Trace-IN625: a source-aware multimodal database of literature images and tensile properties for additively manufactured IN625

Trace-IN625 is a source-aware, literature-derived multimodal database for additively manufactured alloy 625, commonly reported as Inconel 625 or IN625. The dataset links processing conditions, specimen or source-condition records, accepted tensile properties, microstructure evidence, figure/table provenance and licence-screened image crops extracted from peer-reviewed literature. The current release is built from a 75-paper source corpus and includes ten public CSV tables and a released image directory. It contains 103 specimen or source-condition records, 183 accepted tensile-property records from 54 papers, 19 structured microstructure records, 1028 indexed figure/table records, 305 redistributable image crops and 101 candidate multimodal image–property records. The release includes paper-level metadata, specimen/source-condition metadata, accepted tensile-property records, structured microstructure descriptors, figure/table indexes, image-availability decisions, image-crop metadata and candidate multimodal links. Stable identifiers are used to connect source papers, specimens, tensile records, figure/table entries and released image crops. The candidate multimodal records link released image crops to accepted tensile-property records when the paper-level and condition-level context supports a possible association. These records are intended for exploratory or weakly supervised image–property studies and should not be treated as confirmed gauge-volume ground-truth labels without additional source-level review. This dataset supports source-aware meta-analysis, figure discovery, database harmonisation, data-driven alloy research and carefully bounded exploratory microstructure–property modelling for additively manufactured nickel-based alloys. The associated source-aware PDF extraction skill can be used to extend the database to additional papers or other alloy systems by generating auditable candidate records with source locators, confidence scores and review flags. Expanded releases should preserve the same schema, controlled vocabulary, source-traceability rules, image-redistribution checks and validation workflow.