In June a conference put Regeneron's bone-disease drug at a 99.9 percent reduction, with a p-value of .001. The label the FDA approved on Wednesday says that same measurement was not statistically significant. Both are true, and Pasatru works. The approval rests on a different measurement, the count of new bone growths, which the label reports as statistically significant at both doses. Nothing here says otherwise. The disease is fibrodysplasia ossificans progressiva. Muscle and tendon turn to bone, a second skeleton grows over the first, and about 900 people worldwide are diagnosed with it. Sixty three of them were in this trial, and there will never be a bigger one. The measurement that changed its mind is the volume of new bone. At week 56 the placebo group averaged 10.5 cubic centimetres of it. The groups on Pasatru averaged 0.02 and 0.01. Those three numbers are where the 99.8 and 99.9 percent come from. Under the test the trial specified in advance, that difference does not clear significance, and the label says so in one sentence. Under a different model, built after the trial had finished, it becomes 99.9 percent with a p-value of .001. That is the version presented in June and reported on the fourteenth. Neither the FDA's approval announcement nor Regeneron's press release mentions bone volume. The word does not appear in either document. There is a second number doing the same thing. The label permits dropping from 10 milligrams per kilogram to 3 if the higher dose is not tolerated. At 10 the flare-up rate ratio against placebo is 0.12, with a range of 0.03 to 0.42. At 3 it is 0.85, with a range running from a 74 percent fall to a 173 percent rise. Regeneron's release reports the 0.85 as a 15 percent reduction, which is correct, and prints no range beside it. That is exactly why the ranges matter here. When a disease has 900 known patients, the label is the only document that still carries how uncertain the numbers are. The conference slide does not. The press release does not. The coverage does not. By the time a number reaches a patient it has been through three retellings, and the range is gone from all of them.