From b7b46f0888a4551b7597823ce54ac7872938418d Mon Sep 17 00:00:00 2001
From: Nitjsefnie <zmatek.peter@gmail.com>
Date: Thu, 23 Jul 2026 18:07:00 +0200
Subject: [PATCH] test(diagnosis): pin Unicode content-gap keyword extraction
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Add regression tests for #64: content gap analysis must extract accented
Latin keywords (French "différence", German "Größe") and non-empty keyword
sets for CJK questions, rather than dropping them as the old
`[a-zA-Z]{4,}` regex did.

The production fix (switching to `\w{4,}`) already landed on main via
commit 78c92a4 ("Fix #122"), but there was no test guarding it — these
tests fail against `[a-zA-Z]{4,}` and pass against `\w{4,}`, so a
regression back to ASCII-only extraction would be caught.

Refs #64

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 .../test_diagnosis/test_chunking_unicode.py   | 70 +++++++++++++++++++
 1 file changed, 70 insertions(+)
 create mode 100644 tests/unit/test_diagnosis/test_chunking_unicode.py

diff --git a/tests/unit/test_diagnosis/test_chunking_unicode.py b/tests/unit/test_diagnosis/test_chunking_unicode.py
new file mode 100644
index 0000000..50be868
--- /dev/null
+++ b/tests/unit/test_diagnosis/test_chunking_unicode.py
@@ -0,0 +1,70 @@
+"""Regression tests for #64: content gap analysis must handle non-ASCII text.
+
+``_check_content_gaps`` extracts question keywords with ``re.findall(r"\\b\\w{4,}\\b")``.
+With the previous ``[a-zA-Z]{4,}`` pattern, accented Latin words (French
+"différence", German "Größe") were dropped and CJK questions produced zero
+keywords — so content gap detection was silently skipped for most of the
+world's languages. These tests pin the Unicode-aware behavior.
+"""
+
+from __future__ import annotations
+
+from openagent_eval.diagnosis.chunking import ChunkingQualityAnalyzer
+
+
+class TestContentGapUnicode:
+    """Content gap detection across non-ASCII scripts."""
+
+    def setup_method(self) -> None:
+        """Set up test fixtures."""
+        self.analyzer = ChunkingQualityAnalyzer()
+        # Two unrelated English contexts so any missing keyword drives a gap.
+        self.unrelated = [
+            "The weather is sunny today and warm.",
+            "Cats are small domestic mammals.",
+        ]
+
+    def _content_gaps(self, question: str, contexts: list[str]) -> list:
+        return [
+            i
+            for i in self.analyzer.analyze(question, contexts)
+            if i.issue_type == "content_gap"
+        ]
+
+    def test_accented_german_keyword_is_extracted(self) -> None:
+        """The German word 'Größe' must be extracted, not dropped at the umlaut.
+
+        Under the old ``[a-zA-Z]`` regex 'größe' never entered the keyword set,
+        so it could never appear as a missing keyword.
+        """
+        gaps = self._content_gaps(
+            "Warum ist die Größe der Chunks wichtig?", self.unrelated
+        )
+        assert len(gaps) == 1
+        assert "größe" in gaps[0].description.lower()
+
+    def test_accented_french_keyword_is_extracted(self) -> None:
+        """The French word 'différence' must survive the accented 'é'."""
+        gaps = self._content_gaps(
+            "Quelle est la différence entre les modèles?", self.unrelated
+        )
+        assert len(gaps) == 1
+        assert "différence" in gaps[0].description.lower()
+
+    def test_cjk_only_question_still_analyzed(self) -> None:
+        """A CJK-only question must not short-circuit content gap detection.
+
+        Under the old regex ``re.findall`` returned zero matches, the keyword
+        set was empty, and the method returned early with no analysis at all.
+        """
+        gaps = self._content_gaps("什么是量子计算技术在实践", self.unrelated)
+        assert len(gaps) == 1
+
+    def test_no_false_gap_when_accented_keyword_present(self) -> None:
+        """No gap should be flagged when the accented keyword is in the contexts."""
+        present = [
+            "Die Größe der Chunks ist wichtig für die Qualität.",
+            "Modelle brauchen gute Chunks.",
+        ]
+        gaps = self._content_gaps("Warum ist die Größe wichtig?", present)
+        assert gaps == []