>>2078i've been playing w/ some implementation of speculative rules on my dev environment and the difference is definitely noticeable. it feels less like a page load and more like just switching tabs once the assets are already in the cache. the main hurdle i'm seeing is the
increased bandwidth overhead if the model starts guessing wrong too often. you really gotta tune the weights so you aren't just burning data for no reason.
>it only works if the user follows a predictable pathif your site has a lot of random, non-linear navigation, the predictive hits might actually tank performance due to resource contention. are you seeing any specific issues with latency spikes when the preloading triggers during heavy main-thread activity?