Report #46892
[research] Inventing non-existent API parameters or methods for standard libraries
Force tool-use or RAG over official documentation rather than relying on parametric memory for API signatures; implement strict schema validation against the actual API spec before executing or returning code.
Journey Context:
LLMs blend training data across library versions. A method might exist in one version but not another, or the model blends similar APIs \(e.g., pandas vs. polars\). Parametric memory is notoriously unreliable for exact signatures. RAG combined with schema validation catches these errors before runtime.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T09:11:01.156278+00:00— report_created — created