Cool! I had thought about doing the same but in a Chrome extension that would use Llama 3.2 with WebGPU, but when I tested it, it was very slow and sometimes crashed the browser.
I code and document code and imho Claude is superior, try to tell Gpt to draw a mermaid chart to explain a code flow... the mermaid generated will have syntax errors half of the time.
Imho IDs are more explicit, and a bless when working with large codebase apps, when trying to finding the related code to an element just grab the id and search for it.
But for the average person, perception does influence how likely they are themselves to take part in corruption, and that in turn does influence the total amount of corruption in a country. Actual corruption is very hard to measure, so perception of corruption is a reasonable thing to research since it is in many ways "the best you can do" to get some measure of actual corruption.
IMHO not a minor strength, getting pixel perfect designs when SVGs are involved in Figma is near impossible, they are rendered in a very different way and in some cases the differences are huge.