diff --git a/test/alloc_tests.jl b/test/alloc_tests.jl
index 9fa5f29..f8e6935 100644
--- a/test/alloc_tests.jl
+++ b/test/alloc_tests.jl
@@ -1,49 +1,62 @@
 using AllocCheck
 using PoissonRandom
 using Random
+using Test
+
+# AllocCheck's static LLVM-IR analysis is explicitly documented as not guaranteed
+# stable across Julia versions, and on Julia prereleases the in-flux codegen makes
+# it emit architecture-dependent false positives: on 1.13.0-rc1 the aarch64 path
+# flags even `procf` (pure scalar Float64 arithmetic returning an `NTuple{4,Float64}`,
+# which cannot heap-allocate) as allocating, while x86_64 reports zero. The runtime
+# `@allocated == 0` check below is the ground-truth regression guard and is exercised
+# on every version/architecture; the static `@check_allocs` guard is additionally run
+# only on released Julia, where its result is reliable.
+const RELIABLE_STATIC_ALLOC_CHECK = isempty(VERSION.prerelease)
+
+# Assert `f(args...)` performs zero heap allocations at runtime (the real guard), and
+# additionally pass it through AllocCheck's static analysis on released Julia.
+macro test_zero_allocs(call)
+    @assert call.head == :call
+    f = call.args[1]
+    args = call.args[2:end]
+    return quote
+        local f = $(esc(f))
+        local args = ($(map(esc, args)...),)
+        f(args...)                       # warm up / compile
+        @test (@allocated f(args...)) == 0
+        if RELIABLE_STATIC_ALLOC_CHECK
+            local checked = AllocCheck.check_allocs(f, map(typeof, args))
+            @test isempty(checked)
+        end
+    end
+end
 
 @testset "AllocCheck - Zero Allocations" begin
+    rng = Random.default_rng()
+    passthrough = PassthroughRNG()
+
     @testset "count_rand" begin
-        @check_allocs function test_count_rand(rng::TaskLocalRNG, λ::Float64)
-            PoissonRandom.count_rand(rng, λ)
-        end
-        rng = Random.default_rng()
-        @test test_count_rand(rng, 2.0) isa Int
-        @test test_count_rand(rng, 5.0) isa Int
+        @test_zero_allocs PoissonRandom.count_rand(rng, 2.0)
+        @test_zero_allocs PoissonRandom.count_rand(rng, 5.0)
     end
 
     @testset "ad_rand" begin
-        @check_allocs function test_ad_rand(rng::TaskLocalRNG, λ::Float64)
-            PoissonRandom.ad_rand(rng, λ)
-        end
-        rng = Random.default_rng()
-        @test test_ad_rand(rng, 10.0) isa Int
-        @test test_ad_rand(rng, 50.0) isa Int
+        @test_zero_allocs PoissonRandom.ad_rand(rng, 10.0)
+        @test_zero_allocs PoissonRandom.ad_rand(rng, 50.0)
     end
 
     @testset "pois_rand" begin
-        @check_allocs function test_pois_rand(rng::TaskLocalRNG, λ::Float64)
-            pois_rand(rng, λ)
-        end
-        rng = Random.default_rng()
-        @test test_pois_rand(rng, 2.0) isa Int
-        @test test_pois_rand(rng, 10.0) isa Int
+        @test_zero_allocs pois_rand(rng, 2.0)
+        @test_zero_allocs pois_rand(rng, 10.0)
     end
 
     @testset "pois_rand with PassthroughRNG" begin
-        @check_allocs function test_pois_rand_passthrough(rng::PassthroughRNG, λ::Float64)
-            pois_rand(rng, λ)
-        end
-        passthrough = PassthroughRNG()
-        @test test_pois_rand_passthrough(passthrough, 2.0) isa Int
-        @test test_pois_rand_passthrough(passthrough, 10.0) isa Int
+        @test_zero_allocs pois_rand(passthrough, 2.0)
+        @test_zero_allocs pois_rand(passthrough, 10.0)
     end
 
     @testset "procf" begin
-        @check_allocs function test_procf(λ::Float64, K::Int, s::Float64)
-            PoissonRandom.procf(λ, K, s)
-        end
-        @test test_procf(10.0, 5, 3.162) isa NTuple{4, Float64}
-        @test test_procf(10.0, 15, 3.162) isa NTuple{4, Float64}
+        @test_zero_allocs PoissonRandom.procf(10.0, 5, 3.162)
+        @test_zero_allocs PoissonRandom.procf(10.0, 15, 3.162)
     end
 end