test(build): assert metadata survives demotion + unknown-relation is specific; open 0.9.46 (#2803)

Adds a numeric-metadata-preserved test (surviving specific edge keeps its own
weight/confidence, not the demoted generic's) and an unknown-relation test
(a non-denylisted relation is treated as specific in either arrival order, so
the denylist can't drift into an ordering). Dates 0.9.45, opens 0.9.46.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

CHANGELOG.md

@@ -2,7 +2,11 @@
 
 Full release notes with details on each version: [GitHub Releases](https://github.com/safishamsi/graphify/releases)
 
-## 0.9.45 (unreleased)
+## 0.9.46 (unreleased)
+
+- Fix: when two edges connect the same node pair with different relations, the graph builder now keeps the more specific relation (`calls`, `imports`, `inherits`, ...) instead of letting a generic `references`/`uses`/`mentions` overwrite it; previously a real `calls` could be downgraded to `references` and then dropped from the call graph (#2803, thanks @abhay-codes07).
+
+## 0.9.45 (2026-08-16)
 
 - Fix: `graphify install <platform>` now advances the `.graphify_version` stamp only for the platform it actually (re)writes, instead of stamping every installed platform as current; a platform whose skill content was left untouched keeps its old stamp so its staleness warning stays truthful (#2694, thanks @ousamabenyounes). This completes #2694 (the CLAUDE_CONFIG_DIR half shipped in 0.9.44).
 - Fix: an incremental rebuild no longer collapses the whole graph when the `.graphify_root` marker records a subfolder while stored `source_file` paths are relative to the repo root; the marker is validated against the stored paths before it is trusted as their anchor, so a mismatched marker can't make every unchanged source look deleted (#2603, thanks @catpotd). A genuinely deleted source is still evicted, and incremental ids stay identical to a cold build.

tests/test_relation_collapse_precedence.py

@@ -157,3 +157,28 @@ def test_directed_graphs_get_the_same_protection():
                         directed=True)
     assert G.is_directed()
     assert edge_data(G, "a", "b").get("relation") == "calls"
+
+
+def test_specific_edge_numeric_metadata_survives_the_demotion():
+    """When the generic edge is skipped, the surviving specific edge keeps its OWN
+    weight/confidence, not the demoted generic one's."""
+    G = build_from_json(_extraction([
+        _edge("calls", weight=3.0, confidence="EXTRACTED", confidence_score=1.0),
+        _edge("references", weight=9.0, confidence="INFERRED", confidence_score=0.2),
+    ]))
+    d = edge_data(G, "a", "b")
+    assert d.get("relation") == "calls"
+    assert d.get("weight") == 3.0
+    assert d.get("confidence") == "EXTRACTED"
+    assert d.get("confidence_score") == 1.0
+
+
+def test_unknown_relation_is_treated_as_specific():
+    """A relation not on the generic denylist counts as specific: a generic edge
+    must not overwrite it, and it must not be demoted by the guard — so the
+    denylist can't silently drift into an ordering (either arrival order keeps
+    the unknown one)."""
+    for order in ([_edge("custom_rel"), _edge("references")],
+                  [_edge("references"), _edge("custom_rel")]):
+        G = build_from_json(_extraction(order))
+        assert _relation(G) == "custom_rel", f"order {[e['relation'] for e in order]}"