Geopolymer and alkali-activated binder mix-design database: 3,648 designs extracted from 288 peer-reviewed publications
Description
A database of 3,648 unique geopolymer and alkali-activated binder mix designs compiled from 288 peer-reviewed publications (2014-2026). Each row is one distinct design as reported in one source, with every populated cell traceable to a publication that states it. The 61 fields cover precursor identity and oxide composition; activator chemistry (NaOH molarity, silicate modulus Ms, Na2O content as a percentage of binder, activator-to-binder and silicate-to-hydroxide ratios); water-to-binder ratio and aggregate content; curing regime, temperature and duration; mechanical properties (compressive, flexural and splitting-tensile strength, elastic modulus) with their test ages; fresh-state properties (slump, flow diameter, initial and final setting time); and durability outcomes including drying shrinkage, water absorption, sorptivity, elevated-temperature residual strength and chemical resistance. Binder chemistry is partitioned into low-calcium N-A-S-H (1,005 designs), hybrid C-N-A-S-H (1,129) and high-calcium C-A-S-H (1,210) classes by measured bulk CaO content rather than by author label, with 304 designs left unclassified where the source does not support an assignment. The underlying CaO value is published so users can re-partition at their own thresholds. Multi-age strength results are carried inside the row as a series (44.8% of designs) rather than as duplicate rows, so the row count reflects distinct mix designs and not the number of measurements. A field is populated only where the cited source states the value, or where it follows from source-stated values by a documented derivation. A blank means the source did not report that quantity: it is never imputed, estimated or replaced by a typical value, so field completeness measures reporting practice in the alkali-activated materials literature rather than data quality. 99.4% of designs carry a verbatim source excerpt supporting the reported property, with a flag recording whether that excerpt contains the value itself; 3,032 designs report compressive strength. Released with a field-by-field data dictionary giving the type, measured completeness and vocabulary tokens of every column; a controlled-vocabulary reference documenting the binder-classification rule and the permitted and forbidden derivations; and a source corpus giving venue, year, first-author country, open-access status and row contribution for each of the 288 contributing publications. Intended for mix-design selection, meta-analysis across studies, identifying under-reported regions of the design space, and as a training or benchmarking set for property-prediction models. Users modelling from these data should cluster by source publication, since designs from one study are not independent replicates.