Data and code for "Paired curing evaluation connects improvement opportunities with realized gains and batch losses in ultra-high-performance concrete"

Supporting data and analysis evidence for a paired offline evaluation of curing-package selection in fixed-recipe UHPC. The study reuses the public DaKS version-2 dataset (10.48662/daks-56.2). It retains 50 authenticated same-batch pairs, 49 complete pairs, 45 unflagged primary evaluation pairs, four separately audited complete author-flagged pairs, one incomplete pair retained as NA, and 44 historical batches used only for training. The record includes the frozen protocol and configuration, original analysis source, grouped splits, saved predictions and choices, fit logs, all policy results and prespecified contrasts, and labelled verification/packaging utilities. Nominal target events are classifications of reported specimen means relative to 120 MPa, not guarantees of reliable attainment. The package preserves known NOT_SAVED and NOT_RUN boundaries; availability of scripts and a checker does not itself establish an independent full numerical refit. Original experimental attribution and source licensing are retained. Version 1.0.1 — documentation correction (30 October 2026). The deposit-status and licensing statements inside the package were aligned with the published record: new study results and documentation are released under CC BY 4.0 (rights holders: Yuanqiu Zheng, Xiaoqin Zhao, Wei Li, and Xinyu Xiao), and study-owned code is released under the MIT License (Copyright (c) 2026 Yuanqiu Zheng). Original DaKS source data retain CC BY 4.0. No data, code logic, predictions, choices, or results were changed; the frozen scientific payload is byte-identical to version 1.0.0.