diff --git a/HISTORY.md b/HISTORY.md
index ee680fea1..475b345f9 100644
--- a/HISTORY.md
+++ b/HISTORY.md
@@ -2,6 +2,10 @@
 
 Fixed `apply_transform_strategy` to return only the target transform's forward log-Jacobian when converting a `DynamicLink` value to a `FixedTransform` target. It previously added the source's forward Jacobian as well, so `getlogjac`, `getlogjoint_internal`, and `getlogprior_internal` were wrong whenever linked values were fed into a fixed-transform strategy. See [#1407](https://github.com/TuringLang/DynamicPPL.jl/issues/1407).
 
+`marginalize` now constructs its default marginalizer with an explicit `Optim.LBFGS()`. The previous default, `MarginalLogDensities.LaplaceApprox()`, threw an `UndefVarError` once OptimizationOptimJL 0.4.19 stopped re-exporting `LBFGS`. See [#1447](https://github.com/TuringLang/DynamicPPL.jl/issues/1447).
+
+`LogDensityProblems.logdensity(ldf, x)` now throws an `ArgumentError` when `x` does not have the dimension the `LogDensityFunction` was constructed with. A shorter vector previously threw a `BoundsError` from inside model evaluation, and a longer one silently returned the log density of the first `dimension(ldf)` elements.
+
 # 0.42.4
 
 `arraydist` on a vector of univariate distributions now builds its `Distributions.Product` through the inner constructor instead of `Product(dists)`, which is deprecated. The outer constructor calls `Base.depwarn`, and that walks a backtrace on every call, so models with an `arraydist` likelihood paid it once per evaluation. The return type is unchanged.
diff --git a/src/logdensityfunction.jl b/src/logdensityfunction.jl
index 81f373039..a6aa7e386 100644
--- a/src/logdensityfunction.jl
+++ b/src/logdensityfunction.jl
@@ -529,6 +529,13 @@ end
 @inline function LogDensityProblems.logdensity(
     ldf::LogDensityFunction, params::AbstractVector{<:Real}
 )
+    if length(params) != ldf._dim
+        throw(
+            ArgumentError(
+                "The length of the input vector is $(length(params)), but the LogDensityFunction expects a vector of length $(ldf._dim) based on the ranges that were extracted when the LogDensityFunction was constructed.",
+            ),
+        )
+    end
     return logdensity_internal(
         params,
         ldf.model,
diff --git a/test/logdensityfunction.jl b/test/logdensityfunction.jl
index 66fe01ed1..66f5b6194 100644
--- a/test/logdensityfunction.jl
+++ b/test/logdensityfunction.jl
@@ -180,6 +180,18 @@ end
         @test LogDensityProblems.dimension(ldf) == 4
     end
 
+    @testset "logdensity checks the input length" begin
+        @model function m3()
+            x ~ Normal()
+            y ~ Normal()
+            return nothing
+        end
+        ldf = DynamicPPL.LogDensityFunction(m3())
+        @test LogDensityProblems.logdensity(ldf, zeros(2)) isa Real
+        @test_throws ArgumentError LogDensityProblems.logdensity(ldf, zeros(1))
+        @test_throws ArgumentError LogDensityProblems.logdensity(ldf, zeros(3))
+    end
+
     @testset "capabilities" begin
         @model f() = x ~ Normal()
         model = f()