test_that("Monte Carlo simulation is reproducible", {
  x <- agri_pam_example()
  one <- pam_monte_carlo(x, cv = 0.1, n = 100, seed = 99, index = "Paddy")
  two <- pam_monte_carlo(x, cv = 0.1, n = 100, seed = 99, index = "Paddy")

  expect_s3_class(one, "agri_pam_mc")
  expect_equal(one$draws, two$draws)
  expect_equal(nrow(one$draws), 100)
  expect_equal(nrow(one$summary), 10)
  expect_equal(nrow(one$probabilities), 4)
  expect_true(all(one$probabilities$probability >= 0 &
                  one$probabilities$probability <= 1))
})

test_that("a supplied seed does not alter the caller RNG state", {
  set.seed(1234)
  before <- .Random.seed
  invisible(pam_monte_carlo(
    agri_pam_example(), cv = 0.1, n = 20, seed = 5, index = "Paddy"
  ))
  expect_identical(.Random.seed, before)
})

test_that("named CV values hold omitted components fixed", {
  x <- agri_pam_example()
  mc <- pam_monte_carlo(
    x,
    cv = c(social_revenue = 0.2),
    n = 100,
    seed = 2,
    index = "Paddy"
  )
  expect_gt(stats::sd(mc$draws$social_revenue), 0)
  expect_equal(unique(mc$draws$private_revenue), 123400)
  expect_equal(unique(mc$draws$social_domestic_factors), 71000)
})

test_that("zero uncertainty reproduces the deterministic PAM", {
  x <- agri_pam_example()
  mc <- pam_monte_carlo(x, cv = 0, n = 20, seed = 1, index = "Paddy")
  expect_equal(unique(mc$draws$private_profit), 27600)
  expect_equal(unique(mc$draws$drc), pam_indicators(x)$drc[1])
  expect_equal(as.data.frame(mc), mc$draws)
})

test_that("Monte Carlo arguments are validated", {
  x <- agri_pam_example()
  expect_error(pam_monte_carlo(x, cv = -0.1, n = 10), "non-negative")
  expect_error(pam_monte_carlo(x, cv = 0.1, n = 1), "greater than")
  bad <- diag(6)
  bad[1, 2] <- 0.5
  expect_error(pam_monte_carlo(x, n = 10, correlation = bad), "symmetric")
})