How can I apply this system to a random discussion archive page at HN in order to evaluate it more efficiently as a discussion guidance mechanism? I don't want to see usernames in that example, and I don't want a dynamic example either — but I think it would be much easier to convince HN that your AI product is worthwhile if you present an HN-specific example. Specifically, I suggest you take an HN discussion (the HTML is very simply structured), pipe each comment through your engine, and append the <div style="background-color: soft-blue;"> "Your comment etc etc" responses that would have been shown to each comment in the discussion.

Looking at the most popular results for " " on HN Algolia, I would recommend selecting a post that has at least a few hundred comments and is also about HN or YC or YC-adjacent people (since the mods are extra light-touch on such posts), in order to take the best possible sample for unmoderated discussion to evaluate Respectify against. This post is a good example that fits those criteria; I didn't pay attention to it at the time and I haven't assessed the discussion beyond 'total comment count >= 500': https://news.ycombinator.com/item?id=40521657

I recognize that's theoretically a lot of effort, but from a coding standpoint, it's simply `for $comment in $dom.xpath(/blah/blah/comment) { $ai.eval($comment); undef $comment.username; $comment.append($respectify.bulleted_list_with_html_colors); }` for what has the potential to be an extremely convincing demo to the target audience of us here.

▲

vintagedave 5 hours ago | parent [-]

That is an excellent idea. It is 2AM my time, but I may set Claude going and check in the morning. (I tend to keep a closer eye on AI coding than that usually!)

The preset articles and trying out comments were intended to be something similar: see a topic, see how it works. But running it on each and every comment on an existing thread is really powerful.

There may be privacy concerns? General respect? I don't want to tie assessments to specific commenters, who published in good faith not expecting some kind of automated review, nor thereby imply they commented poorly for example. But I'll code it up on my end and see what we can do with it. It's truly a very nice idea.

	▲	altairprime 4 hours ago \| parent [-]
		If you look through the history of the Show HN category, you will see a near-endless stream of "analyzing HN discussions" prior arts that may offer some assurances. I'm not suggesting removing usernames because anonymity is involved — it is not! — but, instead, to specifically focus viewers on the substance of the comments rather than the who of them. That's also why I chose something from two years ago, so that there's no appropriate reason to witchhunt over the past. (Some may still, but nothing short of evaluating AI-generated data will stop them, and you can't make a reasonable case using AI to evaluate AI.)