The Olist and WindowDash simulation inputs are matched in scale and analytical scope, with overlapping distributions but differing duration-rating associations¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Both datasets contain 15,000 unique orders and nine reference patterns; ratings concentrate at four or five stars in both (77.6% in Olist, 79.5% in WindowDash), cumulative-distribution maxima are 0.084 for ratings, 0.127 for prices, and 0.201 for durations, and pooled Spearman correlations are −0.029/−0.006 for price-rating and −0.222/−0.010 for duration-rating. → Bang An 2026

Evidence¶

Bang An 2026¶

Bang An, Maria Hamdani, Joseph Fox. (2026). DataCanvas-EDU: An Agentic Framework for Instructor-Guided Synthetic Data Generation in Business Analytics Education. https://arxiv.org/abs/2609.19617 doi:10.48550/arxiv.2609.19617

q2 · i? · design · r2

Descriptive distributional comparison of the two simulation inputs (Figure 8), using available observations and median-normalized prices and durations. The paper reports maximum empirical cumulative distribution differences of 0.084 (ratings), 0.127 (prices), and 0.201 (durations), and pooled Spearman correlations of −0.029 and −0.006 for price-rating versus −0.222 and −0.010 for duration-rating.

"Ratings concentrate at four or five stars in both cases: 77.6% in Olist and 79.5% in WindowDash (Figure 8)."

Discussion¶

Related Claims¶

• AI agents recover a smaller share of reference patterns from the synthetic WindowDash dataset than from the real-world Olist dataset — related
• Designed patterns can rest on very small supporting subgroups: only three orders meet the Price-Rating Sensitivity conditions, with a mean rating of 2.0 — related