test_that("fit_langmuir recovers approximately correct parameters", {
  set.seed(1)
  Ce <- c(2, 5, 10, 20, 35, 50, 70)
  Qmax_true <- 5; KL_true <- 0.15
  qe <- (Qmax_true * KL_true * Ce) / (1 + KL_true * Ce) + stats::rnorm(length(Ce), 0, 0.03)

  fit <- fit_langmuir(Ce, qe)
  expect_equal(fit$method, "nonlinear")
  expect_true(fit$R2 > 0.9)
  expect_equal(fit$Qmax, Qmax_true, tolerance = 0.5)
})

test_that("fit_langmuir errors with too few points", {
  expect_error(fit_langmuir(c(1, 2), c(1, 2)), "at least 3")
})

test_that("fit_freundlich recovers approximately correct parameters", {
  set.seed(2)
  Ce <- c(2, 5, 10, 20, 35, 50, 70)
  Kf_true <- 1.2; n_true <- 2.5
  qe <- Kf_true * Ce^(1 / n_true) + stats::rnorm(length(Ce), 0, 0.03)

  fit <- fit_freundlich(Ce, qe)
  expect_true(fit$R2 > 0.9)
  expect_equal(fit$n, n_true, tolerance = 0.5)
})

test_that("fit_pfo and fit_pso return sensible R2", {
  set.seed(3)
  t <- c(5, 15, 30, 60, 120, 240)
  qe_true <- 4.2; k1_true <- 0.05
  qt <- qe_true * (1 - exp(-k1_true * t)) + stats::rnorm(length(t), 0, 0.02)

  fit1 <- fit_pfo(t, qt)
  fit2 <- fit_pso(t, qt)
  expect_true(fit1$R2 > 0.85)
  expect_true(fit2$R2 > 0.85)
})