Projects

LAPSE

Memory consolidation flattens the temporal shape of user facts: an ongoing activity is stored as a standing one.

#ConversationalMemory#Evaluation

Overview

Long-term memory systems turn conversations into short stored notes. A note can keep a user fact while losing the evidence of whether the fact still holds: “I am driving a Peugeot” becomes “The user drives a Peugeot”, and the cue that the activity is ongoing is gone. LAPSE is a benchmark of matched user statements that differ only in temporal form, built to measure this.

What we found

Memory writers flatten aspect selectively. Three writer models flattened the progressive statement but kept its simple-present match in 244 of 381 pairs, and never the reverse. The asymmetry holds in all 11 model configurations tested and in the installed pipelines mem0, Graphiti, and Letta.

The lost cue matters to later readers. In exploratory tests, changing only the stored verb shifted all three readers’ estimates that a fact still holds, and when readers could ask the user before acting, two of three acted without asking more often on flattened notes.

Papers

Memory Consolidation Flattens the Temporal Shape of User Facts S. Panthi, M. Yeamin, S. Luo, R. Abdelfattah · arXiv preprint · 2026 [LINK]

Write-ups

How to Create a Simple Synthetic Dataset How the LAPSE benchmark was built: small probes first, minimal pairs, checked invented names, and a smoke test read by hand. Memory Consolidation Turns Temporary Statements Into Standing Facts A LAPSE pilot: memory writers rewrite "I am staying in" as "lives in", and date stamps barely change it.