module Only_in_test:sig..end
    Under the hood, 'a t = (unit -> 'a) and the only thing that ever forces them
    is the run_tests_and_exit function below which actually runs the unit tests.
    For example, suppose in some module, type t is actually an int.  You
    want to keep the type definition opaque, but use the underlying
    representation in unit tests.  You could write in the ml:
let test_to_int t = Only_in_test.return t
    [let test_of_int n = Only_in_test.return n
You would then expose in the mli:
type t
    ... functions for use in regular programming...
    val test_to_int : t -> int Only_in_test.t
    val test_of_int : int -> t Only_in_test.t
Finally, if you have specific values that you might want to use in unit tests, but that have top-level side-effects or take too long to compute, you can delay the side-effects or computation until the unit tests are run by writing, e.g.:
    let (test_special_value : t Only_in_test.t) =
       Only_in_test.of_thunk (fun () ->  (factorial 100))
instead of
    let (test_special_value : t Only_in_test.t) =
       Only_in_test.return (factorial 100)
type 'a t 
include Monad.S
val of_thunk : (unit -> 'a) -> 'a tval force : 'a t -> 'a