Why prompt memory
fails at scale

01 Context window pressure — rules compete with the task

02 Maintenance lag — the file is always behind the codebase

03 No precedence engine — conflicting rules, no resolution

04 Advisory only — the model can always ignore the context file