From f1e998f010fd5e8b01a2ceeb645a0c3b36c7394b Mon Sep 17 00:00:00 2001
From: HeaTTap <aaronlmnop@gmail.com>
Date: Thu, 10 Sep 2026 08:04:11 +0000
Subject: [PATCH] test(brain_ask): cover looks_like_entity and _links_in (#4)

Add unit tests in tests/test_brain_ask.py covering looks_like_entity and _links_in without requiring models or GPU.

Fix _links_in to ignore fenced code blocks so code snippets are not treated as graph edges.

All 49 tests pass.
---
 brain_ask.py            |   4 +-
 tests/test_brain_ask.py | 111 ++++++++++++++++++++++++++++++++++++++++
 2 files changed, 114 insertions(+), 1 deletion(-)
 create mode 100644 tests/test_brain_ask.py

diff --git a/brain_ask.py b/brain_ask.py
index 83f44b0..e346bcb 100644
--- a/brain_ask.py
+++ b/brain_ask.py
@@ -75,12 +75,14 @@ def looks_like_entity(q):
     return False
 
 
+FENCED_CODE_RX = re.compile(r'(?:```[\s\S]*?```|~~~[\s\S]*?~~~)')
 WIKILINK_RX = re.compile(r'\[\[([^\]\|#]+)')
 
 def _links_in(path):
-    """Outgoing wikilink TARGETS in a note (alias/heading stripped)."""
+    """Outgoing wikilink TARGETS in a note (alias/heading stripped, code fences ignored)."""
     try: t = Path(path).read_text(encoding='utf-8', errors='ignore')
     except Exception: return []
+    t = FENCED_CODE_RX.sub('', t)
     return [m.strip() for m in WIKILINK_RX.findall(t)]
 
 
diff --git a/tests/test_brain_ask.py b/tests/test_brain_ask.py
new file mode 100644
index 0000000..c8b6fec
--- /dev/null
+++ b/tests/test_brain_ask.py
@@ -0,0 +1,111 @@
+# -*- coding: utf-8 -*-
+"""Unit tests for pure functions in brain_ask.py.
+
+Covers looks_like_entity() and _links_in() without model downloads or GPUs.
+Addresses issue #4:
+- looks_like_entity(): verifies entity gate switches off graph expansion for
+  proper nouns, CamelCase, snake_case, known tools, and non-Latin (Cyrillic) names,
+  while keeping graph on for themes and acronyms.
+- _links_in(): table-driven parsing of wikilinks in markdown files using tmp_path,
+  including aliases, anchors, multiple links per line, fenced code blocks, and no links.
+"""
+import sys
+from pathlib import Path
+import pytest
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+from brain_ask import looks_like_entity, _links_in  # noqa: E402
+
+
+@pytest.mark.parametrize(
+    ("query", "expected"),
+    [
+        # Known tools and products
+        ("obsidian", True),
+        ("sqlite", True),
+        ("whatsapp", True),
+        # Snake_case identifiers
+        ("turnstate_hook", True),
+        ("my_project", True),
+        # CamelCase tools and libraries
+        ("NotebookLM", True),
+        ("PyTorch", True),
+        ("FastAPI", True),
+        # Proper nouns and names (single word or two tokens)
+        ("Feynman", True),
+        ("Turing", True),
+        ("Alan Turing", True),
+        ("Richard Feynman", True),
+        # Proper noun in short query (idx > 0)
+        ("notes on Feynman", True),
+        ("who is Turing", True),
+        # Non-Latin script (Cyrillic entities and names)
+        ("Тьюринг", True),
+        ("Алан Тьюринг", True),
+        ("«Тьюринг»", True),
+        ("заметки про Тьюринга", True),
+        # Empty / whitespace / None
+        ("", False),
+        ("   ", False),
+        (None, False),
+        # Query with > 5 tokens
+        ("this is a very long query with too many words", False),
+        # Lowercase common-noun phrases / themes (graph stays on)
+        ("graph memory", False),
+        ("approaches to agent memory", False),
+        ("second brain", False),
+        # Non-Latin script lowercase themes
+        ("векторный поиск", False),
+        ("теория графов", False),
+        # Bare topic acronyms (uppercase <= 3 chars)
+        ("AI", False),
+        ("RAG", False),
+        ("DAO", False),
+        ("РФ", False),
+        # Multi-word sentence query where only first word is capitalized
+        ("How does graph memory work", False),
+    ],
+)
+def test_looks_like_entity(query, expected):
+    assert looks_like_entity(query) is expected
+
+
+@pytest.mark.parametrize(
+    ("content", "expected"),
+    [
+        # A file with no links at all
+        ("A plain note with no wikilinks inside.", []),
+        # A link with an alias
+        ("See [[note|display text]] for details.", ["note"]),
+        # A link with a heading anchor
+        ("Refer to [[note#section]] above.", ["note"]),
+        # A link with both heading anchor and alias
+        ("Check [[note#section|display text]] here.", ["note"]),
+        # A line with two links on it
+        ("Two links on one line: [[first]] and [[second]].", ["first", "second"]),
+        # Multiple links across lines
+        ("Line one [[alpha]].\nLine two [[beta]].", ["alpha", "beta"]),
+        # A wikilink inside a fenced code block (backticks) should NOT count
+        ("```python\nx = [[code_note]]\n```", []),
+        # A wikilink inside a fenced code block (tildes) should NOT count
+        ("~~~markdown\n[[ignored_note]]\n~~~", []),
+        # Mixed note: links outside code blocks count, links inside code blocks do not
+        (
+            "Link outside [[valid_note]].\n"
+            "```\n"
+            "[[inside_code]]\n"
+            "```\n"
+            "Another outside [[another_valid#anchor|alias]].",
+            ["valid_note", "another_valid"],
+        ),
+    ],
+)
+def test_links_in_content_cases(tmp_path, content, expected):
+    note_file = tmp_path / "note.md"
+    note_file.write_text(content, encoding="utf-8")
+    assert _links_in(note_file) == expected
+
+
+def test_links_in_nonexistent_file(tmp_path):
+    missing_file = tmp_path / "does_not_exist.md"
+    assert _links_in(missing_file) == []