feat(evidence): structure v3 domain rules for review

harness/tests/test_corpus_pipeline.py

@@ -165,6 +165,50 @@ def test_pipeline_embeds_validated_curated_evidence_as_semantic_fragments(tmp_pa
     assert vectors.records[0].record.metadata["provenance"]["source_file"] == "source/paziente.md"
 
 
+def test_pipeline_strips_domain_rule_presentation_before_embedding(tmp_path):
+    rule = (
+        "Il dominio Ablazione descrive la procedura; indicazioni principali: fibrillazione "
+        "atriale; appartiene all'universo procedurale. Fact centrale: clinical.fact_ablazione."
+    )
+    evidence = CuratedEvidence.model_validate(
+        {
+            "schema_version": 3,
+            "id": "evidence:dominio-ablazione",
+            "title": "Dominio Ablazione",
+            "kind": "domain",
+            "purposes": ["disambiguation"],
+            "language": "it",
+            "provenance": {
+                "source_file": "source/ablazione.md",
+                "source_sha256": "sha256:" + "a" * 64,
+                "supporting_excerpts": ["Il dominio Ablazione descrive la procedura."],
+            },
+            "payload": {"rule": rule},
+        }
+    )
+    source_item = SourceObject(
+        source_id="fs:curated-domain",
+        uri="file:///safe/curated/domain/dominio-ablazione.md",
+        fingerprint="sha256:" + "b" * 64,
+        metadata={"relative_path": "curated/domain/dominio-ablazione.md"},
+    )
+    embedder = Embedder()
+
+    result = pipeline(
+        tmp_path,
+        Source([(source_item, dump_curated_markdown(evidence))]),
+        embedder=embedder,
+        vectors=Vectors(),
+        policy=ChunkPolicy(version="chunk-v1", max_chars=4000),
+    ).run()
+
+    embedded = embedder.calls[0]
+    assert result.status == "succeeded"
+    assert f"Regola: {rule}" in embedded
+    assert "tht:raw-rule" not in embedded
+    assert "### Fact centrale:" not in embedded
+
+
 def test_pipeline_exposes_atomic_content_review_code_when_candidate_is_blocked(tmp_path):
     evidence = CuratedEvidence.model_validate(
         {

harness/tests/test_evidence_authoring.py

@@ -367,9 +367,9 @@ def test_migrate_workspace_evidence_rewrites_v1_units_as_v3_without_a_model_call
     assert report.unchanged == ()
     assert migrated.schema_version == 3
     assert migrated.payload.rule == "La fascia pediatrica comprende i minori."
-    assert "## Regola\n\nLa fascia pediatrica comprende i minori." in curated_path.read_text(
-        encoding="utf-8",
-    )
+    text = curated_path.read_text(encoding="utf-8")
+    assert "## Regola\n\n<!-- tht:raw-rule:" in text
+    assert "La fascia pediatrica comprende i minori." in text
     assert report.findings == ()
 
 
@@ -393,6 +393,41 @@ def test_migrate_workspace_evidence_rewrites_v2_units_as_table_free_v3(tmp_path)
     assert not any(line.startswith("|") for line in text.splitlines())
 
 
+def test_migrate_workspace_evidence_rewrites_legacy_v3_rule_presentation(tmp_path):
+    source_text = "I pazienti sotto i 18 anni sono pediatrici."
+    evidence = _evidence(source_text).model_copy(update={"schema_version": 3})
+    _write_workspace(tmp_path, evidence, source_text)
+    curated_path = tmp_path / "evidence" / "curated" / "domain" / "fascia-pediatrica.md"
+    legacy_text = curated_path.read_text(encoding="utf-8")
+    legacy_text = legacy_text.replace(
+        "## Regola\n\n<!-- tht:raw-rule:",
+        "## Regola\n\n<!-- tht:legacy-raw-rule:",
+        1,
+    )
+    # Recreate the exact pre-structured v3 body from the canonical value.
+    start = legacy_text.index("<!-- tht:field:rule -->")
+    end = legacy_text.index("<!-- /tht:field:rule -->", start)
+    legacy_rule = (
+        "<!-- tht:field:rule -->\n"
+        "## Regola\n\n"
+        f"{evidence.payload.rule}\n"
+    )
+    curated_path.write_text(
+        legacy_text[:start] + legacy_rule + legacy_text[end:],
+        encoding="utf-8",
+    )
+
+    report = migrate_workspace_evidence(tmp_path, git_status=lambda _: ())
+
+    migrated_text = curated_path.read_text(encoding="utf-8")
+    assert report.migrated == ("evidence:fascia-pediatrica",)
+    assert report.unchanged == ()
+    assert "## Regola\n\n<!-- tht:raw-rule:" in migrated_text
+    assert load_curated_tree(tmp_path / "evidence" / "curated")[0].payload.rule == (
+        evidence.payload.rule
+    )
+
+
 def test_migrate_workspace_evidence_rejects_dirty_curated_files_in_a_nested_workspace(tmp_path):
     subprocess.run(["git", "init", "--quiet", str(tmp_path)], check=True)
     workspace_root = tmp_path / "psd-clinical"

harness/tests/test_evidence_canonical.py

@@ -262,6 +262,34 @@ def test_v3_curated_markdown_replaces_frontmatter_tables_with_readable_sections(
     assert parsed == evidence
 
 
+def test_v3_domain_rule_is_structured_without_changing_its_canonical_text(tmp_path):
+    rule = (
+        "Il dominio Ablazione descrive la procedura; indicazioni principali: fibrillazione "
+        "atriale, flutter; appartiene all'universo procedurale. Fact centrale: "
+        "`clinical.fact_ablazione`, collegata a `clinical.dim_patient`. Granularità: una riga "
+        "per procedura. Domande tipiche: numero di procedure per anno; pazienti distinti per "
+        "anno; distribuzione per indicazione."
+    )
+    base = _domain_evidence_v3()
+    evidence = base.model_copy(
+        update={"payload": base.payload.model_copy(update={"rule": rule})},
+    )
+
+    text = dump_curated_markdown(evidence)
+    parsed = parse_curated_markdown(
+        text,
+        path=tmp_path / "curated" / "domain" / "dominio-ablazione.md",
+    )
+
+    assert "## Regola\n\n<!-- tht:raw-rule:" in text
+    assert "- Il dominio Ablazione descrive la procedura;" in text
+    assert "- **indicazioni principali:** fibrillazione atriale, flutter;" in text
+    assert "### Fact centrale:" in text
+    assert "### Granularità:" in text
+    assert "### Domande tipiche:" in text
+    assert parsed.payload.rule == rule
+
+
 def test_v3_curated_markdown_renders_enum_values_as_a_list_instead_of_a_table(tmp_path):
     evidence = CuratedEvidence.model_validate({
         **COMMON,
@@ -294,6 +322,17 @@ def test_v3_curated_markdown_rejects_visible_metadata_that_drifted_from_canonica
         parse_curated_markdown(text)
 
 
+def test_v3_curated_markdown_rejects_rule_presentation_that_drifted_from_raw_text():
+    text = dump_curated_markdown(_domain_evidence_v3()).replace(
+        "La fact centrale è",
+        "La tabella centrale è",
+        1,
+    )
+
+    with pytest.raises(ValueError, match="not canonical"):
+        parse_curated_markdown(text)
+
+
 def test_v2_curated_markdown_renders_domain_content_in_the_markdown_body(tmp_path):
     evidence = _domain_evidence_v2()
     path = tmp_path / "curated" / "domain" / "dominio-ablazione.md"