sorv ("sample of randomized voters") is an algorithm to measure the merit of a piece of content by (a) submitting it to a small random sample of the target audience; (b) having those users rate the content independently of each other; and (c) averaging their votes in order to determine the "merit" of the content (which determines how widely it is promoted to other users).

The algorithm is very simple, but if this were implemented by a major social media site, it would create the first system in history where the a user could submit a piece of content, and the outcome would depend entirely on the merit of the piece of content (as measured by the opinions of the target audience). There would be no more luck, no more gaming the system, and no more depending on favors from users with pre-existing large followings. (Crucially, it also means you don't have to spend weeks or months churning out mediocre submissions to convince the system that you're a "high-volume contributor" so it gives priority to your content, which is a bad policy.) And this applies not just to content but to other types of user "submissions" (in particular: submitting fact-checks against other users' content, and submitting abuse reports against content that breaks the rules).

The "sample of randomized voters" is essentially the algorithm used by peer-reviewed journals to review submitted papers (papers get mailed out to a small random subset of reviewers who independently rate the quality), and a bare-bones version of the same algorithm could be used to solve a wide range of problems that have plagued social media sites for years.

But anecdotally, many people (incorrectly) believe that this is how social media sites already work -- that if content gets a high rate of "likes" from the users who have seen it, it will be propagated above other content that gets a lower rate of likes -- however this is not the case. In a nutshell, suppose that on Twitter (for example) user ABC with 100,000 followers posts a video, and Twitter also shows it to 10,000 of their non-followers, and 5% of the non-followers "like" it. Meanwhile user XYZ with 1,000 followers posts a video, and Twitter also shows it to 1,000 of their non-followers, and 10% of the non-followers "like" it. At that point, Twitter has enough data to conclude that, among people who don't follow either account, the post from user XYZ is actually scoring better, but the default on virtually every social media site is to steer people toward the post from ABC instead, simply because ABC has more followers. This is an easy design mistake to make -- "Let's just steer people toward what's already popular" -- however it has profound implications for the results, rewarding people for cranking out mediocre content at the highest possible volume, rather than individual posts that more people like, while making it much more difficult for new accounts to do catch up, due to the pile-on lottery. Most people already know that "creating great art that is unappreciated in its own time" would not score well on social media, but it's surprising that "pandering" -- creating content which most people actually like -- does not score well either, unless you also get extremely lucky or game the system.

But with sorv, all of that goes away instantly. The only way to "win" in sorv is to create content that does well with its target audience. It's almost eerie to imagine the feeling of sitting down to create something, when none of the previous obstacles to success have been removed -- "Do I have to reach out to some popular users who can share it--"; nope; "Do I have to plan on building a following over weeks or months to reach an audience--;" nope; "Do I have to get a big lucky break with one video--"; nope, none of that matters any more. The only thing that matters is creating something that people like.

The three primary use cases for social media sites are:

Evaluating the merit of new content. Social media sites generally rely on users to rate each other's content, which often rewards gaming the system (groups of friends that all vote up each other's content), and just plain luck (the "pile-on lottery" effect, where users vote for something because they see other users voting for it, produces highly arbitrary outcomes). In a sorv system, a new piece of content could first be released to a random sample of the target audience. The users rate the content independently of each other, the average of their scores becomes the content's initial "rating," and then the content is promoted to the rest of the target audience in proportion to this rating. This reduces the influence of luck and gaming the system. It also means that bad content only wastes the time of a handful of people, while good content gets distributed widely.
Handling abuse reports. Sites such as YouTube, Facebook, and Reddit receive too many abuse reports for their staff to process them quickly and accurately, and users are usually left in the dark about the reasons for a decision. In a sorv system, users of the site could opt in to become "jurors" who adjudicate abuse reports. Then, when an abuse report is filed against a piece of content for violating a rule, the system randomly selects 10 currently online jurors to review the content and determine if it violated the rule; if more than 7 out of 10 vote that the content breaks the rule, then the content is prioritized for removal. Jurors can add comments explaining their votes, and all abuse report decisions could be public record (with juror votes and comments anonymized).
Handling fact-checks. Similarly to handling abuse reports -- in a sorv system, users of the site could opt in as "jurors" to adjudicate fact-checks. Then, when a fact-check is submitted against a piece of content (along with a supporting source), the system randomly selects 10 currently online jurors to review the fact-check; if more than 7 out of 10 vote that the original statement is false (or that the fact-check "adds enough context" that it should be displayed, even if the original post was not strictly false), then the fact-check is displayed (and possibly the reach of the original post can be limited in some way). Once again, jurors can add comments explaining their votes, and votes can be public record.

The primary advantages of this system are:

It's scalable (as more users join the system, more jurors and content-raters sign up as well, to spread the workload).
It's welcoming (the rules have to be written clearly so that jurors can enforce them, which means new users can understand the rules easily as well, instead of worrying about breaking "unwritten rules").
It's non-gameable. You can't "stack the deck" by having your friends falsely report some post or vote up your own content, because the system chooses the voters from the entire eligible population.
It provides high-quality content on average to the user. To oversimplify a bit: Suppose for a given piece of content, the "target audience" is all 100,000 users on the system. If an author submits a piece of "bad" content, and the content is released to a random sample of 100 users and they vote that it is "bad", then only those 100 users will see it. On the other hand, if the initial random sample votes the content is "good", then the system shows it to all 100,000 users. This means, on average, each user will see 1,000 pieces of "good" content for every 1 piece of "bad" content. (It's a bit more complicated, given that the "target audience" is not necessarily all users on the site, and users can rate things on a sliding scale rather than "good"/"bad", but the general conclusion still applies.)
It's as fast as possible. For abuse reports, for example, the system polls jurors who are currently online, and they all review the complaint at the same time, which means an abuse report could be adjudicated in less than 60 seconds. For evaluating the quality of a piece of content, the review time would be more varied -- the quality of a joke can be measured in 10 seconds, a long-form article might take an hour -- but the delay is limited to the time it takes a group of humans to do the evaluation.
It's transparent: the details of the algorithm can be made public, and you can also share publicly the details of how each individual post or abuse report got voted on. (Some systems like Reddit do make their source code public, but that by itself doesn't provide transparency as to why a particular piece of content gets the views that it does. Sorv provides much more straightforward transparency: "This post blew up because 85 out of 100 people in an initial sample thought it was really good!")
It optimizes for accepting criticism. If a random sample of peers gives feedback that your content is poor quality (or that you broke a rule, or that you posted a false fact), that means that the group separately and independently looked at your content and a plurality of them came to that conclusion. This maximizes the chances of a rational person accepting criticism (although of course many people still won't).
It's non-luck-based in terms of how it applies to rating content by quality -- rather than a small number of content submissions "going through the roof" due to the pile-on lottery, the system gives a post a quality rating (and promotes it to an audience in proportion to the rating) based on the average of the votes from a representative random sample of that audience.
It promotes a sense of ownership in the proper functioning of the community, by giving users the opportunity to vote on abuse reports (something that social media sites currently don't do), and also by giving users a way to rate content in such a way that their votes count (most social media sites let users "like" or "rate" content, but with no transparency as to how their votes get counted in determining what becomes popular).

This page makes some more detailed arguments for each of these advantages of the sorv system.

Many of these benefits, however, only apply once the user base is large enough. For example: the argument that the algorithm provides high-quality content on average, depended on the sample numbers that each new piece of content is reviewed by a random sample of 100 users, and only the good content is pushed out to the rest of the 100,000 users. If the system only has 100 users, then every piece of content has to be reviewed by every user on the site, which defeats the purpose. The "non-gameable" argument only applies once the user base is large enough as well; otherwise, a bad actor can sign up enough bots that they control a significant percentage of the user base.