Guardado en:
Detalles Bibliográficos
Autores principales: Panthi, Sugam, Abdelfattah, Rabab
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2605.24060
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866917526342991872
author Panthi, Sugam
Abdelfattah, Rabab
author_facet Panthi, Sugam
Abdelfattah, Rabab
contents Conversational-memory systems increasingly transform dialogue history into facts, summaries, timelines, and other source-linked descendants, so a single source turn can coexist with several derived memories in the same retrieval index. This raises an underspecified evaluation question: which stored form should receive retrieval credit? We show that this scoring-target choice is often left implicit and can materially change benchmark conclusions. We present TIAP, a fixed-output audit that rescores saved ranked outputs under three targets -- Raw, Source, and Canonical -- without rerunning retrieval. On LoCoMo and LongMemEval-S, switching only the credited target changes nDCG on 83.4--94.0 percent of shared queries, flips target orderings on Mem0 and MemoryOS transfer runs, and reverses parser-density recommendations. A 1,902-case semantic audit further shows that relaxed source-linked credit is fully justified only 29.2 percent of the time, despite high rubric reliability in a validation subset. These results reveal target noninvariance: conclusions about memory architectures can silently flip with a single benchmark-design choice. Conversational-memory papers should therefore define and report the scoring target explicitly.
format Preprint
id arxiv_https___arxiv_org_abs_2605_24060
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Same Ranking, Different Winner: How Scoring Targets Shape LLM Memory Benchmarks
Panthi, Sugam
Abdelfattah, Rabab
Information Retrieval
Conversational-memory systems increasingly transform dialogue history into facts, summaries, timelines, and other source-linked descendants, so a single source turn can coexist with several derived memories in the same retrieval index. This raises an underspecified evaluation question: which stored form should receive retrieval credit? We show that this scoring-target choice is often left implicit and can materially change benchmark conclusions. We present TIAP, a fixed-output audit that rescores saved ranked outputs under three targets -- Raw, Source, and Canonical -- without rerunning retrieval. On LoCoMo and LongMemEval-S, switching only the credited target changes nDCG on 83.4--94.0 percent of shared queries, flips target orderings on Mem0 and MemoryOS transfer runs, and reverses parser-density recommendations. A 1,902-case semantic audit further shows that relaxed source-linked credit is fully justified only 29.2 percent of the time, despite high rubric reliability in a validation subset. These results reveal target noninvariance: conclusions about memory architectures can silently flip with a single benchmark-design choice. Conversational-memory papers should therefore define and report the scoring target explicitly.
title Same Ranking, Different Winner: How Scoring Targets Shape LLM Memory Benchmarks
topic Information Retrieval
url https://arxiv.org/abs/2605.24060