diff --git a/include/daScript/ast/compilation_errors.h b/include/daScript/ast/compilation_errors.h
index f1da01ed65..3bcb899697 100644
--- a/include/daScript/ast/compilation_errors.h
+++ b/include/daScript/ast/compilation_errors.h
@@ -353,7 +353,7 @@ namespace das
     ,   exceeds_new_argument                                        =   30509    // 1 site(s)
     ,   exceeds_structure                                           =   30510    // 1 site(s)
     ,   exceeds_tuple_index                                         =   30511    // 1 site(s)
-    ,   exceeds_type                                                =   30512    // 2 site(s)
+    ,   exceeds_type                                                =   30512    // 3 site(s)
     ,   exceeds_type_alias                                          =   30513    // 1 site(s)
     ,   exceeds_typeinfo_sizeof                                     =   30514    // 1 site(s)
     ,   exceeds_constant_range                                      =   30515    // 1 site(s)
diff --git a/include/daScript/simulate/REVIEW.md b/include/daScript/simulate/REVIEW.md
index fb59b65ea4..6c056c670c 100644
--- a/include/daScript/simulate/REVIEW.md
+++ b/include/daScript/simulate/REVIEW.md
@@ -24,6 +24,13 @@ C++ half never opens the daslib checklist on its own.
 - **A diff that changes `KeyHash` (`runtime_table.h`) or `WrapsBuiltinValue` (`cast.h`) states
   in its own PR description which key types change hash value.**
 
+- **A `cvt_*` inline in `aot.h` takes and returns `vec4f`, and none of them is overloaded.**
+  `vec4f` is the SIMD register; the `vec2`/`vec3`/`vec4` types are structs of scalars, so a
+  concrete return spills the lanes through `v_extract_*` and the next conversion reloads them.
+  Overloading is what forced that once: every flavor converts from `vec4f`, so an overload set
+  fed a `vec4f` result is ambiguous - the emitter writes `cvt_pass(cvt_uint3(..))` for
+  `x |> uint3 |> int3`. One name per conversion, `vec4f` throughout, keeps both.
+
 - **A diff that makes the hot path cost more per evaluated expression in the build the repo
   ships is a defect.** The hot path is a `SimNode::eval*` method, any helper such a method
   calls on every evaluation, the dispatchers `Context::callOrFastcall` /
diff --git a/include/daScript/simulate/aot.h b/include/daScript/simulate/aot.h
index 23b5497835..2c53cea7d3 100644
--- a/include/daScript/simulate/aot.h
+++ b/include/daScript/simulate/aot.h
@@ -4517,12 +4517,7 @@ namespace das {
     __forceinline vec4f cvt_uint3 ( float3 f ) { return v_cast_vec4f(v_cvt_vec4i(f)); }
     __forceinline vec4f cvt_uint4 ( float4 f ) { return v_cast_vec4f(v_cvt_vec4i(f)); }
 
-    __forceinline vec4f cvt_pass ( int2 i )  { return i; }
-    __forceinline vec4f cvt_pass ( int3 i )  { return i; }
-    __forceinline vec4f cvt_pass ( int4 i )  { return i; }
-    __forceinline vec4f cvt_pass ( uint2 i ) { return i; }
-    __forceinline vec4f cvt_pass ( uint3 i ) { return i; }
-    __forceinline vec4f cvt_pass ( uint4 i ) { return i; }
+    __forceinline vec4f cvt_pass ( vec4f i ) { return i; }
 
     __forceinline int32_t class_rtti_size ( void * ptr ) {
        auto pti = (TypeInfo **)ptr;
diff --git a/src/ast/ast_infer_type_function.cpp b/src/ast/ast_infer_type_function.cpp
index f7a786fb53..d16170b847 100644
--- a/src/ast/ast_infer_type_function.cpp
+++ b/src/ast/ast_infer_type_function.cpp
@@ -557,6 +557,15 @@ namespace das {
             }
             padded.push_back(newArg);
         }
+        for (int ai = p; ai != k; ++ai) {
+            const auto & argType = winner->arguments[ai]->type;
+            if (!argType->isAutoOrAlias()) continue;
+            AliasMap aliases;
+            OptionsMap options;
+            if (!isMatchingArgument(winner, argType, padded[ai - p]->type, useGenerics, true, &aliases, &options)) {
+                return false;
+            }
+        }
         for (int ai = k - 1; ai >= p; --ai) {
             expr->arguments.insert(expr->arguments.begin() + p, padded[ai - p]);
             types.insert(types.begin() + p, padded[ai - p]->type);
@@ -1908,7 +1917,6 @@ namespace das {
                         if (!anyFailed)
                             break;
                         if (totalAliases == aliases.size()) {
-                            DAS_ASSERTF(0, "we should not be here. function matched arguments!");
                             break;
                         }
                     }
diff --git a/src/ast/ast_lint.cpp b/src/ast/ast_lint.cpp
index 6adf6364d3..b66e786856 100644
--- a/src/ast/ast_lint.cpp
+++ b/src/ast/ast_lint.cpp
@@ -1109,6 +1109,10 @@ namespace das {
                 program->error("[init] is disabled in the options or CodeOfPolicies",  "", "",
                     fn->at, CompilationError::cant_function);
             }
+            if ( !fn->result->isAutoOrAlias() && fn->result->getSizeOf64()>0x7fffffff ) {
+                program->error("function result of '" + fn->name + "' is too big", "", "",
+                    fn->at, CompilationError::exceeds_type);
+            }
         }
         virtual FunctionPtr visit ( Function * fn ) override {
             // Derived class ctor: every CFG path must call super(...) exactly once,
diff --git a/src/ast/ast_typedecl.cpp b/src/ast/ast_typedecl.cpp
index b66d83d4e1..dce905eb21 100644
--- a/src/ast/ast_typedecl.cpp
+++ b/src/ast/ast_typedecl.cpp
@@ -3476,6 +3476,8 @@ namespace das
     uint64_t TypeDecl::getCountOf64() const {
         uint64_t size = 1;
         for ( const TypeDecl * t = this; t && t->baseType==Type::tFixedArray; t = t->firstType ) {
+            // dimAuto (-1) and dimConst (-2) are valid unresolved dims, not counts; uint64_t wraps them
+            if ( t->fixedDim<=0 ) return 0;
             size *= uint64_t(t->fixedDim);
         }
         return size;
diff --git a/tests/language/failed_generic_default_type_mismatch.das b/tests/language/failed_generic_default_type_mismatch.das
new file mode 100644
index 0000000000..6121c0a327
--- /dev/null
+++ b/tests/language/failed_generic_default_type_mismatch.das
@@ -0,0 +1,13 @@
+// A piped call pads a default in for the parameter it shifts across, and lookup admits the
+// candidate on that default's existence alone. When the parameter type is concrete the
+// candidate is still instanced, so the argument-matching loop gives up without progress and
+// the instanced function names the real mistake - which is why that exit falls through.
+options gen2
+expect 30161, 30341
+
+def generic_with_mismatched_default(r : string <- 1, v) => 0
+
+[export]
+def main {
+    generic_with_mismatched_default(){}
+}
diff --git a/tests/language/failed_generic_unknown_arg_type.das b/tests/language/failed_generic_unknown_arg_type.das
new file mode 100644
index 0000000000..13ea43315a
--- /dev/null
+++ b/tests/language/failed_generic_unknown_arg_type.das
@@ -0,0 +1,12 @@
+// A generic whose first argument is an unresolved type expression with a default value. The
+// call matches at lookup; the alias map then stops growing with that argument still
+// unmatched, and the instancing after the loop used to read types the loop never produced.
+options gen2
+expect 30341
+
+def generic_with_unknown_arg_type(r : l() <- 1, v) => 0
+
+[export]
+def call_generic_with_unknown_arg_type {
+    generic_with_unknown_arg_type(){}
+}
diff --git a/tests/language/failed_pointer_math_unsized_pointee.das b/tests/language/failed_pointer_math_unsized_pointee.das
new file mode 100644
index 0000000000..6f2d650730
--- /dev/null
+++ b/tests/language/failed_pointer_math_unsized_pointee.das
@@ -0,0 +1,9 @@
+// Pointer arithmetic whose pointee carries a dimension the parser already rejected. The
+// dimension is negative, so the element count wrapped and the size read as 2^64-48.
+options gen2
+expect 30109, 30239
+
+[export]
+def pointer_math_on_unsized_pointee {
+    var x = new class<block[]>() - 0
+}
diff --git a/tests/language/failed_result_type_by_ref_too_big.das b/tests/language/failed_result_type_by_ref_too_big.das
new file mode 100644
index 0000000000..7a5305ad3b
--- /dev/null
+++ b/tests/language/failed_result_type_by_ref_too_big.das
@@ -0,0 +1,13 @@
+// The oversized result on its own, with no other error to stop compilation first: this one
+// reaches Program::simulate, where the debug-info walk sizes the result and asserts.
+options gen2
+expect 30512
+
+[export]
+def result_too_big_by_ref(p : int[2000000000]?) : int[2000000000]& {
+    unsafe { return *p }
+}
+
+[export]
+def main {
+}
diff --git a/tests/language/failed_result_type_too_big.das b/tests/language/failed_result_type_too_big.das
new file mode 100644
index 0000000000..13630bda24
--- /dev/null
+++ b/tests/language/failed_result_type_too_big.das
@@ -0,0 +1,13 @@
+// An oversized function result. The size gate covered locals, arguments, structures and
+// aliases but not the result, so the type reached the debug-info walk and asserted there.
+options gen2
+expect 30512
+
+// the reported shape - the make-value on the path is diagnosed as well. the crash itself is
+// in failed_result_type_by_ref_too_big.das, where no earlier error stops the walk
+[export]
+def result_too_big => <- tuple<int[2000000000]>(uninitialized)
+
+[export]
+def main {
+}
diff --git a/tests/language/vec_constructors.das b/tests/language/vec_constructors.das
index 6068c9e8ef..85ed0f09e8 100644
--- a/tests/language/vec_constructors.das
+++ b/tests/language/vec_constructors.das
@@ -6,6 +6,12 @@ require dastest/testing_boost public
 // and type conversions between vector types.
 // VEC_SEP is a built-in constant (default ",") used in vector printing.
 
+// globals, so the constant folder leaves the conversion chains below to codegen
+var g_f2 = float2(1, 2)
+var g_f3 = float3(1, 2, 3)
+var g_f4 = float4(1, 2, 3, 4)
+var g_i3 = int3(1, 2, 3)
+
 def test_vec2(tt : T?; a : auto(vec2) -const -&; isRange = false) {
     let f2_0  = vec2()
     let f2_x  = vec2(1.0, 2.0)
@@ -117,4 +123,19 @@ def test_vec_constructors(t : T?) {
         t |> equal(uint3(float3(1, 2, 3)), uint3(1u, 2u, 3u))
         t |> equal(uint4(float4(1, 2, 3, 4)), uint4(1u, 2u, 3u, 4u))
     }
+    t |> run("chained type conversions") @(t : T?) {
+        // AOT writes these as cvt_pass(cvt_uint3(..)), which resolves only while each cvt_
+        // answers in its vector type - a vec4f argument matches every cvt_pass overload
+        t |> equal(g_f2 |> uint2 |> int2, int2(1, 2))
+        t |> equal(g_f3 |> uint3 |> int3, int3(1, 2, 3))
+        t |> equal(g_f4 |> uint4 |> int4, int4(1, 2, 3, 4))
+        t |> equal(g_f2 |> int2 |> uint2, uint2(1u, 2u))
+        t |> equal(g_f3 |> int3 |> uint3, uint3(1u, 2u, 3u))
+        t |> equal(g_f4 |> int4 |> uint4, uint4(1u, 2u, 3u, 4u))
+        t |> equal(g_i3 |> float3 |> int3, int3(1, 2, 3))
+        t |> equal(g_i3 |> uint3 |> float3, float3(1, 2, 3))
+        // three links: cvt_pass(cvt_pass(cvt_uint3(..))), so a cvt_pass feeds a cvt_pass
+        t |> equal(g_f3 |> uint3 |> int3 |> uint3, uint3(1u, 2u, 3u))
+        t |> equal(g_f2 |> int2 |> uint2 |> int2, int2(1, 2))
+    }
 }