filter() right before displaying my data felt easier than getting the right data to begin with.

The issue here is that you don’t know what the right data is to begin with. SAP does what you’re suggesting. They demand you set the filters correctly before requesting any data, which is a terrible user experience.

Imagine you need to debug that and don’t even have an intermediate step to see whether the wrong data is arriving or whether filter/sort logic is wrong for a specific case, or whether its just the rendering.

That’s a strawman. Why would I not know what data arrives in the frontend. That’s what the network debugger is for. That’s what a breakpoint before the filter is for.

But you can’t tell me that applying filter/sort to 8MB of data in the frontend is anything but mega scuffed.

Personally, I find re-transmitting those 8MB of data for every different sorting request way worse. Remember that this isn’t even cacheable, because the data is different for different sorting requests.

Maybe we have different types of frontend and different levels of user proficiency in mind. In my case, I cannot possibly ask the user to know how they want a list sorted and filtered before seeing the list and the options. They’d throw the frontend in my face. If you have very knowledgable users that know what they want from the get-go, then it might be possible to show them a form to sort and filter and only request the data when the user sends the form.

Not to even mention how that would run on low-power devices with bad internet connection.

I don’t see how ‘bad connection’ is an argument in favor of re-requesting data just for it to be displayed in a different order. I’ve made this back-of-the-envelope calculation in another comment. For a good connection, latency is about 20ms. In that time a 1GHz processor can to 20 million operations. Take 10 operations for each comparison (to adjust for more complicated comparisons), and you can use 2 million comparisons to sort the list in the time it takes to re-fetch it. (Keep in mind that the act of rendering the HTML itself is also computationally expensive for tables). 2 million comparisons sorts a list of 120,000 entries.