Data and code for: Continuous property-weighted adaptation with retained domain support for aqueous redox molecular generation

This dataset provides the v66 paper-used data, original code, model checkpoints, frozen predictors and selected archived quantum-chemical records for Continuous property-weighted adaptation with retained domain support for aqueous redox molecular generation. The current comparison includes 16 paired fine-tuning seeds per backbone, 96 runs, 864 decoding cells and 864000 stored attempts across LSTM, TCN and TCN–LSTM. Hard10 and Continuous10 share the backbone-specific pretrained checkpoint, omit ChEMBL replay and use final epoch 10 after 8000 updates. Three prespecified primary tests use Bonferroni correction; only the TCN–LSTM contrast meets the recorded primary criterion. Other endpoints are descriptive. The archive retains the existing predictor cross-validation, replay comparisons and controls, 64-setting sensitivity sweep, earlier architecture/likelihood/motif diagnostics, the separate epoch-2-selected LSTM baseline, and the 50-candidate/8-calibration-molecule DFT records already used in v66. These retain separate protocols and are not pooled with the current comparison. The earlier three-seed no-replay pilot and unrelated campaigns are excluded. Four ZIPs hold the complete licensed local snapshot. File SHA256 hashes, exact source/protocol manifests, editable figures, 26 table CSVs, selected current and historical generator checkpoints and 174 existing Gaussian jobs support inspection and reproduction from stored evidence. Preparation introduced no training, generation, prediction, DFT or new hypothesis test. Licenses are component-specific: MIT for author-owned source code; CC BY 4.0 for author documentation, independent numerical annotations and artwork; CC BY-SA 3.0 for ChEMBL samples, generator weights and generated structure contributions; CC0 retained for original RedDB content. Mixed files are mapped in LICENSE_SCOPE.csv. The ChEMBL source release was not recorded; actual input hashes and preprocessing provenance are supplied. This work was supported by the Korea Institute of Energy Technology Evaluation and Planning (KETEP), and the Ministry of Trade, Industry & Energy (MOTIE) of the Republic of Korea (No. 20224000000020). This work was also supported by the Ministry of Science and ICT (MSIT), Korea, under the ICT Challenge and Advanced Network of HRD (ICAN) support program (IITP-2024-RS-2024-00437186), supervised by the Institute for Information & Communications Technology Planning & Evaluation (IITP).