test_that("named priors are normalized and used by model", {
  custom <- list(
    alpha = list(mean = 1.5, sd = 0.7),
    p = list(shape1 = 2, shape2 = 3),
    component1 = list(k = list(meanlog = 0.1, sdlog = 0.4)),
    component2 = list(theta = list(shape = 2, scale = 0.8))
  )
  got <- ambs:::.normalize_prior(custom, "gw")
  expect_equal(got$alpha, custom$alpha)
  expect_equal(got$p, custom$p)
  expect_equal(got$component1$k$meanlog, 0.1)
  expect_equal(got$component2$theta$scale, 0.8)
  expect_equal(got$component1$theta$shape, 1)
})

test_that("legacy prior remains compatible and records variance semantics", {
  legacy <- list(k = c(2, 3), theta = c(4, 5),
                 alpha = c(1, 9), p = c(6, 7))
  got <- ambs:::.normalize_prior(legacy, "ww")
  expect_true(isTRUE(attr(got, "legacy")))
  expect_equal(got$alpha, list(mean = 1, sd = 3))
  expect_equal(got$component1$theta$shape, 2)
  expect_equal(got$component2$theta$scale, 5)
})

test_that("invalid prior scales and MCMC controls fail early", {
  expect_error(
    ambs:::.normalize_prior(list(alpha = list(sd = 0)), "ll"),
    "positive and finite"
  )
  expect_error(
    alpmixsurv(c(1, 2), c(1, 0), mcmc = list(nburn = 1, nsamp = 0)),
    "nsamp"
  )
})

test_that("alpha sampler is continuous and does not manufacture zeros", {
  set.seed(1804)
  time <- c(0.25, 0.5, 0.8, 1.1, 1.7, 2.3)
  status <- c(1, 1, 0, 1, 0, 1)
  for (model in c("ww", "gw", "ll")) {
    fit <- suppressWarnings(alpmixsurv(
      time, status, model = model,
      mcmc = list(nburn = 20, nsamp = 40, thin = 1)
    ))
    expect_true(all(is.finite(fit$posterior)))
    expect_false(any(fit$posterior[, "alpha"] == 0))
    expect_equal(nrow(fit$posterior), 40)
  }
})

test_that("thinning retains the requested number of posterior samples", {
  set.seed(91)
  fit <- suppressWarnings(alpmixsurv(
    c(0.4, 0.9, 1.4, 2), c(1, 1, 0, 1), model = "ww",
    mcmc = list(nburn = 5, nsamp = 12, thin = 3)
  ))
  expect_equal(nrow(fit$posterior), 12)
  expect_equal(fit$mcmc$n_iter, 41)
})

test_that("alpha remains continuous and near-zero probability is reported", {
  set.seed(1842)
  fit <- suppressWarnings(alpmixsurv(
    c(0.25, 0.5, 0.8, 1.1, 1.7, 2.3), c(1, 1, 0, 1, 0, 1),
    model = "ww", mcmc = list(nburn = 20, nsamp = 40)
  ))
  expect_false("alpha_fixed" %in% names(formals(alpmixsurv)))
  expect_false("alpha_fixed" %in% names(fit))
  expect_false(any(fit$posterior[, "alpha"] == 0))
  expect_equal(fit$alpha_near_zero_band, 0.01)
  expect_equal(
    fit$alpha_near_zero_probability,
    mean(abs(fit$posterior[, "alpha"]) < 0.01)
  )
})

test_that("LL log-time standardization and parameter back-transform are exact", {
  time <- c(0.2, 0.7, 1.5, 4)
  transform <- ambs:::.ll_time_standardization(time)
  original <- c(alpha = 0.3, p = 0.4, mu1 = -0.2, mu2 = 0.8,
                sig1 = 0.5, sig2 = 0.15)
  working <- original
  working[c("mu1", "mu2")] <-
    (original[c("mu1", "mu2")] - transform$center) / transform$scale
  working[c("sig1", "sig2")] <-
    original[c("sig1", "sig2")] / transform$scale
  expect_equal(
    ambs:::.curve_ll(original, time, "s"),
    ambs:::.curve_ll(working, transform$time, "s"),
    tolerance = 1e-12
  )
  prior <- ambs:::.default_prior("ll")
  round_trip <- ambs:::.ll_prior_to_original(
    ambs:::.ll_prior_to_working(prior, transform), transform
  )
  expect_equal(round_trip, prior, tolerance = 1e-12)
})

test_that("LL fit reports original-scale draws and likelihood", {
  set.seed(901)
  time <- c(0.25, 0.4, 0.8, 1.2, 2, 3)
  status <- c(1, 1, 0, 1, 0, 1)
  fit <- suppressWarnings(alpmixsurv(
    time, status, model = "ll", mcmc = list(nburn = 100, nsamp = 20),
    ident = "none"
  ))
  draw <- fit$posterior[1, ]
  expected <- ambs:::.loglik_ll(
    draw["alpha"], draw["p"], 1 - draw["p"], draw["mu1"], draw["mu2"],
    draw["sig1"], draw["sig2"], time, status
  )
  expect_equal(unname(draw["loglik"]), expected, tolerance = 1e-10)
  expect_equal(fit$prior_source, "fixed prior on standardized log-time")
  expect_true(all(c("center", "scale", "time") %in% names(fit$time_transform)))
})

test_that("WW and GW median-time scaling preserves survival curves", {
  time <- c(20, 80, 150, 400)
  transform <- ambs:::.scale_time_standardization(time)
  for (model in c("ww", "gw")) {
    original <- c(alpha = 0.4, p = 0.35, k1 = 1.4, th1 = 90,
                  k2 = 2.2, th2 = 260)
    working <- original
    working[c("th1", "th2")] <-
      original[c("th1", "th2")] / transform$scale
    curve_fun <- if (model == "ww") ambs:::.curve_ww else ambs:::.curve_gw
    expect_equal(curve_fun(original, time, "s"),
                 curve_fun(working, transform$time, "s"),
                 tolerance = 1e-11)

    prior <- ambs:::.default_prior(model)
    round_trip <- ambs:::.scale_prior_to_original(
      ambs:::.scale_prior_to_working(prior, transform), transform
    )
    expect_equal(round_trip, prior, tolerance = 1e-12)
  }
})

test_that("WW fit returns original-scale theta and likelihood", {
  set.seed(902)
  time <- c(25, 40, 80, 120, 200, 300)
  status <- c(1, 1, 0, 1, 0, 1)
  fit <- alpmixsurv(time, status, model = "ww",
                    mcmc = list(nburn = 100, nsamp = 20), ident = "none")
  draw <- fit$posterior[1, ]
  expected <- ambs:::.loglik_ww(
    draw["alpha"], draw["p"], 1 - draw["p"], draw["th1"], draw["th2"],
    draw["k1"], draw["k2"], time, status
  )
  expect_equal(unname(draw["loglik"]), expected, tolerance = 1e-10)
  expect_equal(fit$time_transform$scale, median(time))
  expect_equal(fit$working_time, time / median(time))
  expect_equal(fit$prior_source, "fixed prior on median-standardized time")
  expect_equal(fit$chain_inits[[1]]$th1,
               fit$chain_inits_working[[1]]$th1 * median(time))
})