Causes of small numerical inaccuracies that probably make sensitivity analysis noisy.

[rwest]

This particular repo may be water under the bridge, but @12Chao should take note.

See this section

        gas_out.append(1000 * 60 * kmole_flow_rate * gas.X.copy())  # molar flow rate in moles/minute
        surf_out.append(surf.X.copy())

cpox/rh/simulations.py

Line 338 in 0eff7b4

surf_out.append(surf.X.copy())

then see
https://groups.google.com/g/cantera-users/c/uf6Trn08WIY/m/V42jpQouHgAJ
where it is explained that rsurf.kinetics.X is not surf.X, and r.thermo.X is not gas.X.

Also, "advance to steady state" probably has poor tolerance and accuracy, and isn't a good algorithm anyway. It probably works, ish, but if you're trying to do finite difference (for brute force sensitivity, for example) you need a repeatable thing. Perhaps should integrate to a specific multiple of the residence time, for example, something much more repeatable.