Matter,
translated.
Materials science has a data problem: the databases feeding AI discovery are full of values nobody checked. This publication reads the fine print, with tools built and run by Ibtisam Ahmed Khan, a materials engineer working in data and AI.
House rule: when fixing a bug lowers my model's score, the lower number gets published.
$ audit Fe2O3
band_gap0.00 eVFLAGGEDknown 2.2 eV semiconductor
$ audit ZnO
shear_modulus-46.8 GPaFLAGGEDphysically impossible
$ audit GaAs
synthesis0 recipesVERIFIED30 refs, made by the tonne
every number comes from somewhere.
Three colors, three meanings
- DECODED
- a concept translated until it is actually usable: no jargon left standing unexplained.
- VERIFIED
- a claim I ran, with the code and the output shown, including when it embarrassed me.
- FLAGGED
- a value or belief that looks solid and is not: database artifacts, leaky models, silent assumptions.
The index
The first decode is being written. Subscribe to the RSS feedand it will find you, or follow along onLinkedIn.