Code and derived data for: An Audited Open Record of Three Deep-Blue TADF Host Materials with Benchmark Tasks for Machine Learning

The per-file manifest of the raw record (B_MANIFEST_raw_files.csv, with the experiment record and new file name of every raw file), 116 processed tables, four benchmark tasks with out-of-fold predictions and metrics, technical validation, label inventory and the code. The raw files themselves are in the experiment records. Methods: Scripts are written in Python 3 (NumPy, SciPy, pandas, matplotlib; RDKit for structure drawings where used). Each script reads the tables of this record or the raw files of the experiment records and writes numerical tables and figure panels (vector PDF with a 600-dpi PNG). Inventory and manifest of the raw record built from the laboratory folders (B01-B06: SHA-256 de-duplication, header parsing, path conditions); feature extraction (B10-B13); benchmark tasks T1-T4 with nested cross-validation and group-bootstrap uncertainty (B20); technical-validation and article figures (B21-B28, B33); label harmonisation (B34). Companion paper: Manifest, processed tables, benchmark tasks and code for the measured record of three phosphine-oxide deep-blue TADF host materials (S1, S3, S4), version 1.0 (manuscript prepared for Journal of Chemical Information and Modeling). Raw instrument files are in the experiment records of the series (AFM, UV-Vis, DRS, ellipsometry, PL, time-resolved PL, CV, DSC/TGA, XRD, FTIR, NMR/MS, electron microscopy, J-V, impedance, DFT). Licence: CC BY 4.0 for data and documentation; MIT for code (LICENSE files in the code folders). Version 2.2 (5 November 2026): current figure and TOC/GA scripts and result tables of the journal version; README names the instrument of the 2022 PL series; data and results unchanged.