feat(evidence): build semantic fragments from typed units

2026-08-24 21:33:48 +02:00
parent 8adc085746
commit 3420c57c8b
9 changed files with 642 additions and 14 deletions
@@ -52,6 +52,59 @@ def test_frontmatter_can_end_at_eof_without_inventing_content():
    assert document.content == ""


+def test_normalize_preserves_validated_curated_evidence_for_semantic_projection():
+    source = SourceObject(
+        source_id="filesystem:fascia-pediatrica",
+        uri="file:///safe/evidence/curated/formula/fascia-pediatrica.md",
+        fingerprint="sha256:" + "a" * 64,
+        metadata={"relative_path": "curated/formula/fascia-pediatrica.md"},
+    )
+    raw = (
+        "---\n"
+        "schema_version: 1\n"
+        "id: evidence:fascia-pediatrica\n"
+        "title: Fascia pediatrica\n"
+        "kind: formula\n"
+        "purposes: [sql_generation]\n"
+        "applies_to:\n"
+        "  columns: [clinical.patient.birth_date]\n"
+        "language: it\n"
+        "provenance:\n"
+        "  source_file: source/paziente.md\n"
+        "  source_sha256: sha256:" + "b" * 64 + "\n"
+        "  supporting_excerpts: [Pazienti con età inferiore a 18 anni.]\n"
+        "review_items: []\n"
+        "formula:\n"
+        "  concept: fascia pediatrica\n"
+        "  columns: [clinical.patient.birth_date]\n"
+        "  sql: CASE WHEN age < 18 THEN 'pediatric' END\n"
+        "---\n"
+    ).encode()
+
+    document = normalize(AcquiredDocument(source=source, content=raw, media_type="text/markdown"), "pipe:v1")
+
+    assert document.content == raw.decode()
+    assert document.title == "Fascia pediatrica"
+    assert document.metadata["curated_evidence"]["id"] == "evidence:fascia-pediatrica"
+    assert document.metadata["curated_evidence"]["payload"]["concept"] == "fascia pediatrica"
+
+
+def test_only_curated_markdown_is_treated_as_canonical_evidence_without_relative_metadata():
+    source = SourceObject(
+        source_id="filesystem:notes",
+        uri="file:///safe/evidence/curated/formula/notes.txt",
+        fingerprint="sha256:" + "a" * 64,
+    )
+
+    document = normalize(
+        AcquiredDocument(source=source, content=b"ordinary note", media_type="text/plain"),
+        "pipe:v1",
+    )
+
+    assert document.content == "ordinary note"
+    assert "curated_evidence" not in document.metadata
+
+
@pytest.mark.parametrize(
    "frontmatter",
    [