Which is... reasonable.

Google COULD offer more time machine features and perform diffing on pages. But a reddit "page" will always have content changes, as everything is generated from a database and kept fresh on the page. The ONLY metric therefore Google could use would be whatever meta tag or header tag that reddit provides.

That doesn't explain why Google lists the old search results as being from this month, while Duck correctly lists them as being from years past.

I've always wondered, how search engines get a hold of timestamps. Locally with a cached sample, like I explained above, parsing a page's content or some metadata? It's not like the HTTP protocol sends me a "file created/last modified date" along with the payload, does it?

That could explain the first screenshot, but definitely not the second, where google has it tagged as years old.

One function that takes the output of the page, and renders it so only what's user visible, actually gets indexed. So no headers, no JSON data, no nothing, unless it's actually in the final outcome of the page when rendered. This would require jsdom or some other DOM implementation. Hardly hard for Google (Chrome) to achieve this, and been done multiple times.

Second function is a function that does the same call twice, passing the page to function one each time, then compare them two. If you make two calls right next to each other, and some data is different, you discard that from your search index. Instead you only index data that appears in both calls.

Now you don't have the issue of "dynamic content" anymore...

But I do like your idea.

To go a bit further on your idea - you could apply machine learning to analyse the changes. So for example, ML could determine what is probably the "content area" of the page simply by having built out a NN for each website that self-expires the training data at about 1 month (to account for redesigns over time).

The major problem will still be "ads" in the middle of the content, especially odd scroll designs ads that have a different "picture" at each scroll position, as well as video ads that are likely to be different at each screen shot.

Another form of ad being the "linked" words like when you see random words of the paragraphs becoming links that go to shitty websites that define the word but show a bunch of other ads.

I suppose Google could simply install UBlock in it's training data collector harness to help with that stuff. >()

Obviously, it doesn't need to consider the upvotes directly but maybe the text inside the page. Or the date.

Initially I had the impression that search was hard to implement. However, spending a work week figuring it out with ElasticSearch, Solr and Sphinx changed my mind. Getting the solution to work with the scale of the website would take more work, but all the know-how is there, and they could put a whole team to the task for a month.

Google does use it, last time I used it there was even tooling for it in the 'check how my site is to crawl' console, whatever it's called.

Yet I haven't seen even one instance of this anywhere. :(

If Google gives you what you want straight away then you leave; sure, you come back, but they want to be bad enough to keep you on there and good enough to be better than other searches. Their reach and resources cures the latter.