Address review: section-heading fallback, N+1, alias-registry footgun

- resolver.py: backward-heading fallback now matches the nearest *preceding*
  occurrence (rfind from cand.start) instead of the first in the document,
  so a duplicate heading in an earlier exhibit no longer mis-resolves.
- annotation_queries.py: select_related the FK targets CorpusReferenceType
  resolves, removing the N+1 on resolve_corpus_references.
- authorities.py: authority_alias_registry(user=None) now contributes only
  static defaults (Document.objects.none()) rather than aliases from every
  document — closes a cross-tenant footgun for future callers.
- authorities.py: narrow the read_field_file_text except to (OSError,
  ValueError, AttributeError) so genuine bugs surface.
- extractor.py/constants.py: hoist trailing-punctuation literal to
  C.TRAILING_PUNCT.

Author: JSv4

config/graphql/annotation_queries.py

@@ -80,7 +80,15 @@ def resolve_corpus_references(self, info, corpus_id, **kwargs) -> Any:
             qs = qs.filter(reference_type=kwargs["reference_type"])
         if kwargs.get("canonical_key"):
             qs = qs.filter(canonical_key=kwargs["canonical_key"])
-        return qs
+        # Pull the FK targets the type resolves in one pass — without this each
+        # CorpusReferenceType row fires a separate query per FK (N+1).
+        return qs.select_related(
+            "source_annotation",
+            "corpus",
+            "target_document",
+            "target_annotation",
+            "target_corpus",
+        )
 
     governance_graph = graphene.Field(
         GovernanceGraphType,

opencontractserver/enrichment/authorities.py

@@ -67,8 +67,12 @@ def authority_alias_registry(user=None) -> dict[str, str]:
     documents visible to that user are included.
     """
     mapping: dict[str, str] = dict(C.AUTHORITY_PREFIX)
+    # Fail closed: without a user we contribute only the static defaults rather
+    # than aliases from EVERY document (including private corpora). A future
+    # caller that forgets to thread ``creator`` gets no DB-declared aliases, not
+    # a cross-tenant metadata leak.
     qs = (
-        Document.objects.all()
+        Document.objects.none()
         if user is None
         else Document.objects.visible_to_user(user)
     )
@@ -157,7 +161,10 @@ def bootstrap(
             if existing is not None:
                 try:
                     current = read_field_file_text(existing.txt_extract_file)
-                except Exception:  # unreadable -> rewrite below
+                except (OSError, ValueError, AttributeError):
+                    # Unreadable (missing file / no file associated / decode
+                    # error) -> treat as needs-rewrite below. Narrow on purpose
+                    # so genuine bugs surface instead of being swallowed.
                     current = None
                 if current == sec.text:
                     meta = existing.custom_meta or {}

opencontractserver/enrichment/constants.py

@@ -77,6 +77,9 @@
 # Defaults / thresholds.
 DEFAULT_SAMPLE_N = 10
 MAX_DEFINED_TERMS = 50  # cap to control precision/volume in v1
+# Punctuation stripped from the tail of a captured defined term
+# (e.g. (the "Notes," ...) -> "Notes").
+TRAILING_PUNCT = ",.;:"
 ALL_REFERENCE_TYPES = (REF_LAW, REF_DOCUMENT, REF_SECTION, REF_DEFINED_TERM)
 # Defined-terms are opt-in (precision risk); not scanned/applied by default.
 DEFAULT_REFERENCE_TYPES = (REF_LAW, REF_DOCUMENT, REF_SECTION)

opencontractserver/enrichment/extractor.py

@@ -229,7 +229,7 @@ def _terms(self, text: str) -> Iterator[Candidate]:
                 return
             # Trim trailing punctuation captured inside the quotes
             # (e.g. (the "Notes," ...) -> "Notes").
-            term = m.group("term").strip().rstrip(",.;:")
+            term = m.group("term").strip().rstrip(C.TRAILING_PUNCT)
             slug = _slugify_term(term)
             if not slug or slug in seen:
                 continue

opencontractserver/enrichment/resolver.py

@@ -111,7 +111,11 @@ def resolve_section(
         if heading:
             idx = doc_text.lower().find(heading.lower(), cand.end)
             if idx == -1:
-                idx = doc_text.lower().find(heading.lower())
+                # Backward reference (e.g. "as defined in 'Heading' above"):
+                # match the nearest *preceding* occurrence, not the first in
+                # the document — a duplicate heading in an earlier exhibit
+                # would otherwise mis-resolve to an unrelated section.
+                idx = doc_text.lower().rfind(heading.lower(), 0, cand.start)
             if idx != -1:
                 res.target_offset = idx
                 res.resolution_status = C.STATUS_RESOLVED