Agentic Table Benchmark Reports
1 min read
Purpose
This hub holds the approved benchmark reports in the Product space. Each report mirrors the customer-safe publication body and adds a collapsed internal evidence and maintenance appendix.
Latest benchmark snapshot
Capability | Published headline | Scope | Maintenance use |
|---|---|---|---|
Q&A retrieval | 98.4% answer accuracy; 92.5% answer coverage; 98.4% safe refusal | 325 scored cases | Answer-library methodology, evidence lineage, and rerun governance |
Question extraction | 90.0% accepted on the standard route | 23 documents: 18 standard, 3 fallback, 2 failed | Extraction ladder, hybrid grading, evidence lineage, and update controls |
Published reports
Operational benchmark documentation
Use the established internal pages for test design, setup, execution, and report interpretation. Keep this hub focused on stable publication copies.
Overall Benchmarking Goals
Use case 1 — lookup verbatim
Benchmarking Agentic Table Responses
Update policy
Use stable page titles and update measurements in place. Keep dated PDFs in the benchmark SharePoint libraries and rely on Confluence version history for page-level publication history.