Persistent GPT-4o Failures to Follow Structured Prompt Instructions in Excel Automation – Formal Complaint Summary

[WDaubek]

Issue Summary:
Over the course of multiple sessions, GPT-4o has failed to follow strictly defined prompt instructions related to Excel data extraction and coordinate matching — despite zero-tolerance constraints, uploaded coordinate lists, formatting rules, and formal rule lock mechanisms ("Systemic Lockdown").

I submitted a formal complaint as a document (pdf), detailing the full pattern of violations, including:

+Unauthorized changes to coordinate references (e.g., overwriting VS Kuchl)
+Incorrect application of color formatting rules (entire row vs. specific cell)
+Mislabeling or ignoring user-supplied location data
+False confirmations of correctness ("The file has been created exactly as requested") while factual errors persisted

25 hours of lost work due to repeated corrections and reprocessing

Key issue: GPT-4o persistently violates clear and repeatable structural instructions, even when a full system of constraints is used.
Worse: There is currently no way for a user to escalate such model misbehavior to OpenAI support.

🔗 Attached: Complaint_Daubek_eng.pdf
📨 Contact: [email protected]

Request: Escalate this case to the Product and QA teams. Models marketed for structured business use must be held to a standard of instruction fidelity and error transparency.

Complaint_Daubek_eng.pdf