Model repeatedly fails to fix issue, ignores user guidance, never verifies own work

Behavioral failure report

Summary

The model failed the same task 5+ times in a row, ignoring clear user guidance each time, never verifying its own output, and presenting broken fixes as complete. The user had to repeat the exact same instruction multiple times with increasing frustration.

Failure patterns observed

1. Never verifies its own work
The model made changes and immediately told the user "this should work now" without ever checking whether the fix actually worked. No console.log, no inspecting API responses, no testing the actual data flow. Pure assumption-based development.

2. Ignores the user's explicit guidance
The user repeatedly said "do it the same way as [working feature]" and pointed to a working reference implementation in the same codebase. Instead of copying the exact pattern that works, the model kept inventing its own approaches that didn't work.

3. Layers fixes on broken foundations
When attempt 1 failed, the model didn't investigate WHY it failed. It just added another layer on top. Then another. Five layers of workarounds stacked on an unverified assumption, instead of stepping back after the first failure to understand the root cause.

4. Makes assumptions about runtime data without checking
The model assumed data types, object shapes, and populate behavior without ever looking at the actual data. Every fix was based on what the model THOUGHT the data looked like, not what it actually was.

5. Wastes the user's time with false confidence
After each broken fix, the model confidently explained why it would work now. This is worse than saying "I'm not sure" because it wastes the user's time testing something that was never going to work.

6. Cannot recognize when it's stuck
After 3 failed attempts using variations of the same approach, the model should have recognized it was stuck and fundamentally changed strategy. Instead it kept making minor tweaks to the same broken approach.

What should have happened

• After the FIRST failure: investigate why it didn't work by examining actual data
• After the user says "do it like X": literally copy the pattern from X
• Before saying "fixed": verify the fix actually works
• After 2 failed attempts: completely change approach instead of tweaking

Impact

Severe loss of user trust. The user had to repeat themselves 5+ times for a straightforward issue that should have been fixed in one attempt.