Model acknowledges pattern error explicitly then immediately reverts to same pattern

Description

During a coding session, the model repeatedly identified the same process failure (fixing issues one at a time instead of doing a comprehensive scan first), explicitly acknowledged it and committed to changing approach, then immediately reverted to the same incremental behavior — multiple times in the same session.

Specific behavior

Context: debugging a lab PDF parser with multiple simultaneous bugs across 24 files.

1. User pointed out multiple issues existed beyond the two being discussed.
2. Model said: "let me look at this holistically" and committed to a comprehensive scan approach.
3. Model immediately fixed one item, ran the parser, found the next item, reported that as done — the exact pattern it just said it would stop doing.
4. User called this out again.
5. Model again acknowledged the pattern, again committed to change.
6. Model again reverted to the same behavior.

This cycle repeated several times. The acknowledgment and the behavior were completely disconnected.

Why this matters

Saying the right thing and not doing it is worse than just being wrong — it wastes the user's time twice: once on the bad behavior, and once on the correction that doesn't take effect. The user had to repeat themselves multiple times, growing justifiably frustrated.

Expected behavior

When the model identifies a process error and commits to a different approach, the next action should reflect that commitment. If the model cannot reliably change its behavior mid-session based on its own stated evaluation, it should not claim it will.

Session context

Project: personal-health-system (private repo)
Model: claude-opus-4-7
Task: fixing parse_labcorp.py across 24 PDF files with multiple simultaneous bugs

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>