library(organik)

testthat::test_that("organik returns the expected structure", {
  testthat::skip_on_cran()
  set.seed(1001)
  y <- 10 + cumsum(rnorm(150, sd = 0.4))  # smooth positive-ish series

  obj <- organik(
    ts = y, horizon = 3L,
    n_variants = 4L,
    n_testing = 15L,
    seed = 123
  )

  # Class and top-level fields
  testthat::expect_s3_class(obj, "organik")
  testthat::expect_true(is.list(obj$model_list))
  testthat::expect_true(is.list(obj$growth_pred_funs))
  testthat::expect_true(is.list(obj$level_pred_funs))
  testthat::expect_true(is.matrix(obj$cor_mat))

  # One predictive object per horizon; qfun works
  testthat::expect_equal(length(obj$growth_pred_funs), 3L)
  qg <- obj$growth_pred_funs[[1]]$qfun(c(0.1, 0.5, 0.9))
  ql <- obj$level_pred_funs[[2]]$qfun(c(0.1, 0.5, 0.9))
  testthat::expect_true(is.numeric(qg) && length(qg) == 3L && all(is.finite(qg)))
  testthat::expect_true(is.numeric(ql) && length(ql) == 3L && all(is.finite(ql)))
})

testthat::test_that("organik path_prediction (Gaussian copula) has correct shapes and monotone quantiles", {
  testthat::skip_on_cran()
  set.seed(2002)
  y <- 5 + cumsum(rnorm(150, sd = 0.3))

  obj <- organik(
    ts = y, horizon = 4L,
    n_variants = 3L,
    n_testing = 5L,
    seed = 42
  )

  path <- obj$path_prediction(
    n_paths = 1000L,
    probs = c(0.1, 0.5, 0.9),
    copula = "gaussian",
    seed = 99
  )

  # Shapes: paths are n_paths x horizon
  testthat::expect_equal(dim(path$level_paths), c(1000L, 4L))
  testthat::expect_equal(dim(path$cum_growth_paths), c(1000L, 4L))

  # Quantile matrices: rows = horizons, columns = q10,q50,q90 (as constructed)
  testthat::expect_equal(nrow(path$level_quants), 4L)
  testthat::expect_equal(ncol(path$level_quants), 3L)

  # Monotonic quantiles for each horizon
  for (h in seq_len(4L)) {
    qs <- path$level_quants[h, ]
    testthat::expect_true(qs[1] <= qs[2] && qs[2] <= qs[3])
  }

  # last_level was echoed back
  testthat::expect_equal(path$last_level, tail(y, 1))
})

testthat::test_that("organik path_prediction works with t-copula and validates df", {
  testthat::skip_on_cran()
  set.seed(3003)
  y <- 7 + cumsum(rnorm(160, sd = 0.35))

  obj <- organik(
    ts = y, horizon = 3L,
    n_variants = 3L,
    n_testing = 15L,
    seed = 11
  )

  # df must be > 1
  testthat::expect_error(
    obj$path_prediction(n_paths = 200L, copula = "t", df = 1),
    "df must be > 1",
    fixed = TRUE
  )

  path_t <- obj$path_prediction(n_paths = 500L, copula = "t", df = 5, seed = 202)
  testthat::expect_equal(path_t$copula, "t")
  testthat::expect_equal(path_t$df, 5)

  # correlation used is positive-definite (small eigenvalue > 0)
  ev <- eigen(path_t$cor_used, symmetric = TRUE, only.values = TRUE)$values
  testthat::expect_true(min(ev) > 0)
})

testthat::test_that("organik is reproducible under a fixed seed", {
  testthat::skip_on_cran()
  set.seed(4004)
  y <- 12 + cumsum(rnorm(170, sd = 0.25))

  obj1 <- organik(
    ts = y, horizon = 3L,
    n_variants = 3L,
    n_testing = 15L,
    seed = 777
  )
  obj2 <- organik(
    ts = y, horizon = 3L,
    n_variants = 3L,
    n_testing = 15L,
    seed = 777
  )

  # correlation matrices identical
  testthat::expect_equal(obj1$cor_mat, obj2$cor_mat, tolerance = 0)

  # Path summaries identical with same seed
  p1 <- obj1$path_prediction(n_paths = 600L, probs = c(0.05, 0.5, 0.95), seed = 999)
  p2 <- obj2$path_prediction(n_paths = 600L, probs = c(0.05, 0.5, 0.95), seed = 999)

  testthat::expect_equal(p1$level_quants, p2$level_quants, tolerance = 0)
  testthat::expect_equal(p1$cum_growth_quants, p2$cum_growth_quants, tolerance = 0)
})