The variation gap traces to what the generator was disclosed: no released statistic constrains which band and which week are zero together¶

Claim · All claims
Evidence · 1 study (2 entries) · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? The differentially private summary for the study-habit records contains no statistic constraining joint band-by-week zeros, and the entropy over weeks is invariant to permuting them. → Inoue 2026 q2 i? The excess of all-six-band zeros over what independent bands would give exceeds one in every real cohort (1.40 to 2.57) while synthetic values fall below one in two cohorts of four. → Inoue 2026 (2)

Evidence¶

Inoue 2026¶

Inoue, H., & Yasutake, K. (2026). What Fidelity Metrics Miss: A Structural Check on Synthetic Educational Data. Proceedings of the 34th International Conference on Computers in Education. Asia-Pacific Society for Computers in Education. https://arxiv.org/abs/2609.27265 doi:10.48550/arxiv.2609.27265

q2 · i? · design · r2

Analysis of the platform's disclosure: the private summary comprises the proportion of zeros across the 108 band-by-week combinations and two entropies, one marginalising over weeks and one over bands, so week identity and band-week co-occurrence are unconstrained.

"No released statistic constrains which band and which week are zero together."

Inoue 2026 (2)¶

Inoue, H., & Yasutake, K. (2026). What Fidelity Metrics Miss: A Structural Check on Synthetic Educational Data. Proceedings of the 34th International Conference on Computers in Education. Asia-Pacific Society for Computers in Education. https://arxiv.org/abs/2609.27265. doi:10.48550/arxiv.2609.27265

q2 · i? · design · r2

Comparison of the all-six-band zero co-occurrence ratio: the marginal zero rate is reproduced to within 3% to 13%, but the co-occurrence excess, which no released statistic constrains, differs between real and synthetic cohorts.

"The excess of all-six-band zeros over what independent bands would give is not: it exceeds one in every real cohort, from 1.40 to 2.57, while the synthetic values run from 0.41 to 2.15 and fall below one in two cohorts of four."

Discussion¶

Related Claims¶

• In a control experiment on data of known construction, the check is calibrated, can be passed, and separability from marginals is not fidelity — related
• Synthetic cohorts place week-to-week change in the term's examination weeks, which real cohorts do not — related
• Partition descriptors and the component-membership–study-volume association are reproduced by the synthetic cohorts — related
• Synthetic cohorts reproduce the level and shape of the weekly β0 partition but vary 2.6–4.9 times less across the term than real cohorts at a common working point — related
• Real cohorts separate from marginal-preserving surrogates of themselves in all four years, while three of four synthetic cohorts do not — related