Facts as First Class Objects: Knowledge Objects for Persistent LLM Memory

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zahn, Oliver, Chana, Simran
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914406244286464
author Zahn, Oliver
Chana, Simran
author_facet Zahn, Oliver
Chana, Simran
contents Large language models increasingly serve as persistent knowledge workers, with in-context memory - facts stored in the prompt - as the default strategy. We benchmark in-context memory against Knowledge Objects (KOs), discrete hash-addressed tuples with O(1) retrieval. Within the context window, Claude Sonnet 4.5 achieves 100% exact-match accuracy from 10 to 7,000 facts (97.5% of its 200K window). However, production deployment reveals three failure modes: capacity limits (prompts overflow at 8,000 facts), compaction loss (summarization destroys 60% of facts), and goal drift (cascading compaction erodes 54% of project constraints while the model continues with full confidence). KOs achieve 100% accuracy across all conditions at 252x lower cost. On multi-hop reasoning, KOs reach 78.9% versus 31.6% for in-context. Cross-model replication across four frontier models confirms compaction loss is architectural, not model-specific. We additionally show that embedding retrieval fails on adversarial facts (20% precision at 1) and that neural memory (Titans) stores facts but fails to retrieve them on demand. We introduce density-adaptive retrieval as a switching mechanism and release the benchmark suite.
format Preprint
id arxiv_https___arxiv_org_abs_2603_17781
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Facts as First Class Objects: Knowledge Objects for Persistent LLM Memory
Zahn, Oliver
Chana, Simran
Artificial Intelligence
Large language models increasingly serve as persistent knowledge workers, with in-context memory - facts stored in the prompt - as the default strategy. We benchmark in-context memory against Knowledge Objects (KOs), discrete hash-addressed tuples with O(1) retrieval. Within the context window, Claude Sonnet 4.5 achieves 100% exact-match accuracy from 10 to 7,000 facts (97.5% of its 200K window). However, production deployment reveals three failure modes: capacity limits (prompts overflow at 8,000 facts), compaction loss (summarization destroys 60% of facts), and goal drift (cascading compaction erodes 54% of project constraints while the model continues with full confidence). KOs achieve 100% accuracy across all conditions at 252x lower cost. On multi-hop reasoning, KOs reach 78.9% versus 31.6% for in-context. Cross-model replication across four frontier models confirms compaction loss is architectural, not model-specific. We additionally show that embedding retrieval fails on adversarial facts (20% precision at 1) and that neural memory (Titans) stores facts but fails to retrieve them on demand. We introduce density-adaptive retrieval as a switching mechanism and release the benchmark suite.
title Facts as First Class Objects: Knowledge Objects for Persistent LLM Memory
topic Artificial Intelligence
url https://arxiv.org/abs/2603.17781