In other words, separate "curation/recommendation" from "hosting."

I've been thinking the same thing about Amazon lately: things would be a whole lot better for the experience of buying things "on Amazon" if "Amazon.com, the storefront website" and "Amazon.com, the platform+marketplace that sellers register with" were two separate businesses.

I'd love to use the latter without the former—to use some third-party "Amazon client" that talks to Amazon's marketplace backend but has its own way of searching, browsing, and recommending products, with its own curation, its own reviews, its own ratings, etc. added on top. In the end, you'd still be putting Amazon.com warehouse SKUs into your Amazon.com shopping cart, but those SKUs might be ones Amazon.com the website would never have surfaced for you no matter how much you searched.

For just one thing, someone could build a "classic Amazon" frontend that filters out SKUs coming from "Amazon Marketplace" sellers.

For another, someone could build a version that applies actual modern spam/scam-filtering techniques to the SKU database, paring it down such that any random thing ordered from the "safe" subset should actually be what it seems from the listing. (Not the approach that most benefits you if you're Amazon.com-the-website, because its listings might look empty; but certainly a niche to be filled!)