One thing that I neglected to mention is that since the SRI hashes are calculated, previews can be cached by peers other than the originating host - since they are verifiable, anyone can pass them along and the client can verify the payload.

Not something I really fleshed out in discovery, but if there is interest, that might make it more production ready.

Okay - it would have been nice to learn where the proposal fails, but I understand that I can’t access your expertise for free. Take it easy!

It really feels like you didn’t read my post. I specifically called out caching as a concern, and even if we see more requests, they will be cheap to serve.

Removed by mod

Removed by mod

I’d be curious to know what you think the problem with the ml is? Nothing, or just that it isn’t as good as the online services?

PS: Mozilla themselves argue that this on-device translator is good for privacy. Just sayin’.