fix(evidence): resolve final review findings

harness/tests/test_evidence_formula_migration.py

@@ -1,6 +1,10 @@
 """Migration boundary: legacy formulas become curated evidence or session proposals."""
 
+import hashlib
+
 from tht.evidence import formula_store
+from tht.evidence.authoring import normalize_source_text
+from tht.evidence.canonical import CuratedEvidence
 from tht.evidence.formula_store import ConceptFormula
 
 
@@ -13,17 +17,124 @@ def test_reviewed_formula_migration_has_deterministic_provenance_hash():
         sources=["Regola clinica approvata dal gruppo pediatrico."],
     )
 
+    original_source = """---
+concept: fascia pediatrica
+columns: [clinical.patient.birth_date]
+status: reviewed
+sources:
+  - Regola clinica approvata dal gruppo pediatrico.
+---
+CASE WHEN age < 18 THEN 'pediatrica' ELSE 'adulta' END
+"""
+
     first = formula_store.legacy_formula_to_curated(
-        formula, legacy_path="formulas/fascia-pediatrica-1.sql.md",
+        formula,
+        legacy_path="formulas/fascia-pediatrica-1.sql.md",
+        source_content=original_source,
     )
     second = formula_store.legacy_formula_to_curated(
+        formula,
+        legacy_path="formulas/fascia-pediatrica-1.sql.md",
+        source_content=original_source,
+    )
+
+    assert isinstance(first, CuratedEvidence)
+    assert isinstance(second, CuratedEvidence)
+    assert first.provenance.source_sha256 == second.provenance.source_sha256
+    expected = hashlib.sha256(normalize_source_text(original_source).encode("utf-8")).hexdigest()
+    assert first.provenance.source_sha256 == f"sha256:{expected}"
+
+
+def test_reviewed_formula_without_original_source_fails_closed():
+    formula = ConceptFormula(
+        concept="fascia pediatrica",
+        columns=["clinical.patient.birth_date"],
+        sql="CASE WHEN age < 18 THEN 'pediatrica' ELSE 'adulta' END",
+        status="reviewed",
+        sources=["Regola clinica approvata dal gruppo pediatrico."],
+    )
+
+    outcome = formula_store.legacy_formula_to_curated(
         formula, legacy_path="formulas/fascia-pediatrica-1.sql.md",
     )
 
-    assert first is not None
-    assert second is not None
-    assert first.provenance.source_sha256 == second.provenance.source_sha256
-    assert first.provenance.source_sha256.startswith("sha256:")
+    assert outcome.code == "legacy_formula_requires_manual_review"
+    assert outcome.problems == ("original_source_required",)
+
+
+def test_reviewed_formula_without_verified_supporting_excerpts_fails_closed():
+    formula = ConceptFormula(
+        concept="fascia pediatrica",
+        columns=["clinical.patient.birth_date"],
+        sql="CASE WHEN age < 18 THEN 'pediatrica' ELSE 'adulta' END",
+        status="reviewed",
+        sources=["Nota non presente nel sorgente originale."],
+    )
+    original_source = """---
+concept: fascia pediatrica
+columns: [clinical.patient.birth_date]
+status: reviewed
+sources:
+  - >-
+    Nota non presente nel
+    sorgente originale.
+---
+CASE WHEN age < 18 THEN 'pediatrica' ELSE 'adulta' END
+"""
+
+    outcome = formula_store.legacy_formula_to_curated(
+        formula,
+        legacy_path="formulas/fascia-pediatrica-1.sql.md",
+        source_content=original_source,
+    )
+
+    assert outcome.code == "legacy_formula_requires_manual_review"
+    assert outcome.problems == ("supporting_excerpt_unverified",)
+
+
+def test_reviewed_formula_without_provenance_notes_fails_closed():
+    formula = ConceptFormula(
+        concept="fascia pediatrica",
+        columns=["clinical.patient.birth_date"],
+        sql="CASE WHEN age < 18 THEN 'pediatrica' ELSE 'adulta' END",
+        status="reviewed",
+        sources=[],
+    )
+
+    outcome = formula_store.legacy_formula_to_curated(
+        formula,
+        legacy_path="formulas/fascia-pediatrica-1.sql.md",
+        source_content=formula.dump(),
+    )
+
+    assert outcome.code == "legacy_formula_requires_manual_review"
+    assert outcome.problems == ("supporting_excerpts_required",)
+
+
+def test_reviewed_formula_must_match_the_original_source_record():
+    formula = ConceptFormula(
+        concept="fascia pediatrica",
+        columns=["clinical.patient.birth_date"],
+        sql="CASE WHEN age < 18 THEN 'pediatrica' ELSE 'adulta' END",
+        status="reviewed",
+        sources=["Regola clinica approvata dal gruppo pediatrico."],
+    )
+    different_source = ConceptFormula(
+        concept=formula.concept,
+        columns=formula.columns,
+        sql="CASE WHEN age < 16 THEN 'pediatrica' ELSE 'adulta' END",
+        status=formula.status,
+        sources=formula.sources,
+    ).dump()
+
+    outcome = formula_store.legacy_formula_to_curated(
+        formula,
+        legacy_path="formulas/fascia-pediatrica-1.sql.md",
+        source_content=different_source,
+    )
+
+    assert outcome.code == "legacy_formula_requires_manual_review"
+    assert outcome.problems == ("original_source_mismatch",)
 
 
 def test_incompatible_reviewed_legacy_formula_fails_closed_with_the_original_record():

harness/tests/test_evidence_restructuring_fixture.py

@@ -38,3 +38,22 @@ def test_owner_gate_manual_records_the_narrow_inventory_and_qdrant_baseline_prot
     assert "with_payload:false" in text
     assert "with_payload:true" not in text
     assert "must be repeated through the successful `vector inspect` command" in normalized
+
+
+def test_owner_gate_current_summaries_supersede_stale_pre_inspection_history():
+    root = Path(__file__).parents[2]
+    report = (
+        root / ".superpowers" / "sdd" / "2026-08-24-evidence-restructuring"
+        / "task-12-report.md"
+    ).read_text(encoding="utf-8")
+    state = (root / "PROJECT_STATE.md").read_text(encoding="utf-8")
+
+    current_report = report.split("## Historical record", 1)[0]
+    current_state = state.split("## Modular workflow refactor candidate", 1)[0]
+
+    assert "225 passed, 1 warning" in current_report
+    assert "230 passed, 1 warning" in current_report
+    assert "authorized read-only PSD" in current_report
+    assert "35 moveable source documents" in current_report
+    assert "**225 passed, 1 known pytest deprecation warning**" in current_state
+    assert "**230 passed, 1 known pytest deprecation warning**" in current_state

harness/tests/test_formula.py

@@ -100,7 +100,9 @@ def test_reviewed_legacy_formula_becomes_curated_formula_with_stable_provenance(
     )
 
     migrated = formula_store.legacy_formula_to_curated(
-        formula, legacy_path="formulas/fascia-pediatrica-1.sql.md",
+        formula,
+        legacy_path="formulas/fascia-pediatrica-1.sql.md",
+        source_content=formula.dump(),
     )
 
     assert migrated is not None
@@ -114,7 +116,9 @@ def test_reviewed_legacy_formula_becomes_curated_formula_with_stable_provenance(
     assert migrated.provenance.supporting_excerpts == tuple(formula.sources)
     assert migrated.review_items == ()
     assert formula_store.legacy_formula_to_curated(
-        formula, legacy_path="formulas/fascia-pediatrica-1.sql.md",
+        formula,
+        legacy_path="formulas/fascia-pediatrica-1.sql.md",
+        source_content=formula.dump(),
     ).id == migrated.id
 
 
@@ -124,19 +128,25 @@ def test_reviewed_legacy_formulas_with_the_same_concept_keep_distinct_path_ident
         columns=["clinical.patient.birth_date"],
         sql="CASE WHEN age < 18 THEN 'pediatrica' ELSE 'adulta' END",
         status="reviewed",
+        sources=["Regola legacy revisionata."],
     )
     second = ConceptFormula(
         concept="fascia pediatrica",
         columns=["clinical.patient.birth_date"],
         sql="CASE WHEN age < 16 THEN 'pediatrica' ELSE 'adulta' END",
         status="reviewed",
+        sources=["Regola legacy revisionata."],
     )
 
     first_migration = formula_store.legacy_formula_to_curated(
-        first, legacy_path="formulas/fascia-pediatrica-1.sql.md",
+        first,
+        legacy_path="formulas/fascia-pediatrica-1.sql.md",
+        source_content=first.dump(),
     )
     second_migration = formula_store.legacy_formula_to_curated(
-        second, legacy_path="formulas/fascia-pediatrica-2.sql.md",
+        second,
+        legacy_path="formulas/fascia-pediatrica-2.sql.md",
+        source_content=second.dump(),
     )
 
     assert first_migration is not None

harness/tests/test_preprocess_cli.py

@@ -315,11 +315,35 @@ def test_run_from_config_uses_runtime_identity_workspace_id(monkeypatch, tmp_pat
     command.run_from_config(config)
 
     assert calls["init"]["sparse_language"] == "english"
-    assert callable(calls["init"]["candidate_evaluator"])
+    assert calls["init"]["candidate_evaluator"] is None
     assert calls["run_as_job"]["workspace_id"] == "psd-clinical"
     assert calls["run_as_job"]["input_fingerprint"] != calls["run_as_job"]["config_fingerprint"]
 
 
+@pytest.mark.parametrize("schema_version, source_type", [
+    (1, "filesystem"),
+    (1, "http"),
+    (1, "s3"),
+    (2, "http"),
+    (2, "s3"),
+])
+def test_candidate_evaluation_is_not_required_outside_v2_filesystem_corpora(
+    schema_version, source_type,
+):
+    import tht.cli.preprocess_cmd as command
+
+    cfg = SimpleNamespace(
+        evidence=SimpleNamespace(
+            schema_version=schema_version,
+            source_root=None,
+            sources=[SimpleNamespace(type=source_type)],
+        ),
+        language="it",
+    )
+
+    assert command._candidate_evaluator(cfg, vector_store=object(), embedder=object()) is None
+
+
 def test_preprocess_evidence_gc_json_is_pristine(monkeypatch, tmp_path):
     import tht.cli.preprocess_cmd as command
 

harness/tht/cli/preprocess_cmd.py

@@ -45,8 +45,19 @@ def _evaluation_workspace_root(cfg) -> Path:
     return root
 
 
+def _requires_candidate_evaluation(cfg) -> bool:
+    evidence = cfg.evidence
+    if evidence is None or evidence.schema_version != 2:
+        return False
+    return evidence.source_root is not None or any(
+        source.type == "filesystem" for source in evidence.sources
+    )
+
+
 def _candidate_evaluator(cfg, *, vector_store, embedder):
     """Bind candidate publication to the same read-only retrieval evaluator as the CLI."""
+    if not _requires_candidate_evaluation(cfg):
+        return None
     from tht.evidence.canonical import load_curated_tree
     from tht.evidence.evaluation import evaluate_retrieval, load_evaluation_fixture
 
@@ -77,9 +88,7 @@ def _candidate_evaluator(cfg, *, vector_store, embedder):
 
 def _validate_materialized_curated_corpus(cfg) -> None:
     """Fail closed on a v2 pinned filesystem corpus before any vector write is possible."""
-    if cfg.evidence is None or cfg.evidence.schema_version != 2:
-        return
-    if not any(source.type == "filesystem" for source in cfg.evidence.sources):
+    if not _requires_candidate_evaluation(cfg):
         return
     from tht.evidence import validate_workspace_evidence
 

harness/tht/evidence/formula_store.py

@@ -8,6 +8,7 @@ from __future__ import annotations
 
 import hashlib
 import re
+import unicodedata
 from dataclasses import dataclass
 from pathlib import Path
 from typing import Literal
@@ -121,6 +122,7 @@ def legacy_formula_to_curated(
     formula: ConceptFormula,
     *,
     legacy_path: str,
+    source_content: str | None = None,
 ) -> CuratedEvidence | LegacyFormulaMigrationFailure | None:
     """Convert one reviewed legacy formula into its deterministic curated counterpart.
 
@@ -129,10 +131,39 @@ def legacy_formula_to_curated(
     """
     if formula.status != "reviewed":
         return None
-    source_notes = tuple(formula.sources) or (
-        "Legacy formula migrated without a recorded provenance note.",
-    )
-    source_sha256 = hashlib.sha256(formula.dump().encode("utf-8")).hexdigest()
+    problems: list[str] = []
+    normalized_source: str | None = None
+    if source_content is None:
+        problems.append("original_source_required")
+    else:
+        from tht.evidence.authoring import normalize_source_text
+
+        normalized_source = normalize_source_text(source_content)
+        try:
+            original_formula = ConceptFormula.parse(source_content)
+        except (TypeError, ValidationError, ValueError, yaml.YAMLError):
+            problems.append("original_source_invalid")
+        else:
+            if original_formula != formula:
+                problems.append("original_source_mismatch")
+    source_notes = tuple(formula.sources)
+    if not source_notes:
+        problems.append("supporting_excerpts_required")
+    elif normalized_source is not None and any(
+        unicodedata.normalize("NFC", note.replace("\r\n", "\n").replace("\r", "\n"))
+        not in normalized_source
+        for note in source_notes
+    ):
+        problems.append("supporting_excerpt_unverified")
+    if problems:
+        return LegacyFormulaMigrationFailure(
+            code="legacy_formula_requires_manual_review",
+            legacy_path=legacy_path,
+            formula=formula,
+            problems=tuple(sorted(problems)),
+        )
+    assert normalized_source is not None
+    source_sha256 = hashlib.sha256(normalized_source.encode("utf-8")).hexdigest()
     source_file = legacy_path if legacy_path.startswith("source/") else f"source/{legacy_path}"
     # The legacy path is the immutable identity of this unit during migration. Keeping
     # its full digest avoids a duplicate public ID when the same concept has reviewed