Excel is an ancient and complex beast. I get the appeal of building this project---it sounds fun---but trying to duplicate the Excel engine to the level of producing identical outputs is, frankly, bonkers. The author caught the one discrepancy they noticed, circular reference handling, but how many did they miss? How do they know different inputs won't cause it to deviate from Excel? I didn't get the sense from the blog post that this had extensive test coverage. Putting it into use for a business-critical financial calculation is a massive risk, but I guess that's how Uber rolls ¯\_(ツ)_/¯
It would have been less fun but way, way less risky to wire a headless Excel up to a javascript front-end.
No Windows, no Excel license, hundreds of concurrent users. How confident are you that each user is getting the calculations that they triggered and not someone else's? How are you deploying that spreadsheet? How are you version controlling the mutable parts of the system? How confident are you that the solution would be robust in the face of bad inputs and high load?
I won't say there wasn't risk, but there was quite a bit of testing and a human always made the final call anyway (I never fully understood why we didn't eliminate humans from the processes altogether).
It would have been less fun but way, way less risky to wire a headless Excel up to a javascript front-end.