[Schroedingersat]

Your paper shows a very nearly monotonic increase in costs with no dip at any pont between the peak of construction and when the reactors started during that peak were all finished. It also shows that in the US at least, the negative learning rate was much steeper between the end of the turnkey error and both TMI and the construction peak. It's also not normalised for availability factor or upgrades/reliability retrofits.

You can't just post evidence of the opposite to your claim and then repeat it.