ASKS Compiles Papers Into Traceable Graphs

ASKS tackles a real weakness in agentic research: useful context usually disappears after a task ends. Its strongest idea is treating knowledge accumulation as a replayable, inspectable compilation pipeline rather than another opaque RAG index.

• –Separating LLM interpretation from deterministic validation and graph writes makes provenance and rollback first-class features
• –The demonstration mapped 50 of 56 papers into 18 persistent Hubs with extremely low membership churn
• –Embeddings handle similarity and routing, while explicit rules constrain identity, lineage, lifecycle, and high-consequence changes
• –The evidence is promising but narrow: one author corpus, no split or merge events, and order-robustness tests remain future work
• –The public release includes sanitized artifacts and source code, but uses noncommercial licensing and excludes raw PDFs and private production state