feat(evidence): prepare curated evidence incrementally

2026-08-24 20:08:27 +02:00
parent 6176410f42
commit f5c7cc6198
10 changed files with 899 additions and 7 deletions
@@ -1,4 +1,5 @@
 import hashlib
+import unicodedata

 import pytest
 from pydantic import ValidationError
@@ -6,14 +7,24 @@ from pydantic import ValidationError
 from tht.evidence import (
    CuratedEvidence,
    EvidenceManifest,
+    EvidencePreparationError,
+    EvidenceRestructurer,
+    RestructureCandidate,
    dump_curated_markdown,
    dump_manifest,
+    load_curated_tree,
    load_manifest,
+    prepare_workspace_evidence,
    validate_workspace_evidence,
 )


 def _evidence(source_text: str, *, review_items=()):
+    normalized_source = unicodedata.normalize(
+        "NFC", source_text.replace("\r\n", "\n").replace("\r", "\n"),
+    )
+    if not normalized_source.endswith("\n"):
+        normalized_source += "\n"
    return CuratedEvidence.model_validate({
        "schema_version": 1,
        "id": "evidence:fascia-pediatrica",
@@ -24,7 +35,7 @@ def _evidence(source_text: str, *, review_items=()):
        "language": "it",
        "provenance": {
            "source_file": "source/domain/patient.md",
-            "source_sha256": "sha256:" + hashlib.sha256(source_text.encode()).hexdigest(),
+            "source_sha256": "sha256:" + hashlib.sha256(normalized_source.encode()).hexdigest(),
            "supporting_excerpts": ["I pazienti sotto i 18 anni sono pediatrici."],
        },
        "review_items": list(review_items),
@@ -279,3 +290,150 @@ def test_workspace_validation_reports_unreadable_or_oversized_sources(tmp_path,
    oversized = validate_workspace_evidence(tmp_path)

    assert [finding.code for finding in oversized.findings] == ["source_oversized"]
+
+
+def test_workspace_validation_rejects_a_source_symlink(tmp_path):
+    source_text = "I pazienti sotto i 18 anni sono pediatrici."
+    _write_workspace(tmp_path, _evidence(source_text), source_text)
+    source_path = tmp_path / "evidence" / "source" / "domain" / "patient.md"
+    outside = tmp_path / "outside.md"
+    outside.write_text(source_text, encoding="utf-8")
+    source_path.unlink()
+    source_path.symlink_to(outside)
+
+    report = validate_workspace_evidence(tmp_path)
+
+    assert [finding.code for finding in report.findings] == ["source_unsafe"]
+
+
+class _Restructurer(EvidenceRestructurer):
+    def __init__(self, candidates):
+        self.candidates = candidates
+        self.requests = []
+
+    def restructure(self, request):
+        self.requests.append(request)
+        return tuple(self.candidates)
+
+
+def _candidate(*, title="Fascia pediatrica", existing_id=None):
+    return RestructureCandidate.model_validate({
+        "schema_version": 1,
+        "existing_id": existing_id,
+        "title": title,
+        "kind": "domain",
+        "purposes": ["disambiguation"],
+        "applies_to": {"concepts": ["fascia pediatrica"]},
+        "language": "it",
+        "supporting_excerpts": ["I pazienti sotto i 18 anni sono pediatrici."],
+        "review_items": [],
+        "payload": {"rule": "La fascia pediatrica comprende i minori."},
+    })
+
+
+def test_prepare_changed_source_uses_one_model_call_and_applies_a_valid_batch(tmp_path):
+    source_text = "I pazienti sotto i 18 anni sono pediatrici."
+    _write_workspace(tmp_path, _evidence(source_text), source_text)
+    source_path = tmp_path / "evidence" / "source" / "domain" / "patient.md"
+    source_path.write_text(source_text + " La regola è revisionata.\n", encoding="utf-8")
+    restructurer = _Restructurer([_candidate(existing_id="evidence:fascia-pediatrica")])
+
+    report = prepare_workspace_evidence(tmp_path, restructurer=restructurer, git_status=lambda _: ())
+
+    assert report.model_calls == 1
+    assert report.changed == ("source/domain/patient.md",)
+    assert report.created == ()
+    assert len(restructurer.requests) == 1
+    assert restructurer.requests[0].previous_units[0].id == "evidence:fascia-pediatrica"
+    assert validate_workspace_evidence(tmp_path).publishable is True
+
+
+def test_prepare_unchanged_source_skips_model_and_does_not_write(tmp_path):
+    source_text = "I pazienti sotto i 18 anni sono pediatrici."
+    _write_workspace(tmp_path, _evidence(source_text), source_text)
+    curated_path = tmp_path / "evidence" / "curated" / "domain" / "fascia-pediatrica.md"
+    original = curated_path.read_text(encoding="utf-8")
+    restructurer = _Restructurer([])
+
+    report = prepare_workspace_evidence(tmp_path, restructurer=restructurer, git_status=lambda _: ())
+
+    assert report.model_calls == 0
+    assert report.unchanged == ("source/domain/patient.md",)
+    assert curated_path.read_text(encoding="utf-8") == original
+    assert restructurer.requests == []
+
+
+def test_prepare_rejects_dirty_curated_state_before_model_call(tmp_path):
+    source_text = "I pazienti sotto i 18 anni sono pediatrici."
+    _write_workspace(tmp_path, _evidence(source_text), source_text)
+    restructurer = _Restructurer([])
+
+    with pytest.raises(EvidencePreparationError, match="authoring_worktree_dirty"):
+        prepare_workspace_evidence(
+            tmp_path,
+            restructurer=restructurer,
+            git_status=lambda _: (" M evidence/curated/domain/patient.md",),
+        )
+
+    assert restructurer.requests == []
+
+
+def test_prepare_retains_units_from_a_removed_source_as_blocking_orphans(tmp_path):
+    source_text = "I pazienti sotto i 18 anni sono pediatrici."
+    _write_workspace(tmp_path, _evidence(source_text), source_text)
+    (tmp_path / "evidence" / "source" / "domain" / "patient.md").unlink()
+    restructurer = _Restructurer([])
+
+    report = prepare_workspace_evidence(tmp_path, restructurer=restructurer, git_status=lambda _: ())
+
+    assert report.model_calls == 0
+    assert report.orphaned == ("evidence:fascia-pediatrica",)
+    assert (tmp_path / "evidence" / "curated" / "domain" / "fascia-pediatrica.md").is_file()
+    assert [finding.code for finding in report.findings] == ["orphaned_unit"]
+
+
+def test_prepare_preserves_ids_without_orphans_for_a_unique_source_rename(tmp_path):
+    source_text = "I pazienti sotto i 18 anni sono pediatrici."
+    _write_workspace(tmp_path, _evidence(source_text), source_text)
+    source_root = tmp_path / "evidence" / "source" / "domain"
+    (source_root / "patient.md").rename(source_root / "patient-age.md")
+    restructurer = _Restructurer([])
+
+    report = prepare_workspace_evidence(tmp_path, restructurer=restructurer, git_status=lambda _: ())
+
+    assert report.model_calls == 0
+    assert report.orphaned == ()
+    assert report.unchanged == ("source/domain/patient-age.md",)
+    assert validate_workspace_evidence(tmp_path).publishable is True
+
+
+def test_prepare_rejects_an_unknown_existing_id_without_writing_the_batch(tmp_path):
+    source_text = "I pazienti sotto i 18 anni sono pediatrici."
+    _write_workspace(tmp_path, _evidence(source_text), source_text)
+    source_path = tmp_path / "evidence" / "source" / "domain" / "patient.md"
+    source_path.write_text(source_text + " Revisione.\n", encoding="utf-8")
+    original = (tmp_path / "evidence" / "manifest.yaml").read_text(encoding="utf-8")
+    restructurer = _Restructurer([_candidate(existing_id="evidence:not-supplied")])
+
+    with pytest.raises(EvidencePreparationError, match="unknown_existing_id"):
+        prepare_workspace_evidence(tmp_path, restructurer=restructurer, git_status=lambda _: ())
+
+    assert (tmp_path / "evidence" / "manifest.yaml").read_text(encoding="utf-8") == original
+
+
+def test_prepare_marks_an_omitted_prior_unit_for_human_review(tmp_path):
+    source_text = "I pazienti sotto i 18 anni sono pediatrici."
+    _write_workspace(tmp_path, _evidence(source_text), source_text)
+    (tmp_path / "evidence" / "source" / "domain" / "patient.md").write_text(
+        "La classificazione pediatrica non è documentata.\n", encoding="utf-8",
+    )
+    restructurer = _Restructurer([])
+
+    report = prepare_workspace_evidence(tmp_path, restructurer=restructurer, git_status=lambda _: ())
+
+    assert report.model_calls == 1
+    assert [finding.code for finding in report.findings] == [
+        "supporting_excerpt_missing", "unresolved_review_item",
+    ]
+    retained = load_curated_tree(tmp_path / "evidence" / "curated")[0]
+    assert retained.review_items[0].code == "source_no_longer_supports_unit"