what is xviral fr?

XViral is a tool that grades a draft social media post for X, formerly known as Twitter, the same way the platform's own recommendation algorithm would grade it, so you can see how a post might perform before you actually publish it. The project is built on real code. In May 2026, X published the actual production code behind its For You feed, the system that decides which posts get shown to people, along with a version of its trained ranking model. That release included the exact scoring formula and the categories it checks a post against, but it left out some of the specific numbers the real system uses internally, such as exact weight values and the wording of its internal AI judge prompts. XViral rebuilds that released pipeline so it can run locally on a post you have not published yet. It uses an AI model, either Claude or a model running on your own computer through a tool called Ollama, to stand in for the platform's internal AI judges, checking things like whether a post looks like generic filler content or spam. It then applies the same scoring formula from the real code to estimate how the post would be ranked. Along the way, the project found some things that common advice about the algorithm gets wrong or does not mention, such as every post being checked by a quality and spam judge, small accounts facing an extra spam check, actions like getting muted or reported hurting a post's score far more than a like helps it, and videos below a minimum length getting no credit at all, unlike the widely repeated claim that video always performs far better. The tool can also fetch real public engagement numbers for an already posted tweet, without needing to log in, so its predictions can be checked against what actually happened. The project is upfront that its numeric scores are not yet fully calibrated to match real world outcomes, and sets a specific accuracy target it wants to hit before treating the scores as reliable. To run it, you need Python 3.11 or newer along with either the Claude command line tool or a locally running Ollama model.