"So if I see two videos that were both uploaded "a year ago", it's impossible to know which is more recent without actually opening the videos, going to the description, and looking at the date."

For better or worse (worse, IMHO), this is how YouTube search works. It takes at least two HTTP requests to get the precise dates. One to retreieve the SERP. Another to retrieve the JSON for each videoID. The JSON contains the upload date, etc.

IMO this sort of inefficient design pattern is faciltated by web developers who rely on Javascript, which is seemingly all of them. Webpages become shells for advertising and Javascript is used to fetch non-advertising, i.e., data. Google wants users to make clicks and track them, i.e., collect behavioural data. Thus it makes sense that a user would be forced to "open" a video to see its actual upload date.

The suite of tiny custom utilties I wrote in C can search YT and retrieve all YT videos, this can be upwards of 900 results. The output is CSV, SQL or simple HTML, including upload date and other data I find useful. This is done using only two TCP connections. No browser is needed. The simple HTML is just a page that looks like CSV, with a hyperlink to the video file. It's "YT redesigned" to look like I want it to look. I do the web development before opening the page in a browser.

The custom "downloader" I wrote is smaller, simpler and faster than yt-dlp. No Python required. But most times I don't download. I just use the simple HTML that looks like CSV. IMHO, this is much better than Google's privacy-invasive page laden with Javascript. I can see the precise dates and other useful data. I can sort by vaious columns. And I can exclude garbage videos that Google ("the algorithm") inserts into results, videos that have nothing to do with the search query string I submitted.