Experimental data versus Arrhenius models in literature data.

[FuerstT]

Starting this thread to have a discussion on the best practice for reporting experimental data versus Arrhenius fits for literature values.

My preference would be to report the Arrhenius fit model from the paper if only plot data points in plots are given. The use of a web plot digitizer just adds additional uncertainty and potential human error. It is also difficult to check. However, if data is reported in a table or available in supporting information, those can be added to the database rather than the fit. Or if only data is presented with no Arrhenius fit in the paper.

[RemDelaporteMathurin]

I agree that the best case is when authors provide the raw data in a table, or even better in a downloadable supplementary file.
However it's far from being the common practice.

Data from WPD is less easy to check for sure. Though it's possible by simply loading the .tar file in WPD.

There are cases where the arrhenius fit doesn't match the data presented in graphs , like Fukada 2014. In this case, the fit was wrong but the data was correct (since the authors use it in a follow up publication). Using WPD was the only solution here.

About the uncertainty and human error, copying data from a table can also lead to human error since a lot of copy pastes are required. I think there's never a perfect and 100% automated solution.
The data uncertainty (not having the original data values) is small since the axes are calibrated and usually span over orders of magnitude.