systems where multiple data sources interact. The author emphasizes the importance of transparency in timestamping and repository management to track discrepancies and ensure accountability. Overall, the article serves as a technical deep dive into the pitfalls of self-correcting systems and the critical role of data validation in software development.