This feedback is inspirational, Kudasai. I'll get right back to it, but first a few things from before:
Foreign Art - maybe as it is but giving $3 would be an interesting card as either late game or engine payload? The original intent was very strong payload with self divert as a setback so it plays well every other shuffle through the deck. So here's another go:
Dairy Farm - Action, $5 cost.
+2 Cards
+1 Buy
+ $2
At Clean-up, instead of discarding this, take Diverted and put this on it.
Now it draws cards and is a very strong bonus. But if you do end up drawing the whole deck, getting Your Dairy Farms at the end of the draw will see some of them not much more than Woodcutters, so they play less effectively. Instant self Divert can lead to multiple plays of the same card with enough cycling, so it needs this funny wording to avoid that.
Diversion - It wasn't used very often, since handing out Diverted as an Attack can be rather mild for the cost of a terminal Action. But, I twigged that it was a one-off Attack, and so:
Lose one Treasure, put another in. This set wants a few Treasures in it, and hopefully this one's interesting.
Diverted doesn't look so good a name now so much as Delayed.
Back to Kudasai:
Power - why not indeed? Cards would be simpler to read, and just by thinking of Power tokens on a mat, there's a feeling straight away that a card or two could care for the amount of Power you have (like Gazbag's Barbarian/Warlord in Ice Age with Coffers, but it feels more fitting to me to do it with Action tokens, you can sacrifice playing Actions now for doing bigger things later). And that would give some reason for both Power and Villages in the same game.
Components' Silver gain being optional makes interesting choices with Mule whilst retaining its individual function. With your other suggestions I'm seeing how opening with 2 of them won't be so broken as with 2 Treasure Maps, which was my worry.
Actually using Prototype's self-discard as something you want to do rather than just because you can do self-discard safely makes so much sense. Honestly, that's all my thinking was here.
Your wording for Patent is definitely better, but I think it needs to 'be the card until it leaves play' like Overlord? It's going to be wordy, and I see no way round it. Short of putting the bottom instructions on the Patent mat itself or something.
You question Steam Engine's place in the series; I put it there for the contrast with Mule at first, then Patent joined in because some games would be a pretty sad race for the standout best Patent, and Steam Engine could be a fine consolation prize for the losers. Patent is a lot more flexible now than what it used to be though.
But, I put together your thoughts on Prototype and the idea of Power count and came up with this:
Steam Engine - Reveal a number of cards from the top of your deck equal to your Power. Discard the non-Actions, then play the Actions in any order.
Prototype would give +2 Power, which could either go to playing better Patents or be stored up for this. Is this an improvement?
Spinning Mule did have a Buy on it but it was dominating when you could get it early. So, taking the Buy off this and Patent I was toying with this
new idea:
Replica - Action Duration Machine, 0* cost.
When you play your first Replica this turn, if you have 3 or more Power, gain a Machine from its pile.
At the start of each of your turns for the rest of the game: +1 Buy and + $1.
(This stays in play. This is not in the Supply.)
If you want Buys, wait a bit longer. I put the $ with it to make it feel more worthwhile, and last minute made it count Power to be a bit harder to continue.