Auditing Long-Term Memory Evaluation: Repeated Judging, Reader Variation, and Negative Controls.

arXiv:2609.38021v2 Announce Type: replace-cross Abstract: This report audits evaluation of a long-term-memory retrieval chain on the 500 LongMemEval-S development questions. Its strongest historical reader lane scores 479 and 475 under an adapted GPT-4o rubric