test_that("el engine accepts valid inputs", {
  df <- make_el_test_data()
  el_base <- nmar(
    formula = Y_miss ~ X,
    data = df,
    engine = el_engine(auxiliary_means = c(X = 0), variance_method = 'none')
  )
  expect_s3_class(el_base, "nmar_result_el")
  expect_type(el_base$converged, "logical")

  el_resp_extra <- nmar(
    formula = Y_miss ~ X | Z,
    data = df,
    engine = el_engine(auxiliary_means = c(X = 0), variance_method = 'none')
  )
  expect_s3_class(el_resp_extra, "nmar_result_el")
  expect_type(el_resp_extra$converged, "logical")

  skip_if_not_installed("survey")
  design <- survey::svydesign(ids = ~1, data = df, weights = ~1)
  el_svy <- nmar(
    formula = Y_miss ~ X | Z,
    data = design,
    engine = el_engine(auxiliary_means = c(X = 0), variance_method = 'none')
  )
  expect_s3_class(el_svy, "nmar_result_el")
  expect_true(is.numeric(stats::weights(el_svy)))
})

test_that("el engine rejects faulty inputs", {
  base_df <- make_el_test_data()
  good_engine <- el_engine(auxiliary_means = c(X = 0), variance_method = 'none')

  bad_cases <- list(
    list(
      name = "missing outcome column",
      data = base_df[, c("X", "Z")],
      formula = Y_miss ~ X,
      response_predictors = NULL,
      message = "Variables not found"
    ),
    list(
      name = "missing covariate",
      data = base_df,
      formula = Y_miss ~ W,
      response_predictors = NULL,
      message = "object 'W' not found"
    ),
    list(
      name = "missing response predictor",
      data = base_df,
      formula = Y_miss ~ X | W,
      message = "object 'W' not found"
    ),
    list(
      name = "non numeric outcome",
      data = transform(base_df, Y_miss = ifelse(is.na(Y_miss), NA, as.character(Y_miss))),
      formula = Y_miss ~ X,
      response_predictors = NULL,
      message = "must be numeric"
    ),
    list(
      name = "NA in covariate",
      data = {
        df2 <- base_df
        idx <- which(!is.na(df2$Y_miss))[1]
        df2$X[idx] <- NA_real_
        df2
      },
      formula = Y_miss ~ X,
      response_predictors = NULL,
      message = "contains NA values"
    )
  )

  for (case in bad_cases) {
    if (is.null(case$message)) {
      expect_error(
        nmar(
          formula = case$formula,
          data = case$data,
          engine = good_engine
        ),
        info = case$name
      )
    } else {
      expect_error(
        nmar(
          formula = case$formula,
          data = case$data,
          engine = good_engine
        ),
        regexp = case$message,
        info = case$name
      )
    }
  }
})