diff --git a/R/core-functions.R b/R/core-functions.R
index 3dfeb2e..0d887b9 100644
--- a/R/core-functions.R
+++ b/R/core-functions.R
@@ -200,8 +200,9 @@ fzl <- function(models = "*", calculators = "*", check = FALSE) {
 #'   \code{"algorithms/montecarlo_uniform.py"}.
 #' @param calculators Calculator specification(s). Default \code{NULL}. When
 #'   \code{model} is a function, this must be a single integer (default
-#'   \code{1L}), accepted for API compatibility (see "Direct function model"
-#'   below) — calls are always run sequentially regardless of its value.
+#'   \code{1L}) and is always forced to \code{1L} (see "Direct function
+#'   model" below): R functions are only safe to call from the main thread,
+#'   so a value other than \code{1} triggers a warning and is overridden.
 #' @param algorithm_options Algorithm options as a named list, a JSON string,
 #'   or a path to a JSON file. Default \code{NULL}.
 #' @param analysis_dir Analysis directory. Default \code{"analysis"}.
@@ -209,30 +210,28 @@ fzl <- function(models = "*", calculators = "*", check = FALSE) {
 #' @section Direct function model:
 #' Instead of a file-based model, \code{model} can be an R function (this
 #' requires the \code{main} branch of \code{fz} from GitHub, installed with
-#' \code{fz_install(packages = "git+https://github.com/Funz/fz.git")} — this
+#' \code{fz_install(packages = "git+https://github.com/Funz/fz.git")} -- this
 #' mode is not available in released PyPI versions of \code{funz-fz} yet). In
 #' this mode:
 #' \itemize{
-#'   \item \code{input_path} must be \code{NULL} — there are no input files.
+#'   \item \code{input_path} must be \code{NULL} -- there are no input files.
 #'   \item \code{input_variables} names must match the function's arguments.
 #'   \item \code{output_expression} may be \code{NULL}; the value used is then
 #'     the first element of the function's return value (its return value
 #'     directly if scalar, the first element if a vector/list, or the first
 #'     entry's value if a named list).
-#'   \item \code{calculators} must be a single integer, accepted for API
-#'     compatibility but currently without effect: on the \code{fz} side,
-#'     function-model calls always run sequentially, one at a time in the
-#'     calling thread — never through a thread pool. This is required
-#'     because R functions are called back into the R session via
-#'     \code{reticulate}, which is only safe from the main thread; running
-#'     a Python-side thread pool (which always dispatches to a worker thread,
-#'     even with a single worker) would call the function from a thread
-#'     other than the main one and crash the R session. This safety fix
-#'     requires \href{https://github.com/Funz/fz/pull/73}{Funz/fz#73} on the
-#'     \code{fz} \code{main} branch (not yet in a PyPI release as of
-#'     2026-07-12); without it, direct function models crash regardless of
-#'     \code{calculators}. A value other than \code{1} emits a warning
-#'     noting that it has no effect.
+#'   \item \code{calculators} must be a single integer, but is always forced
+#'     to \code{1} here -- regardless of the value passed in -- before being
+#'     forwarded to \code{fz}. On the \code{fz} (Python) side,
+#'     \code{calculators > 1} now evaluates a Python-function model
+#'     concurrently in a worker-thread pool (\href{https://github.com/Funz/fz/pull/73}{Funz/fz#73});
+#'     that is unsafe here because R functions are called back into the R
+#'     session via \code{reticulate}, which is only safe from the main
+#'     thread -- invoking the function from any other thread crashes the R
+#'     session. Passing a value other than \code{1} therefore emits a
+#'     warning explaining that it is being forced back to \code{1}, and the
+#'     call always proceeds with \code{calculators = 1} (strictly
+#'     sequential, one call at a time, in the calling thread).
 #'   \item each iteration's directory (\code{iterNNN/}) only contains a
 #'     \code{values.csv} of that iteration's function inputs/outputs, since
 #'     there is no file-based execution.
@@ -303,7 +302,7 @@ fzl <- function(models = "*", calculators = "*", check = FALSE) {
 #'   model = rosenbrock,
 #'   output_expression = "result",
 #'   algorithm = "examples/algorithms/bfgs.py",
-#'   calculators = 4L,
+#'   calculators = 1L, # forced to 1L anyway for R functions -- see "Direct function model"
 #'   algorithm_options = list(max_iter = 20, tol = 1e-4)
 #' )
 #' }
@@ -322,12 +321,16 @@ fzd <- function(input_path, input_variables, model, output_expression = NULL, al
     calculators <- as.integer(calculators)
     if (calculators != 1L) {
       warning(
-        "calculators = ", calculators, " has no effect when 'model' is an R ",
-        "function: evaluations always run sequentially (one call at a time), ",
-        "since R functions bridged in via reticulate are only safe to call ",
-        "from the main thread.",
+        "calculators = ", calculators, " was requested but is forced to 1 ",
+        "when 'model' is an R function: on the fz (Python) side, calculators > 1 ",
+        "now evaluates the model concurrently in a worker-thread pool (see ",
+        "https://github.com/Funz/fz#function-models), but R functions bridged in ",
+        "via reticulate are only safe to call from the main thread -- calling them ",
+        "from any other thread crashes the R session. calculators is therefore ",
+        "always reset to 1 here, regardless of the requested value.",
         call. = FALSE
       )
+      calculators <- 1L
     }
   }
   fz_module <- get_fz()
diff --git a/R/install.R b/R/install.R
index 08f40e1..b099793 100644
--- a/R/install.R
+++ b/R/install.R
@@ -68,7 +68,7 @@ fz_available <- function() {
 #' @param global Logical; install system-wide instead of user-level.
 #'   Default \code{FALSE}.
 #'
-#' @return Named list with installation details (path, id, …).
+#' @return Named list with installation details (path, id, ...).
 #' @export
 #'
 #' @examples
@@ -93,7 +93,7 @@ install_model <- function(source, global = FALSE) {
 #' @param global Logical; install system-wide instead of user-level.
 #'   Default \code{FALSE}.
 #'
-#' @return Named list with installation details (path, name, …).
+#' @return Named list with installation details (path, name, ...).
 #' @export
 #'
 #' @examples
diff --git a/R/zzz.R b/R/zzz.R
index c8f1c27..87262b7 100644
--- a/R/zzz.R
+++ b/R/zzz.R
@@ -3,7 +3,7 @@
 .pkg$fz_available <- NULL
 
 .onLoad <- function(libname, pkgname) {
-  # Python is not initialised at load time — deferred to first use via get_fz().
+  # Python is not initialised at load time -- deferred to first use via get_fz().
 }
 
 #' @keywords internal
diff --git a/man/fzd.Rd b/man/fzd.Rd
index 209e27d..792fa32 100644
--- a/man/fzd.Rd
+++ b/man/fzd.Rd
@@ -36,8 +36,9 @@ value is used.}
 
 \item{calculators}{Calculator specification(s). Default \code{NULL}. When
 \code{model} is a function, this must be a single integer (default
-\code{1L}), accepted for API compatibility (see "Direct function model"
-below) — calls are always run sequentially regardless of its value.}
+\code{1L}) and is always forced to \code{1L} (see "Direct function
+model" below): R functions are only safe to call from the main thread,
+so a value other than \code{1} triggers a warning and is overridden.}
 
 \item{algorithm_options}{Algorithm options as a named list, a JSON string,
 or a path to a JSON file. Default \code{NULL}.}
@@ -57,30 +58,28 @@ is useful for sensitivity analysis, surrogate-model fitting, or optimization.
 
 Instead of a file-based model, \code{model} can be an R function (this
 requires the \code{main} branch of \code{fz} from GitHub, installed with
-\code{fz_install(packages = "git+https://github.com/Funz/fz.git")} — this
+\code{fz_install(packages = "git+https://github.com/Funz/fz.git")} -- this
 mode is not available in released PyPI versions of \code{funz-fz} yet). In
 this mode:
 \itemize{
-\item \code{input_path} must be \code{NULL} — there are no input files.
+\item \code{input_path} must be \code{NULL} -- there are no input files.
 \item \code{input_variables} names must match the function's arguments.
 \item \code{output_expression} may be \code{NULL}; the value used is then
 the first element of the function's return value (its return value
 directly if scalar, the first element if a vector/list, or the first
 entry's value if a named list).
-\item \code{calculators} must be a single integer, accepted for API
-compatibility but currently without effect: on the \code{fz} side,
-function-model calls always run sequentially, one at a time in the
-calling thread — never through a thread pool. This is required
-because R functions are called back into the R session via
-\code{reticulate}, which is only safe from the main thread; running
-a Python-side thread pool (which always dispatches to a worker thread,
-even with a single worker) would call the function from a thread
-other than the main one and crash the R session. This safety fix
-requires \href{https://github.com/Funz/fz/pull/73}{Funz/fz#73} on the
-\code{fz} \code{main} branch (not yet in a PyPI release as of
-2026-07-12); without it, direct function models crash regardless of
-\code{calculators}. A value other than \code{1} emits a warning
-noting that it has no effect.
+\item \code{calculators} must be a single integer, but is always forced
+to \code{1} here -- regardless of the value passed in -- before being
+forwarded to \code{fz}. On the \code{fz} (Python) side,
+\code{calculators > 1} now evaluates a Python-function model
+concurrently in a worker-thread pool (\href{https://github.com/Funz/fz/pull/73}{Funz/fz#73});
+that is unsafe here because R functions are called back into the R
+session via \code{reticulate}, which is only safe from the main
+thread -- invoking the function from any other thread crashes the R
+session. Passing a value other than \code{1} therefore emits a
+warning explaining that it is being forced back to \code{1}, and the
+call always proceeds with \code{calculators = 1} (strictly
+sequential, one call at a time, in the calling thread).
 \item each iteration's directory (\code{iterNNN/}) only contains a
 \code{values.csv} of that iteration's function inputs/outputs, since
 there is no file-based execution.
@@ -149,7 +148,7 @@ result <- fzd(
   model = rosenbrock,
   output_expression = "result",
   algorithm = "examples/algorithms/bfgs.py",
-  calculators = 4L,
+  calculators = 1L, # forced to 1L anyway for R functions -- see "Direct function model"
   algorithm_options = list(max_iter = 20, tol = 1e-4)
 )
 }
diff --git a/man/install_algorithm.Rd b/man/install_algorithm.Rd
index 229782d..7900505 100644
--- a/man/install_algorithm.Rd
+++ b/man/install_algorithm.Rd
@@ -14,7 +14,7 @@ or path to a local zip file.}
 Default \code{FALSE}.}
 }
 \value{
-Named list with installation details (path, name, …).
+Named list with installation details (path, name, ...).
 }
 \description{
 Installs an algorithm from a GitHub repository name, URL, or local zip file
diff --git a/man/install_model.Rd b/man/install_model.Rd
index 3680350..296ba1e 100644
--- a/man/install_model.Rd
+++ b/man/install_model.Rd
@@ -14,7 +14,7 @@ path to a local zip file.}
 Default \code{FALSE}.}
 }
 \value{
-Named list with installation details (path, id, …).
+Named list with installation details (path, id, ...).
 }
 \description{
 Installs a model from a GitHub repository name, URL, or local zip file into
diff --git a/tests/testthat/test-fzd-calculators-forced.R b/tests/testthat/test-fzd-calculators-forced.R
new file mode 100644
index 0000000..b448487
--- /dev/null
+++ b/tests/testthat/test-fzd-calculators-forced.R
@@ -0,0 +1,121 @@
+# fzd(model = <R function>, calculators = N) must always be forced to
+# calculators = 1 before being forwarded to the fz (Python) module, since R
+# functions bridged in via reticulate are only safe to call from the main
+# thread. On the fz side, calculators > 1 now genuinely dispatches evaluations
+# to a worker-thread pool (Funz/fz#73) for ordinary Python functions, so this
+# R-side safeguard is what prevents R-function models from crashing the R
+# session when a caller passes calculators > 1.
+
+make_capturing_fz_module <- function(captured) {
+  list(
+    fzd = function(input_path, input_variables, model, output_expression, algorithm,
+                    calculators, algorithm_options, analysis_dir) {
+      captured$calculators <- calculators
+      list(summary = "ok")
+    }
+  )
+}
+
+a_dummy_r_function <- function(x, y) list(z = x + y)
+
+test_that("fzd forces calculators to 1 (with a warning) for an R function model when > 1 requested", {
+  captured <- new.env()
+  testthat::local_mocked_bindings(get_fz = function() make_capturing_fz_module(captured))
+
+  expect_warning(
+    result <- fzd(
+      input_path = NULL,
+      input_variables = list(x = "[0;1]", y = "[0;1]"),
+      model = a_dummy_r_function,
+      output_expression = "z",
+      algorithm = "algorithms/randomsampling.py",
+      calculators = 4L
+    ),
+    "forced to 1"
+  )
+
+  expect_equal(captured$calculators, 1L)
+  expect_equal(result$summary, "ok")
+})
+
+test_that("fzd does not warn and uses calculators = 1 when omitted for an R function model", {
+  captured <- new.env()
+  testthat::local_mocked_bindings(get_fz = function() make_capturing_fz_module(captured))
+
+  expect_no_warning(
+    fzd(
+      input_path = NULL,
+      input_variables = list(x = "[0;1]", y = "[0;1]"),
+      model = a_dummy_r_function,
+      output_expression = "z",
+      algorithm = "algorithms/randomsampling.py"
+    )
+  )
+
+  expect_equal(captured$calculators, 1L)
+})
+
+test_that("fzd does not warn when calculators = 1 is explicitly passed for an R function model", {
+  captured <- new.env()
+  testthat::local_mocked_bindings(get_fz = function() make_capturing_fz_module(captured))
+
+  expect_no_warning(
+    fzd(
+      input_path = NULL,
+      input_variables = list(x = "[0;1]", y = "[0;1]"),
+      model = a_dummy_r_function,
+      output_expression = "z",
+      algorithm = "algorithms/randomsampling.py",
+      calculators = 1L
+    )
+  )
+
+  expect_equal(captured$calculators, 1L)
+})
+
+test_that("fzd rejects a non-scalar/non-numeric calculators for an R function model", {
+  captured <- new.env()
+  testthat::local_mocked_bindings(get_fz = function() make_capturing_fz_module(captured))
+
+  expect_error(
+    fzd(
+      input_path = NULL,
+      input_variables = list(x = "[0;1]"),
+      model = function(x) list(y = x),
+      output_expression = "y",
+      algorithm = "algorithms/randomsampling.py",
+      calculators = c(1L, 2L)
+    ),
+    "single integer"
+  )
+
+  expect_error(
+    fzd(
+      input_path = NULL,
+      input_variables = list(x = "[0;1]"),
+      model = function(x) list(y = x),
+      output_expression = "y",
+      algorithm = "algorithms/randomsampling.py",
+      calculators = "sh://"
+    ),
+    "single integer"
+  )
+})
+
+test_that("fzd forwards calculators unchanged for a file-based (non-function) model", {
+  captured <- new.env()
+  testthat::local_mocked_bindings(get_fz = function() make_capturing_fz_module(captured))
+
+  expect_no_warning(
+    fzd(
+      input_path = "input.txt",
+      input_variables = list(x = "[0;1]"),
+      model = list(output = list(y = "cat output.txt")),
+      output_expression = "y",
+      algorithm = "algorithms/randomsampling.py",
+      calculators = 4L
+    )
+  )
+
+  expect_equal(captured$calculators, 4L)
+})