Release history

Benchmarks, composition ML and a fully graphed pipeline.

• FeatureMatBench standard tasks: official 5-fold splits bundled, Matflow composition baseline scores on the benchmarks page (steels, glass).
• FeatureComposition property prediction with a learned element-embedding model (with uncertainty) and a composition baseline — new Composition ML tab on Materials Search.
• FeatureMatflow candidate sampler on the Data Synthesis stage — generative flow sampling over the discretized design space, biased toward a surrogate-predicted target.
• FeatureStructure-property explorer (chemiscope) on the Atomistic stage — linked scatter plot + 3D viewer over uploaded XYZ structures with property CSV axes.
• FeatureActive learning query strategies: uncertainty sampling and query-by-committee alongside Bayesian acquisition.
• FeatureOptimization search algorithms: Matflow multi-objective search variants alongside the Matflow Adaptive Search.
• FeatureFull Pipeline rendered as an interactive node graph — click-to-jump stages, artifact edges, active-learning loop-back.

Platform redesign and the shared design system.

• FeaturePlatform redesign, blueprint architecture, and a new design system across every page.

The pipeline engine lands.

• FeaturePipeline system introduced allowing end-to-end chaining of stages.
• ImprovementTEA integration directly inside the pipeline for continuous cost evaluation.
• FeatureSynthetic data natively available as an intermediate pipeline node.

Evaluation reports, column stats and lineage.

• FeatureEvaluation tabbed report overhaul with extensive privacy and utility metrics.
• ImprovementInteractive column stats distribution sparklines.
• FeatureData flow visualization lineage graphs.

Optimization constraints and the active-learning loop.

• FeatureOptimization filters and advanced compositional constraints.
• FeatureActive learning loop support via expected information gain.
• ImprovementMulti-objective search algorithms added.

AI Copilot and tabular data enrichment.

• FeatureAI Copilot chat with RAG over platform documentation.
• FeatureData Enrichment tool for tabular lab data ingestion.

New UI foundation and the tiered access system.

• BreakingComplete redesign and new UI layout.
• FeatureTiered access system for quotas and usage caps.

TEA, virtual screening and DoE.

• FeatureTechno-Economic Analysis (TEA) module.
• FeatureVirtual screening tool for rapid in-silico triage.
• FeatureDesign of Experiments (DoE) module for Latin Hypercube and optimal designs.

Initial release of the core stages.

• FeatureInitial release: Core stages including Data Synthesis, Evaluation, Prediction, and Optimization.