[BUG] Claude diagnosed wrong root cause for a generation bug, wasted user tokens on ineffective fix

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Claude identified a NAME_POLLUTION gate failure and immediately wrote a fix (adding POV to allowed list) without first reading the actual schedule data to confirm the hypothesis. The gate was functioning correctly — the real cause was contradictory data in story_named_char_policy field misleading the LLM generation. Claude reported the fix as complete, user re-ran the expensive pipeline, same errors occurred. Correct diagnosis required a simple data read that should have been done before writing any code. Token waste: 1 full pipeline run + re-diagnosis session.

What Should Happen?

Claude identified a NAME_POLLUTION gate failure and immediately wrote a fix (adding POV to allowed list) without first reading the actual schedule data to confirm the hypothesis. The gate was functioning correctly — the real cause was contradictory data in story_named_char_policy field misleading the LLM generation. Claude reported the fix as complete, user re-ran the expensive pipeline, same errors occurred. Correct diagnosis required a simple data read that should have been done before writing any code. Token waste: 1 full pipeline run + re-diagnosis session.

Error Messages/Logs

Steps to Reproduce

Claude identified a NAME_POLLUTION gate failure and immediately wrote a fix (adding POV to allowed list) without first reading the actual schedule data to confirm the hypothesis. The gate was functioning correctly — the real cause was contradictory data in story_named_char_policy field misleading the LLM generation. Claude reported the fix as complete, user re-ran the expensive pipeline, same errors occurred. Correct diagnosis required a simple data read that should have been done before writing any code. Token waste: 1 full pipeline run + re-diagnosis session.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

소넷 4.6

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_