test_that("k posterior target is correctly expressed on log scale", {
  time <- c(0.3, 0.7, 1.2, 2)
  status <- c(1, 1, 0, 1)
  common <- list(alp = 0.4, p1 = 0.45, p2 = 0.55,
                 th1 = 0.8, th2 = 1.6, k2 = 2.1)
  current <- 1.2
  candidate <- 1.7
  implemented <-
    ambs:::.logpost_k1_ww(common$alp, common$p1, common$p2,
                           common$th1, common$th2, candidate, common$k2,
                           time, status, log(2), 1) -
    ambs:::.logpost_k1_ww(common$alp, common$p1, common$p2,
                           common$th1, common$th2, current, common$k2,
                           time, status, log(2), 1)
  expected <-
    ambs:::.loglik_ww(common$alp, common$p1, common$p2,
                       common$th1, common$th2, candidate, common$k2,
                       time, status) -
    ambs:::.loglik_ww(common$alp, common$p1, common$p2,
                       common$th1, common$th2, current, common$k2,
                       time, status) +
    dnorm(log(candidate), log(2), 1, log = TRUE) -
    dnorm(log(current), log(2), 1, log = TRUE)

  expect_equal(implemented, expected, tolerance = 1e-12)
})

test_that("relative Gamma proposal has the stated moments and finite correction", {
  current <- c(1e-10, 0.5, 1e6)
  relative_variance <- 0.1
  shape <- 1 / relative_variance
  scale <- relative_variance * current

  expect_equal(shape * scale, current)
  expect_equal(shape * scale^2, relative_variance * current^2)

  candidate <- current * c(0.8, 1.2, 0.9)
  correction <-
    dgamma(current, shape = shape,
           scale = relative_variance * candidate, log = TRUE) -
    dgamma(candidate, shape = shape,
           scale = relative_variance * current, log = TRUE)
  expect_true(all(is.finite(correction)))
})

test_that("relabeling does not overwrite sampler acceptance rates", {
  set.seed(51)
  time <- c(0.2, 0.35, 0.5, 0.8, 1.1, 1.5, 2, 2.8, 3.5, 5)
  status <- c(1, 1, 0, 1, 1, 0, 1, 0, 1, 0)
  fit <- alpmixsurv(time, status, model = "ww",
                    mcmc = list(nburn = 100, nsamp = 100, thin = 1))
  expect_equal(fit$acceptance, fit$acceptance_raw)
})

test_that("WW generalized-mean survival remains monotone for separated components", {
  time <- seq(8, 2200, length.out = 300)
  draw <- c(alpha = 2, p = 0.58, k1 = 0.0106, th1 = 10.5,
            k2 = 11.66, th2 = 0.058)
  survival <- ambs:::.curve_ww(draw, time, "s")

  expect_true(all(is.finite(survival)))
  expect_true(all(survival >= 0 & survival <= 1))
  expect_true(all(diff(survival) <= 1e-12))
})

test_that("stable alpha-mixture agrees with direct arithmetic mixture", {
  log_s1 <- log(c(0.9, 0.5, 0.1))
  log_s2 <- log(c(0.7, 0.3, 0.02))
  log_h1 <- log(c(0.1, 0.2, 0.4))
  log_h2 <- log(c(0.2, 0.3, 0.5))
  out <- ambs:::.alpmix_log_quantities(1, 0.4, log_s1, log_s2,
                                        log_h1, log_h2)
  expected_survival <- 0.4 * exp(log_s1) + 0.6 * exp(log_s2)
  expected_hazard <-
    (0.4 * exp(log_s1 + log_h1) + 0.6 * exp(log_s2 + log_h2)) /
    expected_survival

  expect_equal(exp(out$log_surv), expected_survival, tolerance = 1e-12)
  expect_equal(exp(out$log_haz), expected_hazard, tolerance = 1e-12)
})