Pre-registration: histogram budget of coverage-aware chain ordering (FedDAC-Order) under record-level differential privacy

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Pre-registration of an experiment for the study 'Grouping or Ordering? Coverage-Aware Sequential Federated Learning under Label Skew'. Section 7 of ORDER_PLAN.md, the analysis script and the job grid were fixed at commit dde633521f3c50962ef1a61d43f51b94dc330020 of the study repository before any run of block L14; this record timestamps them. Primary hypothesis H5: at a total budget eps = 1 split as eps_h + eps_sgd, FedDAC-Order with the histogram budget eps_h selected on the validation split (from 0.05, 0.1, 0.25, 0.5) versus random grouping with random chain order (exact two-sided Wilcoxon over seeds 0-9, Holm over CIFAR-10 and MNIST), with a pre-stated reading of the deficit. Related pre-registration: 10.5281/zenodo.23722974. Code and results of the study: Zenodo DOI 10.5281/zenodo.22965988.

• Pre-registration: replication of coverage-aware chain ordering (FedDAC-Order) on CIFAR-100 with ResNet-18 and FEMNIST

  Zenodo (CERN European Organization for Nuclear Research) 2026 · 0 citations