Temporal Validity on Real Software Histories: Eliminating Stale-Fact Errors in Code-Assistant Memory over GitHub Fixes
Source
arxiv.org
Author
Neeraj Yadav
Date
Why it matters
Retrieval-based code assistants serve stale facts about a third of the time when a value changes mid-session — a renamed function, a moved endpoint — and an LLMA large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.Full definition → reranker doesn't fix it; supersession-aware memory does.
Terms in this piece · Glossary
RAG — Retrieval-augmented generation — fetching relevant documents first and pasting them into the model's context so it answers from your data instead of memory.
SWE-bench — The standard benchmark for AI coding agents: real GitHub issues from real repositories, scored by whether the agent's patch passes the project's own tests.
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
articleWhen Old Facts Return: Re-Reads, Reverts, and the Limits of Temporal MemoryNeeraj Yadav (Called It Inc.)
articleThe Memory Trust Gap: Capability-Dependent Failures in Persistent-Memory AgentsJundong Hu, Shekar Ramachandran
articleAdaRepair-Mem: Adaptive Experience Orchestration for Repository-Level Program RepairZ. C. Luo, J. C. Guo, W. J. He, S. Y. Wang, J. C. Yu, F. M. Zhao, Y. Chen, T. Cao, L. Q. Liu, N. Zheng, W. Xu, J. Jiang, Z. M. Zhao