Earlier quoted context omitted.
If you're loading the content from the originating site, surely there's no benefit at all to signing. If you're loading the content directly from the site, the browser just needs TLS to verify the integrity of the content. And you're also back to the situation where you can't preload the content in a controlled manner or privacy-preserving manner, nor have the page-speed guarantees since the version being served to t…
> And you're also back to the situation where you can't preload the content in a controlled manner or privacy-preserving manner... A few people have pointed out the privacy-preserving aspect of AMP. I'm not sure I get how that's the case. Is this referring to the fact that the page is not being pre-loaded from the content owner's own webserver? The main privacy violators on the internet are Google and Facebook. How i…
If you make a search query, but have not clicked on any results, you have a privacy expectation that the web servers of the search results you have not clicked on will not know you performed this query, your ip address, cookie, etc. For example, if you search for [headache] and then close the window, mayoclinic.com knowing that you made this query would probably be a surprising result.
With naive preloading, you would preload a search result from that origin. Your browser would make an HTTP request to the site and that site (sending an ip address, the URL you are preloading, and any cookies you may have set on that origin). So, this approach would violate your expectation of privacy.
Instead, if the page is delivered from Google's own cache, the HTTP request goes to Google instead of the publisher. Google already knows that you have made this query, and are going to preload it (the search results page instructed your browser to do so in the first place). The request will not have any cookies in it except for Google's origin cookies, which Google already knows as well. Therefore this type of preload does not reveal anything new about you to any party, even Google.
AMP has been doing this for a long time in order to preload results before you click them. However, until Signed Exchanges the only way to do this was that on click the page would need to be from a Google owned cache URL (google.com/amp/...). With Signed Exchanges, that can be fixed. The network events are essentially the same.
Note that once the page has been clicked on, the expectation of privacy from the publisher is no longer there. The page itself can then load resources directly from the publishers origin, etc.
To your last point, if someone posts a link on twitter to an AMP page on a publisher domain, and then you click it, your browser will make a network request to the publisher's origin. Google will not be involved in this transaction in any way. If someone explicitly posts a link to an Google AMP Cache Signed Exchange, then yes this will trigger a request to Google but this will be far less likely going forward as these URLs will never be shown in a browser. For example, try loading https://amppackageexample-com.cdn.ampproject.org/wp/s/amppac... using Chrome 73 or later. This is a signed exchange from one domain being delivered from another. You'll never see that URL in the URL bar for more than a moment, so it's unlikely to ever be shared, like I'm doing now.