fix: snap cross-repo resolver confidence scores to the INFERRED rubric + generalize the guard test

graphify/cross_repo_calls.py

@@ -235,7 +235,7 @@ def link_cross_repo_member_calls(merged: "nx.Graph") -> int:
                 relation="calls",
                 context="cross_repo",
                 confidence="INFERRED",
-                confidence_score=0.8,
+                confidence_score=0.85,
                 source_file=str(caller_data.get("source_file") or ""),
                 source_location=entry.get("line"),
                 weight=1.0,

graphify/cross_repo_types.py

@@ -65,7 +65,7 @@ def link_shared_type_declarations(merged: "nx.Graph") -> int:
                     relation=SHARED_TYPE_RELATION,
                     context="cross_repo",
                     confidence="INFERRED",
-                    confidence_score=0.9,
+                    confidence_score=0.85,
                     source_file=str(merged.nodes[left].get("source_file") or ""),
                     weight=1.0,
                     _src=left,

graphify/extract.py

@@ -3923,7 +3923,7 @@ def _key(label: str) -> str:
             "relation": relation,
             "context": "call",
             "confidence": "EXTRACTED" if type_qualified else "INFERRED",
-            "confidence_score": 1.0 if type_qualified else 0.8,
+            "confidence_score": 1.0 if type_qualified else 0.85,
             "source_file": rc.get("source_file", ""),
             "source_location": rc.get("source_location"),
             "weight": 1.0,
@@ -4215,7 +4215,7 @@ def _key(label: str) -> str:
             "relation": "calls",
             "context": "call",
             "confidence": "EXTRACTED" if type_qualified else "INFERRED",
-            "confidence_score": 1.0 if type_qualified else 0.8,
+            "confidence_score": 1.0 if type_qualified else 0.85,
             "source_file": rc.get("source_file", ""),
             "source_location": rc.get("source_location"),
             "weight": 1.0,
@@ -4381,7 +4381,7 @@ def _key(label: str) -> str:
             "relation": relation,
             "context": "call",
             "confidence": "EXTRACTED" if type_qualified else "INFERRED",
-            "confidence_score": 1.0 if type_qualified else 0.8,
+            "confidence_score": 1.0 if type_qualified else 0.85,
             "source_file": src_file,
             "source_location": rc.get("source_location"),
             "weight": 1.0,
@@ -4603,7 +4603,7 @@ def _park_if_absent(type_name: str | None, caller_node: dict | None, rc: dict) -
             "relation": "calls",
             "context": "call",
             "confidence": "EXTRACTED" if type_qualified else "INFERRED",
-            "confidence_score": 1.0 if type_qualified else 0.8,
+            "confidence_score": 1.0 if type_qualified else 0.85,
             "source_file": src_file,
             "source_location": rc.get("source_location"),
             "weight": 1.0,
@@ -4813,7 +4813,7 @@ def _method_on_type_or_bases(type_nid: str, callee_key: str) -> str | None:
                 "relation": "calls",
                 "context": "call",
                 "confidence": "EXTRACTED" if exact else "INFERRED",
-                "confidence_score": 1.0 if exact else 0.8,
+                "confidence_score": 1.0 if exact else 0.85,
                 "source_file": raw_call.get("source_file", ""),
                 "source_location": raw_call.get("source_location"),
                 "weight": 1.0,
@@ -5004,7 +5004,7 @@ def _field_type_up_chain(cls, receiver):
             "relation": relation,
             "context": "call",
             "confidence": "EXTRACTED" if type_qualified else "INFERRED",
-            "confidence_score": 1.0 if type_qualified else 0.8,
+            "confidence_score": 1.0 if type_qualified else 0.85,
             "source_file": src_file,
             "source_location": rc.get("source_location"),
             "weight": 1.0,

graphify/interface_dispatch.py

@@ -150,7 +150,7 @@ def resolve_interface_dispatch(
                 "relation": DISPATCH_RELATION,
                 "context": "call",
                 "confidence": "INFERRED",
-                "confidence_score": 0.9,
+                "confidence_score": 0.85,
                 "source_file": str(impl_node.get("source_file", "")),
                 "source_location": impl_node.get("source_location"),
                 "weight": 1.0,

tests/test_cross_repo_member_calls.py

@@ -98,6 +98,7 @@ def test_a_parked_call_binds_to_the_one_declaration_in_another_repo():
     data = G.edges["a::app_run", "b::greeter_greet"]
     assert data["relation"] == "calls"
     assert data["confidence"] == "INFERRED"
+    assert data["confidence_score"] == 0.85
     assert data["context"] == "cross_repo"
     assert data["source_location"] == "L10"
 

tests/test_cross_repo_shared_types.py

@@ -67,6 +67,7 @@ def test_same_namespace_and_name_across_repos_are_linked(tmp_path):
     endpoints = {links[0]["source"], links[0]["target"]}
     assert endpoints == {"svc_a::evt", "svc_b::evt"}
     assert links[0]["confidence"] == "INFERRED"
+    assert links[0]["confidence_score"] == 0.85
 
 
 def test_same_name_in_different_namespaces_is_not_linked(tmp_path):

tests/test_csharp_interface_dispatch.py

@@ -76,6 +76,10 @@ def _reachable(r, start: str) -> set[str]:
 def test_single_implementer_links_the_interface_method(tmp_path):
     dispatch, r = _extract(tmp_path, _INJECTED)
     assert (_find(r, ".Build()", "ireport"), _find(r, ".Build()", "report_report")) in dispatch
+    for edge in r["edges"]:
+        if edge["relation"] == "dispatches_to":
+            assert edge["confidence"] == "INFERRED"
+            assert edge["confidence_score"] == 0.85
 
 
 def test_chain_through_an_injected_dependency_becomes_reachable(tmp_path):

tests/test_inferred_confidence_rubric.py

@@ -18,6 +18,7 @@
 what the extractor knows; snapping the scores onto the documented scale fixes
 the stated violation without making that call.
 """
+import ast
 from pathlib import Path
 
 import pytest
@@ -63,19 +64,78 @@ def test_extracted_and_ambiguous_defaults_are_unchanged():
 # No emission site ships an off-rubric literal
 # ---------------------------------------------------------------------------
 
-@pytest.mark.parametrize("rel_path", [
-    "extract.py",
-    "symbol_resolution.py",
-    "extractors/engine.py",
-    "extractors/resolution.py",
+def _literal_scores(expr):
+    """Read only emitted literals, including both branches of a ternary.
+
+    Calls and lookups are consumers of scores, not literal emission sites.
+    Parsing Python also avoids matching examples in strings or comments.
+    """
+    if isinstance(expr, ast.Constant) and type(expr.value) in (int, float):
+        yield expr.value
+    elif isinstance(expr, ast.IfExp):
+        yield from _literal_scores(expr.body)
+        yield from _literal_scores(expr.orelse)
+
+
+def _emitted_score_exprs(tree):
+    for node in ast.walk(tree):
+        if isinstance(node, ast.Dict):
+            for key, value in zip(node.keys, node.values):
+                if isinstance(key, ast.Constant) and key.value == "confidence_score":
+                    yield value
+        elif isinstance(node, ast.keyword) and node.arg == "confidence_score":
+            yield node.value
+        elif isinstance(node, ast.Assign) and any(
+            isinstance(target, ast.Name) and target.id == "confidence_score"
+            for target in node.targets
+        ):
+            yield node.value
+        elif isinstance(node, ast.AnnAssign) and (
+            isinstance(node.target, ast.Name) and node.target.id == "confidence_score"
+        ) and node.value is not None:
+            yield node.value
+
+
+def test_no_module_hardcodes_an_off_rubric_score():
+    """Cover every Python module and spelling, not a fixed list of resolvers.
+
+    EXTRACTED/1.0 and AMBIGUOUS/0.2 are valid too. Keep the separate
+    runtime checks below to verify that INFERRED edges use the INFERRED set.
+    """
+    allowed = RUBRIC | {1.0, 0.2}
+    violations = []
+    for path in sorted(SRC.rglob("*.py")):
+        tree = ast.parse(path.read_text(encoding="utf-8"), filename=str(path))
+        for expr in _emitted_score_exprs(tree):
+            for score in _literal_scores(expr):
+                if score not in allowed:
+                    violations.append(f"{path.relative_to(SRC)}:{expr.lineno}: {score}")
+    assert not violations, "Off-rubric emitted scores:\n" + "\n".join(violations)
+
+
+@pytest.mark.parametrize("source", [
+    '{"confidence_score": 0.8}',
+    'emit(confidence_score=0.8)',
+    'confidence_score = 0.8',
+    'confidence_score: float = 0.8',
+    '{"confidence_score": 1.0 if exact else 0.8}',
 ])
-def test_no_module_hardcodes_an_off_rubric_inferred_score(rel_path):
-    """0.8 was the value in the tree and is not on the scale. Catch it and the
-    forbidden 0.5 as literals, so a future edit cannot reintroduce either."""
-    text = (SRC / rel_path).read_text(encoding="utf-8")
-    for forbidden in ('"confidence_score": 0.8,', '"confidence_score": 0.5,',
-                      "confidence_score = 0.8\n", "confidence_score = 0.5\n"):
-        assert forbidden not in text, f"{rel_path} still emits {forbidden.strip()}"
+def test_emission_guard_recognizes_all_literal_spellings(source):
+    scores = [
+        score
+        for expr in _emitted_score_exprs(ast.parse(source))
+        for score in _literal_scores(expr)
+    ]
+    assert 0.8 in scores
+
+
+def test_emission_guard_ignores_comments_strings_and_score_consumers():
+    tree = ast.parse(
+        '# "confidence_score": 0.8\n'
+        'example = \'{"confidence_score": 0.8}\'\n'
+        'score = edge.get("confidence_score", 0.5)\n'
+    )
+    assert list(_emitted_score_exprs(tree)) == []
 
 
 # ---------------------------------------------------------------------------

tests/test_swift_cross_file_calls.py

@@ -91,7 +91,7 @@ def test_swift_cross_file_member_calls_have_correct_confidence_and_resolve(tmp_p
         if e.get("relation") != "calls":
             continue
         if tgt_label in inferred_targets:
-            assert e["confidence"] == "INFERRED" and e["confidence_score"] == 0.8
+            assert e["confidence"] == "INFERRED" and e["confidence_score"] == 0.85
             assert e["target"] in node_ids and src_by_id.get(e["target"])
             seen_inferred.add(tgt_label)
         elif tgt_label in extracted_targets:
@@ -295,7 +295,7 @@ def test_environment_attribute_typed_receiver_resolves(tmp_path: Path):
                  and _label(result, e["target"]) == ".reset()"), None)
     assert edge is not None, "store.reset() must resolve to Store.reset"
     assert _label(result, edge["source"]) == ".go()"
-    assert edge["confidence"] == "INFERRED" and edge["confidence_score"] == 0.8
+    assert edge["confidence"] == "INFERRED" and edge["confidence_score"] == 0.85
 
 
 def test_environment_keypath_and_dotted_forms_are_skipped(tmp_path: Path):
@@ -368,7 +368,7 @@ def test_factory_returned_receiver_resolves(tmp_path: Path):
     assert (".local()", "calls", ".go()") in calls    # method-local receiver
     for e in result["edges"]:
         if e.get("relation") == "calls" and _label(result, e["target"]) == ".go()":
-            assert e["confidence"] == "INFERRED" and e["confidence_score"] == 0.8
+            assert e["confidence"] == "INFERRED" and e["confidence_score"] == 0.85
 
 
 def test_factory_receiver_resolves_through_cross_file_extension(tmp_path: Path):

tests/test_swift_protocol_dispatch.py

@@ -105,7 +105,7 @@ def test_dispatch_edge_shape(tmp_path):
     _, r = _extract(tmp_path, _INJECTED)
     edge = next(e for e in r["edges"] if e["relation"] == "dispatches_to")
     assert edge["confidence"] == "INFERRED"
-    assert edge["confidence_score"] == 0.9
+    assert edge["confidence_score"] == 0.85
     assert edge["context"] == "call"
     assert edge["source_file"].endswith("RemoteStore.swift")