Does the domain resolve at all. A surprising share of any exported or scraped list is parked domains, expired registrations, and sites that no longer answer. This costs nothing to check and removes rows before anything else has to look at them.
If I could keep only one filter, it would be outbound links per page. Editorial pages link out a few times. Link shops link out constantly.
Does the page return a genuine 200. Watch for soft 404s, meaning pages that return a success code while displaying a not found message, and for holding pages dressed up as content. A status code alone is not proof of a real page, which is why presence of actual content matters alongside it.
Is the site indexable. A sitewide noindex, or a robots configuration that blocks crawling of the sections where links would sit, makes any placement worthless from a ranking perspective. Rare but catastrophic when missed, since the link looks perfect to a human reviewer.
Does the language match your market. Cheap to detect and easy to skip when you are moving fast. A high authority link on a site your customers cannot read sends no useful traffic and carries a weak relevance signal.
Is the site still alive editorially. Check whether anything has been published in the last few months. A site that stopped publishing in 2021 may still have authority on paper, but its pages are drifting out of relevance, its audience has gone, and nobody is going to answer your email anyway.
How many outbound links per page. This is the filter I would keep if I could only keep one. Editorial pages link out a handful of times because that is what serving a reader looks like. Pages built to monetize links carry dozens or hundreds. The ratio of external to internal links tells the same story from another angle.
Only now does enrichment make sense. Authority scores, traffic estimates, and referring domain counts all cost credits or rate limit, and the whole point of the sequence above is that you are spending them on a list that has already been cut down to plausible candidates rather than on raw export rows.