test_that("awe_registry lists candidates without fitting", {
  reg_c <- awe_registry("continuous")
  reg_b <- awe_registry("binary")
  expect_true(all(c("model", "has_probabilities") %in% names(reg_c)))
  expect_true("linear_regression" %in% reg_c$model)
  expect_true("logistic_or_multinomial" %in% reg_b$model)
})

test_that("expanded model bank includes new regression candidates", {
  m <- awe_model(mtcars, target = "mpg", k = 3, seed = 1)
  expect_true("knn_regression" %in% m$leaderboard$model)
})

test_that("expanded model bank includes new classification candidates", {
  m <- awe_model(iris, target = "Species", k = 3, seed = 1)
  expect_true(any(c("lda", "qda") %in% m$leaderboard$model))
})

test_that("awe_ensemble builds a working weighted ensemble", {
  m <- awe_model(mtcars, target = "mpg", k = 3, seed = 1)
  ens <- awe_ensemble(m, top_n = 3)
  expect_s3_class(ens, "AWEEnsemble")
  expect_s3_class(ens, "AWEWrapper")
  p <- predict(ens, newdata = mtcars)
  expect_length(p, nrow(mtcars))
  expect_true(all(is.finite(p)))
})

test_that("awe_ensemble errors with too few valid candidates", {
  m <- awe_model(mtcars, target = "mpg", k = 3, models = "linear_regression", seed = 1)
  expect_error(awe_ensemble(m, top_n = 3))
})

test_that("awe_compare stacks leaderboards across runs", {
  m1 <- awe_model(mtcars, target = "mpg", k = 3, seed = 1)
  m2 <- awe_model(mtcars, target = "mpg", k = 3, models = c("linear_regression", "decision_tree"), seed = 1)
  cmp <- awe_compare(full = m1, restricted = m2)
  expect_s3_class(cmp, "AWECompare")
  expect_setequal(unique(cmp$leaderboard$run), c("full", "restricted"))
})

test_that("awe_tune finds a best hyperparameter combination", {
  tn <- awe_tune(mtcars, target = "mpg", model = "decision_tree", k = 3, seed = 1)
  expect_s3_class(tn, "AWETune")
  expect_true("cp" %in% names(tn$best_params))
  expect_s3_class(tn$wrapper, "AWEWrapper")
  p <- predict(tn$wrapper, newdata = mtcars)
  expect_length(p, nrow(mtcars))
})

test_that("awe_save/awe_load round-trip an AWEModel", {
  m <- awe_model(mtcars, target = "mpg", k = 3, seed = 1)
  f <- tempfile(fileext = ".rds")
  awe_save(m, f)
  m2 <- awe_load(f)
  expect_equal(m$best_name, m2$best_name)
})

test_that("plot_roc computes an AUC in [0, 1] for a binary target", {
  d <- iris[iris$Species != "setosa", ]
  d$Species <- droplevels(d$Species)
  m <- awe_model(d, target = "Species", k = 3, seed = 1)
  w <- awe_wrap(m)
  res <- plot_roc(w)
  expect_true(res$auc >= 0 && res$auc <= 1)
  expect_true(all(c("fpr", "tpr") %in% names(res$roc)))
})

test_that("plot_confusion runs without error on a classification evaluation", {
  m <- awe_model(iris, target = "Species", k = 3, seed = 1)
  w <- awe_wrap(m)
  ev <- awe_evaluate(w, newdata = iris)
  expect_error(plot_confusion(ev), NA)
})

test_that("plot_confusion errors on a regression evaluation", {
  m <- awe_model(mtcars, target = "mpg", k = 3, seed = 1)
  w <- awe_wrap(m)
  ev <- awe_evaluate(w, newdata = mtcars)
  expect_error(plot_confusion(ev))
})

test_that("plot_residuals runs without error on a regression wrapper", {
  m <- awe_model(mtcars, target = "mpg", k = 3, seed = 1)
  w <- awe_wrap(m)
  expect_error(plot_residuals(w), NA)
})

test_that("eda_report runs without error", {
  expect_error(eda_report(mtcars, vars = c("mpg", "hp", "wt")), NA)
})

test_that("theme_awe and awe_palette are usable", {
  expect_true(inherits(theme_awe(), "theme"))
  expect_length(awe_palette(4), 4)
  expect_length(awe_palette(20), 20)
})

test_that("plot.awe supports roc/confusion/residuals types", {
  fit_c <- awe(mtcars, target = "mpg", k = 3, seed = 1)
  expect_error(plot(fit_c, type = "residuals"), NA)

  d <- iris[iris$Species != "setosa", ]
  d$Species <- droplevels(d$Species)
  fit_b <- awe(d, target = "Species", k = 3, seed = 1)
  expect_error(plot(fit_b, type = "roc"), NA)
  expect_error(plot(fit_b, type = "confusion"), NA)
})