PILLAR is a mission-knowledge capture and retrieval framework designed to make NASA data systems usable, trustworthy, and future-ready in the era of large language models. Its central premise is that expert knowledge is often clearest not as a generic description of how to use a dataset, but as a catalog of hard-won mistakes: the tempting but wrong analyses, invalid use cases, misleading shortcuts, calibration misunderstandings, silent failure modes, brittle code patterns, and community folklore that can lead users or AI systems to confident but incorrect results. PILLAR captures these negative examples together with expert correctives—what went wrong, why it fails, what assumptions were violated, and what a valid workflow should do instead—then organizes them into an authoritative, attributed, versioned record that AI systems can retrieve at the point of need. Rather than training new models or preserving static code snippets, PILLAR encodes mission-specific judgment through do/don’t patterns, use-case boundaries, MCP skills, and future-safe code-generation guidance so onboarding researchers, digital twins, autonomous geospatial systems, and AI assistants can learn from NASA’s documented expertise before repeating preventable errors.