SuperIntelligencefor Science
Home/AI for HPC/Recall Is Not Honesty
SI² — SuperIntelligence for Science

Recall Is Not Honesty

Recall Is Not Honesty: Benchmarking Multi-Agent Architectures for Trustworthy HPC Operational Analytics

AcceptedSC ’26 · 2026

Research question

How should multi-agent architectures be benchmarked for trustworthy HPC operational analytics?

Overview

This research examines trustworthiness in multi-agent analytics for high-performance computing operations. The work frames recall and honesty as distinct concerns in evaluating operational analytics. Detailed methods, benchmark results, and artifacts will be added with the manuscript.

Research team

Authors & affiliations.

Kartik Shinde*1, Tirthankar Ghosal*2, Sreekantam Sai Venkat1, Siddhant Senapati1, Sanjay Das2, Tilak Kasturi1, Khaled Z. Ibrahim3, Hugo Hernandez4, Kanad Basu5, Sreenivas Rangan Sukumar6

  1. Predii Inc.
  2. Oak Ridge National Laboratory
  3. Lawrence Berkeley National Laboratory
  4. spinTwo LLC
  5. Rensselaer Polytechnic Institute
  6. Advanced Micro Devices

* Equal contribution, as indicated in the supplied author list.