fix(evidence): resolve final review findings

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