First, THANK YOU for making a concrete proposal!

Specific comments:

So we'd get to 2MB blocks in the year 2021. I think that is much too conservative, and the most likely effect of being that conservative is that the main blockchain becomes a settlement network, affordable only for large-value transactions.

I don't think your proposal strikes the right balance between centralization of payments (a future where only people running payment hubs, big merchants, exchanges, and wallet providers settle on the blockchain) and centralization of mining.

I'll comment on using median time generally in Jorge's thread, but why does monotonically increasing matter for max block size? I can't think of a reason why a max block size of X bytes in block N followed by a max size of X-something bytes in block N+1 would cause any problems.