We've gone through a bunch of iterations. One of those iterations was canvas, but we care a lot about the user experience, and found getting the frame times we wanted on huge workbooks wasn't practical, so we graduated to webgpu/webgl, and that was better, but we kept hitting bugs in the web rendering engines that we couldn't tolerate, so now we've ended up going down to the metal and writing an engine on top of metal/vulkan/opengl (we still have a webgpu/webgl backend, but we're not using it at the moment).
You didn't ask a specific question, but it basically comes down to getting really good at calculating layout fast, getting good at figuring out how to incrementally maintain your rendered view, then getting really good at figuring out how to optimally use the hardware you're targeting.
You didn't ask a specific question, but it basically comes down to getting really good at calculating layout fast, getting good at figuring out how to incrementally maintain your rendered view, then getting really good at figuring out how to optimally use the hardware you're targeting.