Principled Context Engineering for RAG: Statistical Guarantees via Conformal Prediction

Post time 7 hour(s) ago

Title: Principled Context Engineering for RAG: Statistical Guarantees via Conformal Prediction
Authors: Debashish Chakraborty, Eugene Yang, Daniel Khashabi, Dawn Lawrie, and Kevin Duh
Publication date: 29 March 2026
Pages: 537–546
Conference proceedings: Advances in Information Retrieval: 48th European Conference on Information Retrieval (ECIR 2026), Delft, The Netherlands, March 29 – April 2, 2026, Proceedings, Part II
Publisher: Springer-Verlag
ISBN: 978-3-032-21299-3
DOI:
https://doi.org/10.1007/978-3-032-21300-6_45
ACM Digital Library link:
https://dl.acm.org/doi/10.1007/978-3-032-21300-6_45
The paper studies context engineering for Retrieval-Augmented Generation (RAG) using conformal prediction. It proposes a statistically controlled filtering framework that removes irrelevant retrieved content while preserving a specified level of coverage for supporting evidence.
The authors evaluate both embedding-based and LLM-based scoring functions on the NeuCLIR and RAGTIME collections. The approach consistently satisfies its target coverage while substantially reducing the amount of context retained. On NeuCLIR, strict filtering also improves downstream factual accuracy measured by ARGUE F1, suggesting that much of the removed context is redundant or irrelevant.
The paper currently appears to require publisher or institutional access. Could the library, or anyone with legitimate Springer access, please help me obtain the PDF through a university subscription or document-delivery service?
Thank you very much for your support.

[Other] Principled Context Engineering for RAG: Statistical Guarantees via Conformal Prediction

Reply

Same Category

Latest Reply

Announcement

Daily Top Contributors

[Other] Principled Context Engineering for RAG: Statistical Guarantees via Conformal Prediction Copy

Reply

Same Category

Latest Reply

Announcement

Daily Top Contributors

[Other] Principled Context Engineering for RAG: Statistical Guarantees via Conformal Prediction