Fix Swift adapter property assignment flow and initializer applicability

.github/workflows/ci.yml

@@ -100,6 +100,10 @@ jobs:
           python3 scripts/verify-swift-resolved-append.py
           python3 scripts/test-swift-resolved-conversion.py
           python3 scripts/verify-swift-resolved-conversion.py
+          python3 scripts/test-swift-conversion-body.py
+          python3 scripts/verify-swift-conversion-body.py
+          python3 scripts/test-swift-conversion-diagnostics.py
+          python3 scripts/verify-swift-conversion-diagnostics.py
       - name: Test Swift fixture validation controls
         run: python3 scripts/test-swift-fixtures.py
       - name: Test isolated release validation gate

adapters/codeql/swift-resolved-native-v1/patches/ssa-property-rhs-v1.json

@@ -0,0 +1,7 @@
+{
+  "path": "codeql/swift-all/6.8.4/codeql/swift/dataflow/Ssa.qll",
+  "original_sha256": "50259f376c2246c6bcd163088e9644de81e48b8d5de0969e149ca5ef66013d4d",
+  "patched_sha256": "fe435a0656f93afbc813faab91c2544761a776b54f68708e8e2876522e400591",
+  "patch_sha256": "d61a606ed559388abf83babef9828f3d1d9bebc5549e447077636136b94cd75e",
+  "scope": "generic CFG property-getter RHS mapping, independent of fixture identity"
+}

adapters/codeql/swift-resolved-native-v1/patches/ssa-property-rhs-v1.patch

@@ -0,0 +1,11 @@
+--- a/codeql/swift-all/6.8.4/codeql/swift/dataflow/Ssa.qll
++++ b/codeql/swift-all/6.8.4/codeql/swift/dataflow/Ssa.qll
+@@ -176,7 +176,7 @@
+       exists(AssignExpr a, BasicBlocks::BasicBlock bb, int i |
+         this.definesAt(_, bb, i) and
+         a = bb.getNode(i).getNode().asAstNode() and
+-        value.getNode().asAstNode() = a.getSource()
++        value.getAst() = a.getSource()
+       )
+       or
+       exists(BasicBlocks::BasicBlock bb, int blockIndex, NamedPattern np |

docs/swift-conversion-body.md

@@ -0,0 +1,77 @@
+# Swift conversion body and initializer controls
+
+The initial catalog applicability patch removed local summaries and the original
+constant-body false flow. It did not establish ordinary-body preservation. This
+follow-up adds a transparent local Data/String body chain, a matching safe chain,
+and a constant three-parameter String initializer called with real tainted Data.
+The original independent evidence remains unchanged.
+
+The new source is typechecked with pinned Swift 6.3.3 and is never executed.
+One extraction is shared by separately locked stock and patched query packs.
+The source, queries, runtime manifests, runners, and expectations are committed
+before execution. Extraction took 91.163 seconds under the 150-second limit.
+Queries retain 60-second / 2048-MiB requests. These are non-scored diagnostics;
+aggregate memory containment and complete semantics remain unproven.
+
+## Expanded result
+
+| Sink | Intended flow | Stock assisted | Patched catalog |
+| --- | --- | --- | --- |
+| Real raw roundtrip | yes | yes | yes |
+| Real safe roundtrip | no | no | no |
+| Local constant chain | no | yes | no |
+| Local transparent body chain | yes | yes | **missing** |
+| Local safe body chain | no | no | no |
+| Constant wrong-arity String initializer | no | yes | **false flow** |
+
+The stock body's reported flow is not evidence of body propagation: its local
+calls receive summaries. In the patched lane all 13 local calls have no manual
+summaries. A separate query also confirms no other summary attachment. Eight
+real calls retain their summaries. Both remaining failures block qualification
+and the canonical summary pair.
+
+## Separate diagnostics
+
+A preregistered all-node sink configuration records intermediate reachable
+nodes on the same hash-verified database. It reaches the local encoder result
+and the `.value` input of the local String initializer, but not the final local
+body-chain sink. This is diagnostic evidence of where propagation is missing,
+not proof of a specific internal cause or a production flow configuration.
+
+A separate structural query witnesses a native additional taint edge from the
+`data` argument to the constant wrong-arity initializer call. The pinned
+`InitializerFromDataStep` admits any `data` argument. This is independent of
+catalog summary attachment, which is absent on that call.
+
+The next preregistered experiment filters only the additional heuristic for
+resolved Swift.String data initializers, retaining the proved Foundation
+initializer identity. Other owners remain stock; ordinary method bodies and
+summary steps remain available. The ordinary-body failure remains a separate
+blocker even if the false flow is removed. No fixture-specific barrier,
+unsupported classification, score activation, or release is authorized by a
+successful diagnostic command.
+
+The resolved-initializer experiment completed within its query limit. The
+catalog-only lane retained real sink 20 and false sink 42. The additional
+resolved filter retained only real sink 20. It removes this false flow but still
+misses body sink 35; conversion qualification remains blocked. Portable body
+verification reports the original two blockers, and the separate diagnostic
+verifier records the narrower correction without rewriting the original result.
+
+## Structural SSA repair
+
+The follow-up resolved-AST/SSA query found the literal assignment on line 11
+connected through an SSA definition to the initializer return, while the
+property-getter assignment on line 12 did not. The pinned `WriteDefinition.assigns`
+compared `value.getNode().asAstNode()` with the assignment RHS, omitting the
+property-getter CFG representation. The isolated `6.8.4-dfb.3` adapter library
+uses the existing `value.getAst()` mapping instead. This one-line structural
+change has no fixture name, location, source, or sink condition.
+
+With that patch, both assignments have the expected SSA and return links.
+The catalog-plus-SSA lane reports real sink 20, legitimate body sink 35, and
+wrong-arity false sink 42. Adding the resolved initializer filter retains only
+20 and 35. All three safe/constant sinks remain absent. This separates the
+body repair from the heuristic correction and preserves every earlier failure.
+Fresh extraction and canonical summary controls remain necessary before
+family qualification; these retained-database results do not activate scores.