SOR-Bench Workday TextQA
The first suite in the SOR-Bench family, and the standard for measuring whether AI can answer the Workday questions your admins field every day, accurately and safely enough to act on in production.
Questions
2,214
Held-out, expert-verified
Difficulty levels
3
Easy · Medium · Hard
Workday modules
31
SKUs & admin domains
The suite
How Workday TextQA was built
SOR-Bench Workday TextQA evaluates text-based question answering across 2,214 
Workday questions collected from real customer-permissioned scenarios and verified by Workday experts.
The suite spans Workday SKUs, admin domains, task types, and difficulty levels. Each item is tagged for analysis and scored against an expert reference answer.
The full corpus is held out. Public samples are representative, but not part of the scoring set.
Real-world questions
Collected from customer-approved Workday admin scenarios, not synthetic hypotheticals.
Expert-verified answers
Reviewed by Workday-experienced experts from Mando's network of 200+ consultants.
Rubric-based scoring
Scores reflect accuracy, evidence, specificity, safety, hallucination risk, and admin usefulness.
Held-out evaluation
The scoring corpus and reference answers stay private to reduce contamination and preserve value.
Coverage
Built around Workday modules run by real enterprises
Every major SKU and admin domain is represented, so a leaderboard score reflects performance on real customer surface area (Payroll, Security, Integrations, Reporting), not a generic average.
2,214 questions across 31 modules in 14 Workday product families.
Difficulty mix
Weighted toward the hard problems that eat admin time
Most questions are Hard: multi-SKU, release-sensitive, and audit-grade. A system that only nails easy lookups won't actually lighten your team's load, and the score reflects that.
Easy
Medium
Hard
Scoring
Scored the way an admin decides whether to trust an answer
It goes beyond 'sounds right': correctness, Workday specificity, evidence, and configuration safety are all graded, because an admin needs an answer they can defend in an audit and act on in a live tenant.