If you have a bunch of observations, you can fit an exponential through them. You can do that in Excel. Here, let me just quickly do that for the half life of t

If you have a bunch of observations, you can fit an exponential through them. You can do that in Excel. Here, let me just quickly do that for the half life of the buildings in Europe. ... Ok, done, it's 63 years. Should I publish an article? What conclusions does one draw? The usefulness of a model is how well it can extrapolate. 63 years means I should not find any buildings that are 2000 years old, and still the Pantheon is standing there just fine. Then what exactly is the claim? That it's possible to fit an exponential, but the exponential can't be used for inferring anything outside of my training sample? And, no you are not being rude. But since you seem to have read the article, can you give me a few pointers why you think it's a good investment of someone else's time to go ahead and read it too? And don't be shy to use Machine Learning or Statistics jargon, we are on Hacker News here, everyone understands concepts like in-sample, out-of-sample, goodness of fit, overfitting and such.