feat(evidence): structure v3 domain rules for review

harness/tht/evidence/canonical.py

@@ -319,6 +319,10 @@ _V2_EMPTY_LIST = "<!-- tht:empty-list -->"
 _V2_REVIEW_SEPARATOR = "<!-- tht:review-separator -->"
 _V2_REVIEW_FIELD = "<!-- tht:review-field -->"
 _V3_METADATA = re.compile(r"\A<!-- tht:metadata:([A-Za-z0-9+/=]+) -->\n")
+_V3_RAW_RULE = re.compile(
+    r"\A<!-- tht:raw-rule:([A-Za-z0-9+/=]+) -->\n\n(.+)\Z",
+    re.DOTALL,
+)
 _V3_KIND_LABELS = {
     "en": {
         "glossary": "Glossary",
@@ -507,6 +511,72 @@ def _parse_v3_values(value: str, labels: dict[str, str]) -> dict[str, str]:
     return {key: "\n".join(lines) for key, lines in parsed.items()}
 
 
+def _leading_rule_label(value: str) -> tuple[str, str] | None:
+    match = re.match(r"\A([^:;\n]{2,80}:)\s+(.+)\Z", value, re.DOTALL)
+    if match is None:
+        return None
+    label = match.group(1)
+    if len(label.removesuffix(":").split()) > 10:
+        return None
+    if label.lower() in {"http:", "https:"}:
+        return None
+    return label, match.group(2)
+
+
+def _render_v3_rule_detail(value: str) -> str:
+    clauses = re.split(r"(?<=;)\s+", value)
+    if len(clauses) < 3:
+        return value
+    rendered: list[str] = []
+    for clause in clauses:
+        labelled = _leading_rule_label(clause)
+        if labelled is None:
+            rendered.append(f"- {clause}")
+        else:
+            label, detail = labelled
+            rendered.append(f"- **{label}** {detail}")
+    return "\n".join(rendered)
+
+
+def _render_v3_rule_presentation(value: str) -> str:
+    if re.search(r"(?m)^\s*(?:[-*+] |#{1,6} |```|>)", value):
+        return value
+    rendered: list[str] = []
+    for paragraph in re.split(r"\n\s*\n", value):
+        sentences = re.split(
+            r"(?<=[.!?])\s+(?=[A-ZÀ-ÖØ-Þ0-9(`])",
+            paragraph,
+        )
+        for sentence in sentences:
+            labelled = _leading_rule_label(sentence)
+            if labelled is None:
+                rendered.append(_render_v3_rule_detail(sentence))
+                continue
+            label, detail = labelled
+            rendered.append(f"### {label}\n\n{_render_v3_rule_detail(detail)}")
+    return "\n\n".join(rendered)
+
+
+def _render_v3_rule(value: str) -> str:
+    if "<!-- tht:raw-rule:" in value:
+        raise ValueError("curated evidence rule contains a reserved marker")
+    encoded = base64.b64encode(value.encode("utf-8")).decode("ascii")
+    return (
+        f"<!-- tht:raw-rule:{encoded} -->\n\n"
+        f"{_render_v3_rule_presentation(value)}"
+    )
+
+
+def _parse_v3_rule(value: str) -> str:
+    match = _V3_RAW_RULE.fullmatch(value)
+    if match is None:
+        return value
+    try:
+        return base64.b64decode(match.group(1), validate=True).decode("utf-8")
+    except (binascii.Error, UnicodeDecodeError, ValueError) as error:
+        raise ValueError("curated evidence v3 rule metadata is malformed") from error
+
+
 def _render_v2_payload(value: CuratedEvidence, labels: dict[str, str]) -> list[str]:
     payload = value.payload
     if value.kind == "glossary":
@@ -568,6 +638,12 @@ def _render_v2_payload(value: CuratedEvidence, labels: dict[str, str]) -> list[s
 
 
 def _render_v3_payload(value: CuratedEvidence, labels: dict[str, str]) -> list[str]:
+    if value.kind == "domain":
+        return [_render_v2_field(
+            "rule",
+            labels["rule"],
+            _render_v3_rule(value.payload.rule),
+        )]
     if value.kind != "enum":
         return _render_v2_payload(value, labels)
     payload = value.payload
@@ -705,14 +781,22 @@ def _render_v3_metadata(value: CuratedEvidence) -> str:
     return f"<!-- tht:metadata:{encoded} -->"
 
 
-def _render_v3_body(value: CuratedEvidence) -> str:
+def _render_v3_body(value: CuratedEvidence, *, structured_domain_rule: bool = True) -> str:
     if "\n" in value.title:
         raise ValueError("curated evidence title must be single-line in v3")
     labels = _v2_labels(value.language)
     fields = [
         _render_v3_overview(value, labels),
         _render_v3_scope(value, labels),
-        *_render_v3_payload(value, labels),
+        *(
+            _render_v3_payload(value, labels)
+            if structured_domain_rule
+            else (
+                [_render_v2_field("rule", labels["rule"], value.payload.rule)]
+                if value.kind == "domain"
+                else _render_v3_payload(value, labels)
+            )
+        ),
         _render_v2_field(
             "supporting_excerpts",
             labels["supporting_excerpts"],
@@ -827,6 +911,8 @@ def _parse_v2_payload(
 def _parse_v3_payload(
     kind: str, fields: dict[str, str], labels: dict[str, str],
 ) -> tuple[dict, set[str]]:
+    if kind == "domain":
+        return {"rule": _parse_v3_rule(fields.get("rule", ""))}, {"rule"}
     if kind != "enum":
         return _parse_v2_payload(kind, fields, labels)
     return {
@@ -985,7 +1071,9 @@ def parse_curated_markdown(text: str, *, path: Path | None = None) -> CuratedEvi
                 data["payload"] = data.pop(kind, None)
     evidence = CuratedEvidence.model_validate(data)
     if evidence.schema_version == 3 and dump_curated_markdown(evidence) != text:
-        raise ValueError("curated evidence v3 presentation is not canonical")
+        legacy = f"{_render_v3_metadata(evidence)}\n{_render_v3_body(evidence, structured_domain_rule=False)}"
+        if legacy != text:
+            raise ValueError("curated evidence v3 presentation is not canonical")
     if path is not None:
         _validate_kind_directory(path, evidence.kind)
     return evidence