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
@@ -2,6 +2,7 @@ from datetime import UTC, datetime, timedelta

 import pytest

+from tht.evidence.canonical import CuratedEvidence, dump_curated_markdown
 from tht.evidence.contracts import AcquiredDocument, SourceObject
 from tht.evidence.corpus.chunk import ChunkPolicy
 from tht.evidence.corpus.models import CanonicalChunk, CanonicalDocument, CorpusManifest
@@ -116,6 +117,131 @@ def pipeline(tmp_path, source, *, embedder=None, vectors=None, model="model-a",
    )


+def test_pipeline_embeds_validated_curated_evidence_as_semantic_fragments(tmp_path):
+    evidence = CuratedEvidence.model_validate(
+        {
+            "schema_version": 1,
+            "id": "evidence:fascia-pediatrica",
+            "title": "Fascia pediatrica",
+            "kind": "formula",
+            "purposes": ["sql_generation"],
+            "applies_to": {"columns": ["clinical.patient.birth_date"]},
+            "language": "it",
+            "provenance": {
+                "source_file": "source/paziente.md",
+                "source_sha256": "sha256:" + "a" * 64,
+                "supporting_excerpts": ["Pazienti con età inferiore a 18 anni."],
+            },
+            "review_items": [],
+            "payload": {
+                "concept": "fascia pediatrica",
+                "columns": ["clinical.patient.birth_date"],
+                "sql": "CASE WHEN age < 18 THEN 'pediatric' END",
+            },
+        }
+    )
+    source_item = SourceObject(
+        source_id="fs:curated-formula",
+        uri="file:///safe/curated/formula/fascia-pediatrica.md",
+        fingerprint="sha256:" + "b" * 64,
+        metadata={"relative_path": "curated/formula/fascia-pediatrica.md"},
+    )
+    embedder = Embedder()
+    vectors = Vectors()
+
+    result = pipeline(
+        tmp_path,
+        Source([(source_item, dump_curated_markdown(evidence))]),
+        embedder=embedder,
+        vectors=vectors,
+        policy=ChunkPolicy(version="chunk-v1", max_chars=4000),
+    ).run()
+
+    assert result.status == "succeeded"
+    assert len(result.manifest.chunks) == 1
+    assert "Formula: Fascia pediatrica" in embedder.calls[0]
+    assert vectors.records[0].record.metadata["evidence_id"] == evidence.id
+    assert vectors.records[0].record.metadata["provenance"]["source_file"] == "source/paziente.md"
+
+
+def test_pipeline_exposes_atomic_content_review_code_when_candidate_is_blocked(tmp_path):
+    evidence = CuratedEvidence.model_validate(
+        {
+            "schema_version": 1,
+            "id": "evidence:formula-lunga",
+            "title": "Formula lunga",
+            "kind": "formula",
+            "purposes": ["sql_generation"],
+            "language": "it",
+            "provenance": {
+                "source_file": "source/paziente.md",
+                "source_sha256": "sha256:" + "a" * 64,
+                "supporting_excerpts": ["Una formula molto lunga."],
+            },
+            "review_items": [],
+            "payload": {"concept": "formula lunga", "columns": [], "sql": "x" * 200},
+        }
+    )
+    source_item = SourceObject(
+        source_id="fs:formula-lunga",
+        uri="file:///safe/curated/formula/formula-lunga.md",
+        fingerprint="sha256:" + "b" * 64,
+        metadata={"relative_path": "curated/formula/formula-lunga.md"},
+    )
+
+    result = pipeline(
+        tmp_path,
+        Source([(source_item, dump_curated_markdown(evidence))]),
+        policy=ChunkPolicy(version="chunk-v1", max_chars=120),
+    ).run()
+
+    assert result.status == "blocked"
+    assert result.published is False
+    assert [item.code for item in result.review_items] == ["atomic_content_too_large"]
+    assert result.review_items[0].field == "formula.sql"
+
+
+def test_job_pipeline_persists_atomic_content_review_item_when_candidate_is_blocked(tmp_path):
+    evidence = CuratedEvidence.model_validate(
+        {
+            "schema_version": 1,
+            "id": "evidence:formula-lunga-job",
+            "title": "Formula lunga",
+            "kind": "formula",
+            "purposes": ["sql_generation"],
+            "language": "it",
+            "provenance": {
+                "source_file": "source/paziente.md",
+                "source_sha256": "sha256:" + "a" * 64,
+                "supporting_excerpts": ["Una formula molto lunga."],
+            },
+            "review_items": [],
+            "payload": {"concept": "formula lunga", "columns": [], "sql": "x" * 200},
+        }
+    )
+    source_item = SourceObject(
+        source_id="fs:formula-lunga-job",
+        uri="file:///safe/curated/formula/formula-lunga-job.md",
+        fingerprint="sha256:" + "b" * 64,
+        metadata={"relative_path": "curated/formula/formula-lunga-job.md"},
+    )
+
+    result = pipeline(
+        tmp_path,
+        Source([(source_item, dump_curated_markdown(evidence))]),
+        policy=ChunkPolicy(version="chunk-v1", max_chars=120),
+    ).run_as_job(
+        workspace_id="demo",
+        workspace_root=tmp_path,
+        config_fingerprint="sha256:" + "1" * 64,
+        input_fingerprint="sha256:" + "2" * 64,
+    )
+
+    assert result.status == "blocked"
+    assert result.published is False
+    assert [item.code for item in result.review_items] == ["atomic_content_too_large"]
+
+
 def test_pipeline_routes_source_io_through_evidence_facade(tmp_path, monkeypatch):
    import tht.evidence.acquisition as evidence_acquisition