fix(build): normalize legacy numeric edge confidences to INFERRED on load

A pre-enum graph.json stored a numeric edge confidence (a float) instead of a
string tag, which tripped the per-edge 'invalid confidence' validator warning on
every incremental reload. _fold_edge_aliases now moves a numeric confidence into
confidence_score (when none is present) and sets the tag to INFERRED — never
EXTRACTED, since a recovered legacy float is not structural provenance. Modern
string tags are untouched (no churn) and the raw float survives in
confidence_score.

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

graphify/build.py

@@ -210,9 +210,23 @@ def _fold_edge_aliases(edge: dict) -> None:
     is not provenance) and never a threshold mapping of the float. The
     ``confidence_score`` key itself is NOT popped: it is a legitimate companion
     field that the edge loop sanitizes and to_json round-trips.
+
+    A NUMERIC ``confidence`` (pre-enum graphs stored the LLM pass's float —
+    1.0/0.95/0.9/0.85 — directly in the field) normalizes to ``INFERRED``:
+    numeric confidences only ever came from the LLM semantic pass, and
+    LLM-derived edges are INFERRED by definition. The original float moves to
+    ``confidence_score`` unless an explicit one is already present (the
+    companion field is the authority). Without this fold, every reload of a
+    pre-enum graph re-warns once per legacy edge, forever. ``bool`` is
+    excluded despite subclassing ``int``: ``True`` is not a score.
     """
     if not edge.get("relation") and isinstance(edge.get("type"), str) and edge["type"]:
         edge["relation"] = edge.pop("type")
+    _conf = edge.get("confidence")
+    if isinstance(_conf, (int, float)) and not isinstance(_conf, bool):
+        if edge.get("confidence_score") is None:
+            edge["confidence_score"] = float(_conf)
+        edge["confidence"] = "INFERRED"
     if not edge.get("confidence") and edge.get("confidence_score") is not None:
         edge["confidence"] = "INFERRED"
 

tests/test_build.py

@@ -172,6 +172,62 @@ def test_legacy_edge_type_confidence_score_aliases_folded():
     assert "type" not in data
 
 
+def test_legacy_numeric_confidence_normalized_to_inferred(capsys):
+    """Pre-enum graphs stored the LLM pass's float directly in `confidence`
+    (1.0/0.95/0.9/0.85). Reloading such a graph must not warn once per edge on
+    every load — the number normalizes to INFERRED (numeric confidences only
+    ever came from the LLM semantic pass, and LLM-derived edges are INFERRED
+    by definition) with the original float preserved in confidence_score."""
+    ext = {"nodes": [{"id": "n1", "label": "A", "file_type": "code", "source_file": "a.py"},
+                     {"id": "n2", "label": "B", "file_type": "code", "source_file": "b.py"},
+                     {"id": "n3", "label": "C", "file_type": "code", "source_file": "c.py"}],
+           "edges": [{"source": "n1", "target": "n2", "relation": "calls",
+                      "confidence": 0.95, "source_file": "a.py"},
+                     {"source": "n2", "target": "n3", "relation": "references",
+                      "confidence": 1.0, "source_file": "b.py"}],
+           "input_tokens": 0, "output_tokens": 0}
+    G = build_from_json(ext)
+    err = capsys.readouterr().err
+    assert "invalid confidence" not in err
+    assert "Extraction warning" not in err
+    d12 = edge_data(G, "n1", "n2")
+    assert d12["confidence"] == "INFERRED"
+    assert d12["confidence_score"] == 0.95
+    d23 = edge_data(G, "n2", "n3")
+    assert d23["confidence"] == "INFERRED"
+    assert d23["confidence_score"] == 1.0
+
+
+def test_legacy_numeric_confidence_existing_score_wins():
+    """A numeric `confidence` next to an explicit `confidence_score` must not
+    overwrite the explicit score — the companion field is the authority."""
+    ext = {"nodes": [{"id": "n1", "label": "A", "file_type": "code", "source_file": "a.py"},
+                     {"id": "n2", "label": "B", "file_type": "code", "source_file": "b.py"}],
+           "edges": [{"source": "n1", "target": "n2", "relation": "calls",
+                      "confidence": 0.9, "confidence_score": 0.4, "source_file": "a.py"}],
+           "input_tokens": 0, "output_tokens": 0}
+    G = build_from_json(ext)
+    data = edge_data(G, "n1", "n2")
+    assert data["confidence"] == "INFERRED"
+    assert data["confidence_score"] == 0.4
+
+
+def test_legacy_numeric_confidence_links_spelling_reload(capsys):
+    """The on-disk shape of the defect: a NetworkX-serialized graph.json
+    (`links` spelling) from a pre-enum version reloads without a validator
+    warning per edge and its edges land INFERRED."""
+    raw = {"nodes": [{"id": "n1", "label": "A", "file_type": "code", "source_file": "a.py"},
+                     {"id": "n2", "label": "B", "file_type": "code", "source_file": "b.py"}],
+           "links": [{"source": "n1", "target": "n2", "relation": "calls",
+                      "confidence": 0.85, "source_file": "a.py"}]}
+    G = build_from_json(raw)
+    err = capsys.readouterr().err
+    assert "invalid confidence" not in err
+    data = edge_data(G, "n1", "n2")
+    assert data["confidence"] == "INFERRED"
+    assert data["confidence_score"] == 0.85
+
+
 def test_node_alias_canonical_field_wins():
     """#2194: when both the canonical field and its alias are present, the
     canonical value wins and the alias key is left untouched."""