## ---- ai_test_retest -------------------------------------------------

test_that("test-retest ICC is high for stable repeated measurements", {
  # 5 prompts, 2 occasions, near-identical across occasions -> high ICC
  responses <- cbind(
    c(1, 2, 3, 4, 5),
    c(1.1, 2.0, 3.1, 3.9, 5.0)
  )
  out <- ai_test_retest(responses)
  expect_type(out, "list")
  expect_true(all(c("icc", "msb", "msj", "mse") %in% names(out)))
  expect_true(is.finite(out$icc))
  expect_gt(out$icc, 0.90)
})

test_that("test-retest ICC is low when occasions are unrelated noise", {
  set.seed(1)
  responses <- cbind(rnorm(30), rnorm(30))  # independent columns
  out <- ai_test_retest(responses)
  expect_lt(out$icc, 0.5)
})

test_that("test-retest requires at least 2 rows and 2 columns", {
  expect_error(ai_test_retest(matrix(1:3, nrow = 1)), "at least 2 rows")
  expect_error(ai_test_retest(matrix(1:3, ncol = 1)), "at least 2 rows")
})

test_that("test-retest drops rows with missing values and messages", {
  responses <- cbind(c(1, 2, 3, NA, 5), c(1, 2, 3, 4, 5))
  expect_message(ai_test_retest(responses), "dropping")
})

test_that("test-retest errors when too few complete rows remain", {
  responses <- cbind(c(1, NA, NA), c(1, 2, 3))
  expect_error(suppressMessages(ai_test_retest(responses)), "complete rows")
})

## ---- ai_internal_consistency ---------------------------------------

test_that("Cronbach's alpha is high for strongly correlated items", {
  set.seed(2)
  latent <- rnorm(50)
  responses <- sapply(1:4, function(i) latent + rnorm(50, sd = 0.2))
  out <- ai_internal_consistency(responses)
  expect_gt(out$alpha, 0.90)
  expect_equal(out$n_items, 4)
})

test_that("internal consistency errors with fewer than 2 items", {
  expect_error(ai_internal_consistency(matrix(1:5, ncol = 1)),
               "at least 2 items")
})

test_that("internal consistency returns NA alpha when total variance is zero", {
  responses <- matrix(rep(c(1, 1), each = 5), ncol = 2)  # constant rows
  expect_warning(out <- ai_internal_consistency(responses),
                 "variance is zero")
  expect_true(is.na(out$alpha))
})

## ---- ai_reliability (wrapper) --------------------------------------

test_that("ai_reliability returns test_retest and optional consistency", {
  set.seed(3)
  responses <- matrix(rnorm(30, mean = 5), nrow = 10, ncol = 3)
  out <- ai_reliability(responses)
  expect_s3_class(out, "aiEvalR_reliability")
  expect_true("test_retest" %in% names(out))
  expect_false("internal_consistency" %in% names(out))

  out2 <- ai_reliability(responses, items = TRUE)
  expect_true("internal_consistency" %in% names(out2))
})

## ---- ai_bootstrap_reliability --------------------------------------

test_that("bootstrap reliability returns an interval bracketing the estimate", {
  set.seed(4)
  latent <- rnorm(20)
  responses <- cbind(latent + rnorm(20, sd = 0.3), latent + rnorm(20, sd = 0.3))
  out <- ai_bootstrap_reliability(responses, n_boot = 200, seed = 42)
  expect_true(all(c("estimate", "ci_lower", "ci_upper", "n_valid") %in% names(out)))
  expect_lte(out$ci_lower, out$estimate)
  expect_gte(out$ci_upper, out$estimate)
})

test_that("bootstrap reliability is reproducible with a fixed seed", {
  set.seed(5)
  responses <- cbind(rnorm(15), rnorm(15))
  a <- ai_bootstrap_reliability(responses, n_boot = 100, seed = 99)
  b <- ai_bootstrap_reliability(responses, n_boot = 100, seed = 99)
  expect_equal(a$ci_lower, b$ci_lower)
  expect_equal(a$ci_upper, b$ci_upper)
})