User blog comment:Upquark11111/An Explanation of Loader's Number/@comment-11227630-20171209072636

It seems that Loader used a different inference ruleset to this one (see page 89). Loader's ruleset lacks "weakening rule" and "conversion rule", and the rule judging the type of $$\Pi x:A.X$$ is different from the "(s1, s2) rule" in the pdf.