Cross-source confidence scoring explained
When multiple independent computational databases agree on a material's properties, that agreement is a powerful signal. This whitepaper explains the methodology behind Lattice Graph's cross-source confidence scoring.
The problem with single-source trust
Every DFT database makes approximation choices: exchange-correlation functionals, pseudopotentials, convergence criteria, and post-processing pipelines. These choices create systematic biases that are invisible within a single database. A formation energy from Materials Project and the same quantity from AFLOW may differ by 0.05-0.2 eV/atom for the same material, depending on computational settings.
Cross-source agreement as a trust signal
Lattice Graph normalizes properties across 9+ independent databases (Materials Project, AFLOW, JARVIS, OQMD, NOMAD, Alexandria, COD, and others) into a common schema. For each material-property pair, we measure the spread across sources. Tight agreement indicates robust results; wide disagreement flags properties that are sensitive to computational methodology.
Confidence scoring methodology
The confidence score combines source count (how many databases report the property), agreement (standard deviation across normalized values), provenance depth (whether raw data, derived quantities, or both are available), and recency. The score is reported on a 0-100 scale with clear thresholds: 80+ for high confidence, 50-79 for moderate, and below 50 for low confidence requiring manual review.
Practical applications
Teams use confidence scores to prioritize experimental validation, set readiness gates for workflow advancement, and build defensible evidence packages for IP filings and diligence reports. Materials with high confidence scores across multiple properties are the strongest candidates for fast-tracking through qualification pipelines.
Ready to try it yourself?
Start exploring materials data with cross-source confidence, simulation workflows, and the tools described in this resource.