What “not in the databases” does and doesn’t mean

When we say a candidate compound is “not in the Materials Project or COD”, we mean exactly that: two public databases have no entry for it. The Materials Project (MP) holds computed crystal structures. The Crystallography Open Database (COD) holds structures measured by experiment.

A starting point, not a novelty claim

It does not mean the compound is new. Other structure databases, including paid ones we did not search, as well as papers and patents, can describe it. Lithium hafnate is an example: COD has no entry for it, yet published papers describe lab-made samples. Until those sources are searched, “not in MP or COD” means no synthesis record in the sources we checked, and nothing more.

The reverse is much stronger

If COD holds a measured structure, someone has made the compound. Of the 371 entries on our internal candidate list, 116 have an experimental structure in COD, so we do not treat any of them as a new compound. Another 85 are not in COD; 16 compounds we generated ourselves are in neither MP nor COD; and 154 have not been checked, mostly because they are families of compounds, mixtures or processes rather than one formula.

COD keeps our calculations honest

Machine-learned stability checks sometimes call a real, measured compound unstable. Of 62 compounds that have a structure in COD and a recomputed phonon verdict, 4 were called unstable. Each case has three possible explanations: the model is wrong about a compound someone has made; COD holds a different crystal form of the same compound; or the measured form is stable only at higher temperatures, which these calculations leave out. So a machine-learned “unstable” is evidence, not proof.