RAFT retrieves at the timeline-entry level so a troubleshooting agent matches your current state, not the whole closed case

arXiv 2609.20754·low signalXBlueskyLinkedInCopy link

Standard RAG treats a support case as a static document, which ignores that cases are multi-stage and stateful. RAFT abstracts each closed case into a directed chain of timeline entries, retrieves at entry level, and returns the parent case trajectory anchored at the matched intermediate state, with an optional case-level graph linking cases by a configurable similarity representation. It beats vanilla RAG and GraphRAG on Case Hit at every stage of case progress, evaluated on a synthetic benchmark from Microsoft Learn Windows Server docs plus real Apache Jira issues with human duplicate labels; the benchmark, implementation and Jira eval set are released. Single-paper, vendor-adjacent evaluation, so treat the margins as directional.