Based on other details of analysis, I believe this means 2% of front pages posts are ShowHN. Not that 98% of ShowHN posts fail to reach front page. But the author doubles down on this confusion with flawed methodology:
> Second, and perhaps more importantly, the dataset doesn’t record the attempts made to get to the front page, i.e. all posts on Hacker News in a given day. It’s possible that there are orders of magnitude more blog posts posted but fewer that make the FP, whereas 95% of any academic paper submitted makes the front page (extreme figures used for illustrative purposes). So for simplicity, I’ll say “the likelihood of making the FP” which assumes a constant rate of conversion from post to FP across different categories.
and also doubles down on the logical fallacy in their conclusion:
> ShowHN is very valuable, but is not likely to land your product on the front page.