test_that("Pareto screen removes dominated rows", {
  d <- data.frame(quality = c(.9, .8, .7), cost = c(.02, .03, .01), latency = c(1, 2, 1))
  p <- pareto_ai(d)
  expect_lt(nrow(p), nrow(d))
  expect_true(.9 %in% p$quality)
})