cume_dist() and percent_rank() computed values within mutate do not ignore NA values

I'm expecting values as in the column correct when computing these two functions on mutate-generated gain, just as it is correct when computed directly on b-a. I don't think the results of the ecdf and correct columns should differ. I include pct_rank to show that function is also affected.

library(tidyverse)
set.seed(0)
df <- tibble(a = runif(1000, -1, 1), b = runif(1000, -1, 1))
df[df < 0] <- NA
df <- df %>%
  mutate(gain = b - a)
df <- df %>%
  mutate(ecdf = cume_dist(gain),
         correct = cume_dist(b - a),
         pct_rank = percent_rank(gain))