I’m building an open map of literature values for MXenes (and MOF/COF surface areas). Every value links to its source sentence: MXene Atlas — capacitance · conductivity · shielding
Values are AI-extracted and checked by two AI judges. No domain expert has checked them yet.
Three things we found:
- A number being in the paper doesn’t make it plausible. One MOF composite paper reports “about 37,500 m²/g”, far above the ~7,800 m²/g record (DUT-60). Both AI judges accepted it.
- Two-electrode device capacitances get compared with three-electrode values. 118 of 930 gravimetric capacitance values on our map come from two-electrode cells; the map now labels them and can hide them.
- “BET” and “pore volume” columns mix different things (BET from CO₂ isotherms, areas with no model named, micropore vs total volume). We now keep them apart.
Accuracy so far: against an outside hand-made answer key in another field (LitXAlloy, 19 alloy papers: [2604.07649] LitXBench: A Benchmark for Extracting Experiments from Scientific Literature ), 66 of our 67 yield-strength values matched. For the MXene values there is no expert check yet.
Two questions:
- If you spot a wrong value, what would make it easy to tell us?
- Should two-electrode values be hidden by default?