Sadly nothing in Scott's blog post about how they obtained the source. Was it still in Microsoft's archives? Did they happen upon some tractor-feed print-outs they had to type in by hand?
I assume today typing in by hand is no longer needed, with text parsing from images being table stakes for LLMs.
It would also be interesting why it was open-sourced now. I assume if they had done the same last year, the resulting loss of revenue would not have destroyed the plucky little $3T upstart.