Heaps, Top-K & Intervals

Everything X shows you is ranked. The timeline picks a few hundred posts out of the thousands your follows produced. Trending surfaces the hottest hashtags of the last hour. A dashboard reports the median latency of a live request stream. Underneath all of it sit three plain moves this family drills. A heap pulls the extreme of a set without ordering the rest. Two heaps run back to back to track the running middle of a stream. And sorting then sweeping folds a pile of overlapping time windows into a tidy few. None of the three is exotic, which is exactly why the loop expects them to be automatic.

The pressure X applies here is the pressure it applies everywhere in the coding loop. The code has to run and clear the cases on a shared editor, you reach for the edges without being told, and you give a spoken read of the time and space. That holds across both versions of the company: the earlier Twitter screen still live on some roles, and the harder xAI-run bar that came with the merger. Reports vary on the exact rounds. What stays constant is that a clean heap you can write without fumbling buys the room for the follow-up that almost always lands next.

The X angle

Every problem here is a scale model of something ranking your feed right now. The kth largest engagement score is the cutoff a post has to beat to make a top-k shelf, the same threshold trending computes over a firehose of hashtags. The running median is how a live dashboard reports the p50 of an engagement or latency stream without re-sorting on every event. Coalescing overlapping windows is what a rate limiter and a trending pipeline both do when they fold bursts of activity into the spans that matter. Build the small version cleanly and you have shown the shape of the real one.