Since October 13, the part of X's ranker that shows your posts to people who don't follow you has been trained on whether they read to the end. X added that line to the changelog in its open-source algorithm repo this week, and @ThePeterMick flagged it on October 28 in a post that has passed 93,000 views.
So a stranger only carries a crypto post forward if they actually finish it. The same changelog says ranking got better at predicting time spent from October 8, and Peter notes that no score weights moved this round.
That changes how an on-chain post should be built. A Hyperliquid funding breakdown that gives away the number in line one and then pads three paragraphs gets opened and abandoned, and Peter's recap of the open weights puts a quick open-and-leave at only 0.3 for the tap. A post that saves the one figure readers can check for themselves until the last line only pays off for readers who reach that line.
His recap also lists bookmarks at zero, a copy-link share at 20, and a second original in the same feed at about 62% of the first. The 62% figure is why I now keep my own originals 90 to 120 minutes apart.
X's changelog dates the read-to-end training for non-follower reach to October 13, about two weeks before it was written down.
So a stranger only carries a crypto post forward if they actually finish it. The same changelog says ranking got better at predicting time spent from October 8, and Peter notes that no score weights moved this round.
That changes how an on-chain post should be built. A Hyperliquid funding breakdown that gives away the number in line one and then pads three paragraphs gets opened and abandoned, and Peter's recap of the open weights puts a quick open-and-leave at only 0.3 for the tap. A post that saves the one figure readers can check for themselves until the last line only pays off for readers who reach that line.
His recap also lists bookmarks at zero, a copy-link share at 20, and a second original in the same feed at about 62% of the first. The 62% figure is why I now keep my own originals 90 to 120 minutes apart.
X's changelog dates the read-to-end training for non-follower reach to October 13, about two weeks before it was written down.
