diff --git a/NAMESPACE b/NAMESPACE
index 151aa6149..72a69df36 100644
--- a/NAMESPACE
+++ b/NAMESPACE
@@ -72,6 +72,7 @@ export(get_grouping)
 export(get_metrics)
 export(get_pairwise_comparisons)
 export(get_pit_histogram)
+export(get_unique_values)
 export(interval_coverage)
 export(is_forecast)
 export(is_forecast_binary)
diff --git a/R/get-unique-values.R b/R/get-unique-values.R
new file mode 100644
index 000000000..4bb4c03fa
--- /dev/null
+++ b/R/get-unique-values.R
@@ -0,0 +1,69 @@
+#' @title Get unique value counts per forecast unit column
+#'
+#' @description
+#' Given a forecast object, this function counts the number of unique values
+#' in each column that defines the forecast unit (as determined by
+#' [get_forecast_unit()]). This is useful for getting a quick overview of the
+#' structure and scope of a forecast dataset.
+#'
+#' @param by character vector (default: `NULL`). If specified, results are
+#'   grouped by this column, showing unique value counts for each forecast
+#'   unit column per group. The grouping column itself is excluded from the
+#'   column-level counts.
+#'
+#' @returns A data.table with columns `"column"` (the name of the forecast unit
+#' column) and `"N_unique"` (the number of unique values in that column).
+#' If `by` is specified, there is an additional column for the grouping
+#' variable.
+#'
+#' @inheritParams score
+#' @importFrom checkmate assert_subset
+#' @export
+#' @keywords gain-insights
+#' @examples
+#' \dontshow{
+#'   data.table::setDTthreads(2) # restricts number of cores used on CRAN
+#' }
+#'
+#' example_quantile |>
+#'   as_forecast_quantile() |>
+#'   get_unique_values()
+#'
+#' example_quantile |>
+#'   as_forecast_quantile() |>
+#'   get_unique_values(by = "model")
+get_unique_values <- function(forecast, by = NULL) {
+  forecast <- clean_forecast(forecast, copy = TRUE, na.omit = TRUE)
+  forecast_unit <- get_forecast_unit(forecast)
+
+  if (!is.null(by)) {
+    assert_subset(by, forecast_unit, empty.ok = FALSE)
+  }
+
+  forecast <- as.data.table(forecast)
+
+  cols <- if (!is.null(by)) setdiff(forecast_unit, by) else forecast_unit
+
+  if (is.null(by)) {
+    out <- data.table(
+      column = cols,
+      N_unique = vapply(
+        cols,
+        function(col) length(unique(forecast[[col]])),
+        integer(1)
+      )
+    )
+  } else {
+    out_list <- forecast[,
+      {
+        vals <- lapply(cols, function(col) length(unique(.SD[[col]])))
+        list(column = cols, N_unique = unlist(vals))
+      },
+      by = by,
+      .SDcols = cols
+    ]
+    out <- out_list
+  }
+
+  return(out[])
+}
diff --git a/R/z-globalVariables.R b/R/z-globalVariables.R
index 5229126e4..06b2ee0b5 100644
--- a/R/z-globalVariables.R
+++ b/R/z-globalVariables.R
@@ -51,6 +51,7 @@ globalVariables(c(
   "metrics",
   "metrics_select",
   "model",
+  "N_unique",
   "n_obs",
   "n_obs wis_component_name",
   "observed",
diff --git a/man/get_unique_values.Rd b/man/get_unique_values.Rd
new file mode 100644
index 000000000..cdf82fb20
--- /dev/null
+++ b/man/get_unique_values.Rd
@@ -0,0 +1,43 @@
+% Generated by roxygen2: do not edit by hand
+% Please edit documentation in R/get-unique-values.R
+\name{get_unique_values}
+\alias{get_unique_values}
+\title{Get unique value counts per forecast unit column}
+\usage{
+get_unique_values(forecast, by = NULL)
+}
+\arguments{
+\item{forecast}{A forecast object (a validated data.table with predicted and
+observed values).}
+
+\item{by}{character vector (default: \code{NULL}). If specified, results are
+grouped by this column, showing unique value counts for each forecast
+unit column per group. The grouping column itself is excluded from the
+column-level counts.}
+}
+\value{
+A data.table with columns \code{"column"} (the name of the forecast unit
+column) and \code{"N_unique"} (the number of unique values in that column).
+If \code{by} is specified, there is an additional column for the grouping
+variable.
+}
+\description{
+Given a forecast object, this function counts the number of unique values
+in each column that defines the forecast unit (as determined by
+\code{\link[=get_forecast_unit]{get_forecast_unit()}}). This is useful for getting a quick overview of the
+structure and scope of a forecast dataset.
+}
+\examples{
+\dontshow{
+  data.table::setDTthreads(2) # restricts number of cores used on CRAN
+}
+
+example_quantile |>
+  as_forecast_quantile() |>
+  get_unique_values()
+
+example_quantile |>
+  as_forecast_quantile() |>
+  get_unique_values(by = "model")
+}
+\keyword{gain-insights}
diff --git a/tests/testthat/test-get-unique-values.R b/tests/testthat/test-get-unique-values.R
new file mode 100644
index 000000000..895d9acee
--- /dev/null
+++ b/tests/testthat/test-get-unique-values.R
@@ -0,0 +1,125 @@
+# ==============================================================================
+# `get_unique_values()` # nolint: commented_code_linter
+# ==============================================================================
+test_that("get_unique_values() works with a quantile forecast", {
+  forecast <- suppressMessages(as_forecast_quantile(example_quantile))
+  result <- get_unique_values(forecast)
+
+  expect_s3_class(result, c("data.table", "data.frame"), exact = TRUE)
+
+  # Should have a column for the column name and one for the unique count
+  expect_true("column" %in% names(result) || "Column" %in% names(result) ||
+                length(names(result)) == 2)
+  expect_true(ncol(result) == 2)
+
+  # The column names in the result should be exactly the forecast unit columns
+  fu <- get_forecast_unit(forecast)
+  col_col <- names(result)[1]
+  expect_setequal(result[[col_col]], fu)
+
+  # Check some expected unique value counts (after clean_forecast with na.omit)
+  count_col <- names(result)[2]
+  location_count <- result[result[[col_col]] == "location", ][[count_col]]
+  expect_identical(location_count, 4L)
+  target_type_count <- result[result[[col_col]] == "target_type", ][[count_col]]
+  expect_identical(target_type_count, 2L)
+  location_name_count <- result[
+    result[[col_col]] == "location_name",
+  ][[count_col]]
+  expect_identical(location_name_count, 4L)
+})
+
+test_that("get_unique_values() works with different forecast types", {
+  forecast_binary <- suppressMessages(as_forecast_binary(example_binary))
+  result_binary <- get_unique_values(forecast_binary)
+
+  expect_s3_class(result_binary, c("data.table", "data.frame"), exact = TRUE)
+
+  # Should only contain forecast unit columns, not protected columns
+  col_col <- names(result_binary)[1]
+  expect_false("predicted" %in% result_binary[[col_col]])
+  expect_false("observed" %in% result_binary[[col_col]])
+
+  forecast_sample <- suppressMessages(
+    as_forecast_sample(example_sample_continuous)
+  )
+  result_sample <- get_unique_values(forecast_sample)
+
+  expect_s3_class(result_sample, c("data.table", "data.frame"), exact = TRUE)
+
+  # sample_id should not appear (it is a protected column)
+  col_col_s <- names(result_sample)[1]
+  expect_false("sample_id" %in% result_sample[[col_col_s]])
+})
+
+test_that("get_unique_values() returns correct output structure", {
+  dt <- data.table::data.table(
+    location = c("A", "A", "B"),
+    model = c("m1", "m2", "m1"),
+    observed = c(1, 2, 3),
+    predicted = c(1.1, 2.1, 3.1)
+  )
+  forecast <- suppressMessages(as_forecast_point(dt))
+  result <- get_unique_values(forecast)
+
+  expect_s3_class(result, c("data.table", "data.frame"), exact = TRUE)
+
+  col_col <- names(result)[1]
+  count_col <- names(result)[2]
+
+  # Should have exactly 2 rows: location and model
+  expect_identical(nrow(result), 2L)
+  expect_setequal(result[[col_col]], c("location", "model"))
+
+  # observed and predicted should NOT appear
+  expect_false("observed" %in% result[[col_col]])
+  expect_false("predicted" %in% result[[col_col]])
+
+  # Unique counts
+  location_count <- result[result[[col_col]] == "location", ][[count_col]]
+  model_count <- result[result[[col_col]] == "model", ][[count_col]]
+  expect_identical(location_count, 2L)
+  expect_identical(model_count, 2L)
+})
+
+test_that("get_unique_values() accepts a `by` argument for grouping", {
+  forecast <- suppressMessages(as_forecast_quantile(example_quantile))
+  result <- get_unique_values(forecast, by = "model")
+
+  expect_s3_class(result, c("data.table", "data.frame"), exact = TRUE)
+
+  # Should have a model column in the result
+
+  expect_true("model" %in% names(result))
+
+  # "model" should not appear in the column-name column since it's the
+
+  # grouping variable
+  col_col <- setdiff(names(result), c("model", names(result)[ncol(result)]))[1]
+  if (is.na(col_col)) {
+    col_col <- names(result)[1]
+  }
+
+  # Should have rows for each model x column combination
+  # After na.omit there are 4 models
+  fu <- setdiff(get_forecast_unit(forecast), "model")
+  n_models <- length(unique(
+    scoringutils:::clean_forecast(forecast, copy = TRUE, na.omit = TRUE)$model
+  ))
+  expect_gte(nrow(result), n_models)
+})
+
+test_that("get_unique_values() errors on non-forecast input", {
+  expect_error(get_unique_values("not a forecast"))
+  expect_error(get_unique_values(42))
+})
+
+test_that("get_unique_values() handles data with NAs correctly", {
+  dt <- data.table::copy(example_quantile)
+  dt[1:5, location := NA]
+  forecast <- suppressMessages(as_forecast_quantile(dt))
+
+  # Should not error
+  expect_no_error(result <- get_unique_values(forecast))
+  expect_s3_class(result, c("data.table", "data.frame"), exact = TRUE)
+})