feat(evidence): prepare curated evidence incrementally

harness/tests/test_evidence_pi_restructurer.py

@@ -0,0 +1,80 @@
+import json
+from types import SimpleNamespace
+
+import pytest
+
+from tht.evidence import (
+    EvidencePreparationError,
+    PiEvidenceRestructurer,
+    RestructureRequest,
+)
+
+
+def _request():
+    return RestructureRequest.model_validate({
+        "source_file": "source/domain/patient.md",
+        "source_sha256": "sha256:" + "a" * 64,
+        "normalized_text": "I pazienti sotto i 18 anni sono pediatrici.\n",
+    })
+
+
+def _candidate():
+    return {
+        "schema_version": 1,
+        "title": "Fascia pediatrica",
+        "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_pi_restructurer_uses_an_ephemeral_no_tools_invocation(tmp_path, monkeypatch):
+    from tht.evidence import authoring
+
+    calls = []
+
+    def run(argv, **kwargs):
+        calls.append((argv, kwargs))
+        kwargs["stdout"].write(json.dumps({"candidates": [_candidate()]}))
+        return SimpleNamespace(returncode=0)
+
+    monkeypatch.setattr(authoring.subprocess, "run", run)
+    restructurer = PiEvidenceRestructurer("pi-test", tmp_path / "skill.md", timeout_seconds=12)
+
+    candidates = restructurer.restructure(_request())
+
+    assert candidates[0].title == "Fascia pediatrica"
+    argv, kwargs = calls[0]
+    assert argv[:8] == [
+        "pi-test", "--mode", "text", "--print", "--no-session", "--no-tools", "--no-extensions", "--no-context-files",
+    ]
+    assert "--skill" in argv
+    assert any(argument.startswith("@") for argument in argv)
+    assert kwargs["timeout"] == 12
+    assert kwargs["shell"] is False
+    assert kwargs["stderr"] is authoring.subprocess.DEVNULL
+
+
+@pytest.mark.parametrize("result", [
+    SimpleNamespace(returncode=1, stdout="secret", stderr="secret"),
+    SimpleNamespace(returncode=0, stdout="not-json", stderr=""),
+])
+def test_pi_restructurer_returns_a_bounded_error_without_model_output(tmp_path, monkeypatch, result):
+    from tht.evidence import authoring
+
+    def run(*args, **kwargs):
+        kwargs["stdout"].write(result.stdout)
+        return result
+
+    monkeypatch.setattr(authoring.subprocess, "run", run)
+    restructurer = PiEvidenceRestructurer("pi-test", tmp_path / "skill.md")
+
+    with pytest.raises(EvidencePreparationError) as error:
+        restructurer.restructure(_request())
+
+    assert error.value.code in {"pi_restructure_failed", "pi_restructure_invalid"}
+    assert "secret" not in str(error.value)