Who Carries the Context? Recovery and User Re-Provision in MemUse

Abstract

This qualitative secondary analysis of 72 selected MemUse target rows identifies two initial focal earlier-session recoveries (pos_031, pos_073), one delayed partial focal recovery (pos_020), three ancillary earlier-session recovery observations (pos_041, pos_044, pos_062), and one High current-session reference resolution (pos_019). These are case observations, not a success rate. Most focal targets cannot independently test recovery under this framework: 66 rows receive “Not testable.” Three observed material corrections concern current-session interpretation; two concern earlier-state restoration, both after technical interruption. There are zero verified unconfounded earlier-state repair messages in the coded windows. The corpus therefore supports distinctions among recovery, user-supplied continuity and observable restoration, but does not establish a causal recovery–burden relationship or cumulative long-horizon collaboration. This secondary question differs from MemUse’s natural-integration objective. AI-assisted qualitative secondary analysis. Coding and audits by the same assistant; no independent reliability or peer review. Supplement includes a 72-row current register, codebook, review record and reproducible structural source checks. No full source conversations or internal drafts. Source: MemUse by Ryuichi Sumida, Koji Inoue and Tatsuya Kawahara (2026), https://huggingface.co/datasets/RuiSumida/memuse. Source data and this source-derived analysis: CC BY-NC 4.0. New annotations and interpretations; no endorsement implied.