*Yes, I understand it's inflation adjusted. There are so many possible explanations for the observed negative curve that go beyond the bold, broad claim that learning curve theory doesn't hold in nuclear.

In my mind, an (at least) equally reasonable explanation is that the conditions for the learning curve weren't met. (This probably sounds like "no true Scotsman". I admit that the learning curve is a function of scale and relative to mass-production examples, the "signal" for the learning curve is probably weaker to begin with given how many facilities of the same design were actually built.)

-Changes in design pull you backward on the curve. There were lots of changes in French design

-Unsteady expansion timeline messes with the workforce expertise part of the hypothesis. You want ideally an accelerated or at least constant build rate, not large gaps where the workforce either respecializes in another field or retires.

- regulations increase over time. Part of the conditions for the theory are implicitly "all else being equal".

-while inflation adjustment partially accounts for this, labor becomes more expensive as gdp per capital increases (see, for example, low skill manufacturing leaving China as it becomes wealthier). I don't know the details, but given the rapid post-war growth, I'm guessing gdp/capital was growing pretty quickly during the French build out

For relatively low volume manufacturing, the learning curve effects are probably smaller to begin with, so it's easier to get an effective negative learning rate. With so many confounding factors that violate the premise of theory, I find it rather unscientific to definitively claim the theory is just wrong in an entire industry.