FDAP-artifact v1.0: Where Does a Learned Task Profiler Help? A Controlled Boundary Study of Federated Drift-Aware Task Profiling

Artifact for the measurement-study submission "Where Does a Learned Task Profiler Help? A Controlled Boundary Study of Federated Drift-Aware Task Profiling". Contains every experiment and build script (seeds fixed in code), the raw result files, the parsed Alibaba PAI GPU production trace (732,691 tasks with measured four-channel resource intensity), the family (k-means) assignment, both built PDFs (manuscript and supplementary) and an md5 manifest mapping each table and figure to its script. v2.4 re-measures all wall-clock columns one process at a time (concurrent runs roughly double them), releases the per-block errors behind Table XII so every interval is recomputable, adds a moving-block bootstrap that preserves the serial dependence of the block deltas, adds the matched-information-set ablation of Section VI-K, and renders the supplementary material from the same result files as the manuscript. v2.3 corrected the placement objective: its time term is now the remaining time normalised by the shortest competing remaining time, so the effective objective no longer drifts with the campaign clock. Every schedule-dependent number was regenerated. Environment: CPython 3.13.14, numpy 2.5.4, pandas 3.1.1, scipy 1.18.1, MiKTeX for the PDFs. See README.md for the regeneration pipeline.