diff --git a/evidence/test-tex-tools-20261004/SHA256SUMS b/evidence/test-tex-tools-20261004/SHA256SUMS
new file mode 100644
index 0000000000..43e4675d15
--- /dev/null
+++ b/evidence/test-tex-tools-20261004/SHA256SUMS
@@ -0,0 +1,4 @@
+6562728634d6d20eec32c3061a1f940c28a7ed6cd1f405e40a9b73f6714c6892  tests/tools/test_tex_tools.py
+4250de984a3e972ea1e62f4de3de7733b3a640b05f137277773949796a59c818  evidence/test-tex-tools-20261004/report.md
+7a24468e97305f4e2f59cb3758fcabec57a3cbdeb8e091e8ed66060bf6294aef  evidence/test-tex-tools-20261004/pytest-tex-tools.txt
+bb2fd9469ceafeb2cf14f725bc8117e2988ca6d278a6a77c7a405d5eb108262d  evidence/test-tex-tools-20261004/pytest-tools-regression.txt
diff --git a/evidence/test-tex-tools-20261004/pytest-tex-tools.txt b/evidence/test-tex-tools-20261004/pytest-tex-tools.txt
new file mode 100644
index 0000000000..614147e170
--- /dev/null
+++ b/evidence/test-tex-tools-20261004/pytest-tex-tools.txt
@@ -0,0 +1,2 @@
+............................................                             [100%]
+44 passed, 5 warnings in 0.47s
diff --git a/evidence/test-tex-tools-20261004/pytest-tools-regression.txt b/evidence/test-tex-tools-20261004/pytest-tools-regression.txt
new file mode 100644
index 0000000000..e7e883f0b1
--- /dev/null
+++ b/evidence/test-tex-tools-20261004/pytest-tools-regression.txt
@@ -0,0 +1,6 @@
+........................................................................ [ 24%]
+........................................................................ [ 48%]
+........................................................................ [ 73%]
+........................................................................ [ 97%]
+......                                                                   [100%]
+294 passed, 5 warnings in 2.67s
diff --git a/evidence/test-tex-tools-20261004/report.md b/evidence/test-tex-tools-20261004/report.md
new file mode 100644
index 0000000000..4aca8b1edb
--- /dev/null
+++ b/evidence/test-tex-tools-20261004/report.md
@@ -0,0 +1,59 @@
+# test: tex_downloader / tex_chunker 工具链补测 — 证据说明
+
+- 日期：2026-10-04
+- 分支：`test/tex-tools-20261004`（基线 `origin/main` @ `ef2d9e5c3`，release v1.6.12）
+- 运行环境：`/Users/Shared/DeepTutor/.venv`（Python 3.13.13，pytest 9.1.1，tiktoken 已装）
+- 新增文件：`tests/tools/test_tex_tools.py`（44 个用例，全部本地 mock，无真实网络）
+- 产品代码改动：无（`git status` 仅新增上述测试文件与本证据目录）
+
+## 测试命令与结果
+
+```
+/Users/Shared/DeepTutor/.venv/bin/python -m pytest tests/tools/test_tex_tools.py -q
+→ 44 passed, 5 warnings in 0.48s
+```
+
+回归确认（未影响既有工具测试）：
+
+```
+/Users/Shared/DeepTutor/.venv/bin/python -m pytest tests/tools/ -q
+→ 294 passed, 5 warnings in 2.86s
+```
+
+## 覆盖清单
+
+tex_downloader（TexDownloader / TexDownloadResult / read_tex_file）：
+- 正常下载分块：tar 源包落盘 `paper_<id>/main.tex`、临时目录清理；zip 源包；非压缩字节按单 tex 文件处理；arxiv_id 从 URL 提取（abs/pdf/带版本号）与显式传入
+- 失败分支（全部 mock）：无法提取 ID；`ConnectionError` / `HTTPError` → "Download failed"；处理异常 → "Processing failed"；压缩包内无 tex → "Main tex file not found"
+- 边界输入：TarSlip 路径穿越成员（`../evil.tex`）被安全过滤且不落盘；损坏字节流；`main.tex` 选择优先级（命名 > documentclass > 最大文件）；大小写混合 `Main.tex` 可命中；`MAIN.TEX` 因 glob 大小写敏感完全不可见（已知限制，见下）
+- `read_tex_file`：UTF-8 内容；非法字节以 errors="ignore" 容错
+
+tex_chunker（TexChunker）：
+- 构造：无 model → 回退 cl100k_base；不支持的 model 名 → 回退 cl100k_base；支持的 model → 正常编码器
+- estimate_tokens：空串为 0；单调性；编码器异常时按 len//4 兜底
+- _clean_text：连续空白折叠（>100 次折叠为 10）；单行 >10000 字符截断加 "...[truncated]" 标记；正常文本不变
+- split_tex_into_chunks：短文本原样透传；空输入返回 `[""]`；多 section 文档按 max_tokens 切分且每块不超限；单 section 超长走段落切分；无 section 文档走段落合并；overlap>0 时相邻块首尾衔接（chunk[1] 开头 40 字符出现在 chunk[0] 尾部）；极小 max_tokens 不崩溃
+- _get_overlap_text：请求 overlap 大于整块 token 时整块返回；正常 overlap 为前块尾部子串
+
+## 发现（只记录，未改产品代码）
+
+1. 失效 docstring 引用（DT-22 §4 确认项）：
+   - `deeptutor/tools/tex_downloader.py:11` — "Based on: TODO.md specification"，仓库根目录无 `TODO.md`（`git ls-tree origin/main --name-only | grep -i todo` 为空）
+   - `deeptutor/tools/tex_chunker.py:11` — 同上
+
+2. tarfile 兼容性风险（测试过程中发现的额外信号）：
+   - `tex_downloader.py` 的 `_extract_tar` 调用 `tar.extractall(extract_dir, members=...)` 未传 `filter=` 参数。Python 3.12+ 触发 DeprecationWarning（本机 3.13.13 可复现）；在 `-W error::DeprecationWarning` 下 5 个 tar 路径用例全部转为 "Processing failed" 失败
+   - pyproject 声明 `requires-python = ">=3.11,<3.15"`，即 3.14 在支持范围内，该调用存在前向兼容风险。建议后续单独开卡处理（如传 `filter="data"`，可同时替代手写 safe_members 过滤）
+
+3. 边界行为基线（测试已固化，供后续回归对照）：
+   - `_find_main_tex` 的发现层 `rglob("*.tex")` 大小写敏感，`MAIN.TEX` 这类全大写扩展名文件永远不会成为候选；命中层 `name.lower()` 的大小写不敏感仅对已匹配 glob 的文件生效
+   - 无句读的超大单段落无法继续切分，会输出超过 max_tokens 的单块（可能伴随一个空白尾块），为当前策略的已知退化行为
+   - 空输入 `split_tex_into_chunks("")` 返回 `[""]` 而非空列表
+
+## 产物文件
+
+- `tests/tools/test_tex_tools.py`
+- `evidence/test-tex-tools-20261004/report.md`（本文件）
+- `evidence/test-tex-tools-20261004/pytest-tex-tools.txt`（本卡用例运行输出）
+- `evidence/test-tex-tools-20261004/pytest-tools-regression.txt`（tests/tools 回归输出）
+- `evidence/test-tex-tools-20261004/SHA256SUMS`
diff --git a/tests/tools/test_tex_tools.py b/tests/tools/test_tex_tools.py
new file mode 100644
index 0000000000..e9f67b886c
--- /dev/null
+++ b/tests/tools/test_tex_tools.py
@@ -0,0 +1,529 @@
+"""Tests for deeptutor.tools.tex_downloader and deeptutor.tools.tex_chunker.
+
+Covers the happy paths (download + extract + locate main tex; section/paragraph
+chunking), boundary inputs (empty, oversized, malformed archives, path-traversal
+members) and the mocked network-failure branches. No real network access.
+"""
+
+from __future__ import annotations
+
+import io
+from pathlib import Path
+import tarfile
+from unittest.mock import MagicMock
+import zipfile
+
+import pytest
+import requests
+
+from deeptutor.tools.tex_chunker import TexChunker
+import deeptutor.tools.tex_downloader as tex_downloader_module
+from deeptutor.tools.tex_downloader import (
+    TexDownloader,
+    TexDownloadResult,
+    read_tex_file,
+)
+
+MIN_TEX = "\\documentclass{article}\n\\begin{document}\nHello arxiv.\n\\end{document}\n"
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+def _make_tar_bytes(files: dict[str, bytes]) -> bytes:
+    buf = io.BytesIO()
+    with tarfile.open(fileobj=buf, mode="w") as tar:
+        for name, data in files.items():
+            info = tarfile.TarInfo(name=name)
+            info.size = len(data)
+            tar.addfile(info, io.BytesIO(data))
+    return buf.getvalue()
+
+
+def _make_zip_bytes(files: dict[str, bytes]) -> bytes:
+    buf = io.BytesIO()
+    with zipfile.ZipFile(buf, "w") as zf:
+        for name, data in files.items():
+            zf.writestr(name, data)
+    return buf.getvalue()
+
+
+def _mock_response(content: bytes) -> MagicMock:
+    response = MagicMock()
+    response.content = content
+    response.raise_for_status.return_value = None
+    return response
+
+
+@pytest.fixture()
+def downloader(tmp_path: Path) -> TexDownloader:
+    return TexDownloader(workspace_dir=str(tmp_path / "ws"))
+
+
+# ---------------------------------------------------------------------------
+# TexDownloader — construction and result object
+# ---------------------------------------------------------------------------
+
+
+class TestTexDownloaderBasics:
+    def test_workspace_dir_created_recursively(self, tmp_path: Path) -> None:
+        nested = tmp_path / "a" / "b" / "ws"
+        TexDownloader(workspace_dir=str(nested))
+        assert nested.is_dir()
+
+    def test_existing_workspace_reused(self, tmp_path: Path) -> None:
+        ws = tmp_path / "ws"
+        ws.mkdir()
+        marker = ws / "keep.txt"
+        marker.write_text("keep", encoding="utf-8")
+        TexDownloader(workspace_dir=str(ws))
+        assert marker.read_text(encoding="utf-8") == "keep"
+
+    def test_result_defaults(self) -> None:
+        result = TexDownloadResult(success=False)
+        assert result.success is False
+        assert result.tex_path is None
+        assert result.tex_content is None
+        assert result.error is None
+
+
+# ---------------------------------------------------------------------------
+# TexDownloader — ArXiv ID extraction
+# ---------------------------------------------------------------------------
+
+
+class TestExtractArxivId:
+    def test_abs_url(self, downloader: TexDownloader) -> None:
+        assert downloader._extract_arxiv_id("https://arxiv.org/abs/1706.03762") == "1706.03762"
+
+    def test_pdf_url_with_version(self, downloader: TexDownloader) -> None:
+        assert downloader._extract_arxiv_id("https://arxiv.org/pdf/2401.12345v2") == "2401.12345"
+
+    def test_unrelated_url_returns_none(self, downloader: TexDownloader) -> None:
+        assert downloader._extract_arxiv_id("https://example.com/paper") is None
+
+    def test_empty_url_returns_none(self, downloader: TexDownloader) -> None:
+        assert downloader._extract_arxiv_id("") is None
+
+
+# ---------------------------------------------------------------------------
+# TexDownloader — download + extract happy paths (network mocked)
+# ---------------------------------------------------------------------------
+
+
+class TestDownloadHappyPaths:
+    def test_tar_source_lands_in_paper_dir(self, downloader: TexDownloader, tmp_path: Path) -> None:
+        payload = _make_tar_bytes({"main.tex": MIN_TEX.encode("utf-8")})
+
+        with pytest.MonkeyPatch.context() as mp:
+            mp.setattr(
+                tex_downloader_module.requests, "get", lambda *a, **k: _mock_response(payload)
+            )
+            result = downloader.download_arxiv_source(
+                "https://arxiv.org/abs/1706.03762", arxiv_id="1706.03762"
+            )
+
+        assert result.success is True, result.error
+        assert result.tex_path is not None
+        assert Path(result.tex_path) == tmp_path / "ws" / "paper_1706.03762" / "main.tex"
+        assert result.tex_content == MIN_TEX
+        # temporary extraction dir is cleaned up, only the permanent paper dir remains
+        assert [p.name for p in (tmp_path / "ws").iterdir()] == ["paper_1706.03762"]
+
+    def test_zip_source_picks_paper_tex(self, downloader: TexDownloader, tmp_path: Path) -> None:
+        payload = _make_zip_bytes({"paper.tex": MIN_TEX.encode("utf-8")})
+
+        with pytest.MonkeyPatch.context() as mp:
+            mp.setattr(
+                tex_downloader_module.requests, "get", lambda *a, **k: _mock_response(payload)
+            )
+            result = downloader.download_arxiv_source(
+                "https://arxiv.org/pdf/2401.12345", arxiv_id="2401.12345"
+            )
+
+        assert result.success is True, result.error
+        assert Path(result.tex_path).name == "main.tex"  # always finalised to main.tex
+        assert result.tex_content == MIN_TEX
+
+    def test_plain_bytes_treated_as_single_tex_file(self, downloader: TexDownloader) -> None:
+        # Not a tar, not a zip → stored as <arxiv_id>.tex and still found.
+        payload = b"some raw tex body without documentclass"
+
+        with pytest.MonkeyPatch.context() as mp:
+            mp.setattr(
+                tex_downloader_module.requests, "get", lambda *a, **k: _mock_response(payload)
+            )
+            result = downloader.download_arxiv_source(
+                "https://arxiv.org/abs/2101.00001", arxiv_id="2101.00001"
+            )
+
+        assert result.success is True, result.error
+        assert result.tex_content == "some raw tex body without documentclass"
+
+    def test_arxiv_id_taken_from_url_when_arg_missing(self, downloader: TexDownloader) -> None:
+        payload = _make_tar_bytes({"main.tex": MIN_TEX.encode("utf-8")})
+
+        with pytest.MonkeyPatch.context() as mp:
+            mp.setattr(
+                tex_downloader_module.requests, "get", lambda *a, **k: _mock_response(payload)
+            )
+            result = downloader.download_arxiv_source("https://arxiv.org/abs/2401.12345v2")
+
+        assert result.success is True, result.error
+        assert "paper_2401.12345" in result.tex_path
+
+
+# ---------------------------------------------------------------------------
+# TexDownloader — failure branches
+# ---------------------------------------------------------------------------
+
+
+class TestDownloadFailureBranches:
+    def test_unextractable_id_fails_fast(self, downloader: TexDownloader) -> None:
+        result = downloader.download_arxiv_source("https://example.com/no-id")
+        assert result.success is False
+        assert result.error == "Unable to extract ArXiv ID"
+
+    def test_network_error_maps_to_download_failed(self, downloader: TexDownloader) -> None:
+        with pytest.MonkeyPatch.context() as mp:
+            mp.setattr(
+                tex_downloader_module.requests,
+                "get",
+                lambda *a, **k: (_ for _ in ()).throw(
+                    requests.exceptions.ConnectionError("conn refused")
+                ),
+            )
+            result = downloader.download_arxiv_source(
+                "https://arxiv.org/abs/1706.03762", arxiv_id="1706.03762"
+            )
+
+        assert result.success is False
+        assert result.error is not None
+        assert result.error.startswith("Download failed")
+        assert "conn refused" in result.error
+
+    def test_http_error_maps_to_download_failed(self, downloader: TexDownloader) -> None:
+        response = MagicMock()
+        response.raise_for_status.side_effect = requests.exceptions.HTTPError("404")
+
+        with pytest.MonkeyPatch.context() as mp:
+            mp.setattr(tex_downloader_module.requests, "get", lambda *a, **k: response)
+            result = downloader.download_arxiv_source(
+                "https://arxiv.org/abs/1706.03762", arxiv_id="1706.03762"
+            )
+
+        assert result.success is False
+        assert result.error is not None
+        assert result.error.startswith("Download failed")
+        assert "404" in result.error
+
+    def test_processing_error_maps_to_processing_failed(self, downloader: TexDownloader) -> None:
+        payload = _make_tar_bytes({"main.tex": MIN_TEX.encode("utf-8")})
+
+        with pytest.MonkeyPatch.context() as mp:
+            mp.setattr(
+                tex_downloader_module.requests, "get", lambda *a, **k: _mock_response(payload)
+            )
+            mp.setattr(
+                downloader,
+                "_find_main_tex",
+                lambda *a, **k: (_ for _ in ()).throw(RuntimeError("boom")),
+            )
+            result = downloader.download_arxiv_source(
+                "https://arxiv.org/abs/1706.03762", arxiv_id="1706.03762"
+            )
+
+        assert result.success is False
+        assert result.error == "Processing failed: boom"
+
+    def test_archive_without_tex_reports_main_not_found(self, downloader: TexDownloader) -> None:
+        payload = _make_tar_bytes({"README.txt": b"no tex here"})
+
+        with pytest.MonkeyPatch.context() as mp:
+            mp.setattr(
+                tex_downloader_module.requests, "get", lambda *a, **k: _mock_response(payload)
+            )
+            result = downloader.download_arxiv_source(
+                "https://arxiv.org/abs/1706.03762", arxiv_id="1706.03762"
+            )
+
+        assert result.success is False
+        assert result.error == "Main tex file not found"
+
+    def test_tar_path_traversal_member_is_skipped(
+        self, downloader: TexDownloader, tmp_path: Path
+    ) -> None:
+        # TarSlip guard: "../evil.tex" must not be written anywhere in the workspace.
+        payload = _make_tar_bytes({"../evil.tex": b"malicious"})
+
+        with pytest.MonkeyPatch.context() as mp:
+            mp.setattr(
+                tex_downloader_module.requests, "get", lambda *a, **k: _mock_response(payload)
+            )
+            result = downloader.download_arxiv_source(
+                "https://arxiv.org/abs/1706.03762", arxiv_id="1706.03762"
+            )
+
+        assert result.success is False
+        assert result.error == "Main tex file not found"
+        assert list(tmp_path.rglob("evil.tex")) == []
+
+
+# ---------------------------------------------------------------------------
+# TexDownloader — main tex selection priority
+# ---------------------------------------------------------------------------
+
+
+class TestFindMainTex:
+    def test_prefers_main_tex_over_documentclass(
+        self, downloader: TexDownloader, tmp_path: Path
+    ) -> None:
+        (tmp_path / " appendix.tex").write_text("\\documentclass{book}", encoding="utf-8")
+        (tmp_path / "main.tex").write_text("main", encoding="utf-8")
+        found = downloader._find_main_tex(tmp_path)
+        assert found is not None and found.name == "main.tex"
+
+    def test_falls_back_to_documentclass_file(
+        self, downloader: TexDownloader, tmp_path: Path
+    ) -> None:
+        (tmp_path / "chapter.tex").write_text("no marker", encoding="utf-8")
+        (tmp_path / "chapter-a.tex").write_text("\\documentclass{article}", encoding="utf-8")
+        found = downloader._find_main_tex(tmp_path)
+        assert found is not None and found.name == "chapter-a.tex"
+
+    def test_falls_back_to_largest_tex(self, downloader: TexDownloader, tmp_path: Path) -> None:
+        (tmp_path / "a.tex").write_text("x", encoding="utf-8")
+        (tmp_path / "b.tex").write_text("x" * 500, encoding="utf-8")
+        found = downloader._find_main_tex(tmp_path)
+        assert found is not None and found.name == "b.tex"
+
+    def test_no_tex_files_returns_none(self, downloader: TexDownloader, tmp_path: Path) -> None:
+        (tmp_path / "notes.txt").write_text("nothing", encoding="utf-8")
+        assert downloader._find_main_tex(tmp_path) is None
+
+    def test_mixed_case_main_tex_matched(self, downloader: TexDownloader, tmp_path: Path) -> None:
+        # Name comparison is case-insensitive for files that match the *.tex glob.
+        (tmp_path / "Main.tex").write_text("mixed", encoding="utf-8")
+        found = downloader._find_main_tex(tmp_path)
+        assert found is not None and found.name == "Main.tex"
+
+    def test_uppercase_extension_file_is_invisible(
+        self, downloader: TexDownloader, tmp_path: Path
+    ) -> None:
+        # Known limitation: discovery globs "*.tex" case-sensitively, so a file
+        # named MAIN.TEX is never even considered a candidate.
+        (tmp_path / "MAIN.TEX").write_text("upper", encoding="utf-8")
+        assert downloader._find_main_tex(tmp_path) is None
+
+
+# ---------------------------------------------------------------------------
+# read_tex_file convenience function
+# ---------------------------------------------------------------------------
+
+
+class TestReadTexFile:
+    def test_reads_utf8_content(self, tmp_path: Path) -> None:
+        tex = tmp_path / "paper.tex"
+        tex.write_text("\\section{引言}\n内容", encoding="utf-8")
+        assert read_tex_file(str(tex)) == "\\section{引言}\n内容"
+
+    def test_invalid_utf8_bytes_are_ignored_not_raised(self, tmp_path: Path) -> None:
+        tex = tmp_path / "broken.tex"
+        tex.write_bytes(b"ok \\documentclass \xff\xfe end")
+        content = read_tex_file(str(tex))
+        assert content.startswith("ok \\documentclass")
+
+
+# ---------------------------------------------------------------------------
+# TexChunker — construction
+# ---------------------------------------------------------------------------
+
+
+def _cl100k_chunker(monkeypatch: pytest.MonkeyPatch) -> TexChunker:
+    """Chunker forced onto the cl100k_base fallback, independent of local LLM config."""
+    import deeptutor.tools.tex_chunker as chunker_module
+
+    def _raise() -> None:
+        raise RuntimeError("no llm config in tests")
+
+    monkeypatch.setattr(chunker_module, "resolve_llm_runtime_config", _raise)
+    return TexChunker()
+
+
+class TestTexChunkerInit:
+    def test_no_model_resolves_fallback_encoding(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+        assert chunker.encoder.name == "cl100k_base"
+
+    def test_unsupported_model_falls_back_to_cl100k(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        monkeypatch.setattr(
+            "deeptutor.tools.tex_chunker.resolve_llm_runtime_config",
+            lambda: MagicMock(model="definitely-not-a-real-model"),
+        )
+        chunker = TexChunker(model="definitely-not-a-real-model")
+        assert chunker.encoder.name == "cl100k_base"
+
+    def test_supported_model_uses_its_encoding(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        monkeypatch.setattr(
+            "deeptutor.tools.tex_chunker.resolve_llm_runtime_config",
+            lambda: MagicMock(model="gpt-4o"),
+        )
+        chunker = TexChunker(model="gpt-4o")
+        assert chunker.encoder.name in {"o200k_base", "cl100k_base"}
+
+
+# ---------------------------------------------------------------------------
+# TexChunker — token estimation and text cleaning
+# ---------------------------------------------------------------------------
+
+
+class TestEstimateTokens:
+    def test_empty_string_is_zero_tokens(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        assert _cl100k_chunker(monkeypatch).estimate_tokens("") == 0
+
+    def test_estimation_is_positive_and_monotonic(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+        short = chunker.estimate_tokens("one two three")
+        longer = chunker.estimate_tokens("one two three four five six seven eight")
+        assert short > 0
+        assert longer > short
+
+    def test_encoder_failure_falls_back_to_char_estimate(
+        self, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+
+        def _boom(text: str) -> list[int]:
+            raise RuntimeError("encode failed")
+
+        monkeypatch.setattr(chunker.encoder, "encode", _boom)
+        text = "x" * 40
+        assert chunker.estimate_tokens(text) == len(text) // 4
+
+
+class TestCleanText:
+    def test_collapses_repeated_whitespace(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+        cleaned = chunker._clean_text("a" + "\n" * 200 + "b")
+        assert "\n" * 10 in cleaned
+        assert "\n" * 11 not in cleaned
+
+    def test_truncates_overlong_single_line(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+        cleaned = chunker._clean_text("x" * 15000)
+        assert cleaned.endswith("...[truncated]")
+        assert len(cleaned) == 10000 + len("...[truncated]")
+
+    def test_normal_lines_pass_through(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+        text = "line one\nline two\nline three"
+        assert chunker._clean_text(text) == text
+
+
+# ---------------------------------------------------------------------------
+# TexChunker — chunking
+# ---------------------------------------------------------------------------
+
+SECTION_DOC = (
+    "\\documentclass{article}\n\\begin{document}\n"
+    + "\n\n".join(
+        f"\\section{{Section {i}}}\nParagraph {i} explains topic {i} with words.\n"
+        f"Second paragraph of section {i} adds more detail words here."
+        for i in range(12)
+    )
+    + "\n\\end{document}\n"
+)
+
+
+class TestSplitTexIntoChunks:
+    def test_short_content_passes_through_verbatim(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+        content = "\\section{Intro}\nShort body."
+        assert chunker.split_tex_into_chunks(content, max_tokens=10_000) == [content]
+
+    def test_empty_content_returns_single_empty_chunk(
+        self, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+        result = chunker.split_tex_into_chunks("", max_tokens=100)
+        assert result == [""]
+
+    def test_sectioned_document_splits_into_bounded_chunks(
+        self, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+        max_tokens = 120
+        chunks = chunker.split_tex_into_chunks(SECTION_DOC, max_tokens=max_tokens, overlap=0)
+        assert len(chunks) > 1
+        for chunk in chunks:
+            assert chunker.estimate_tokens(chunk) <= max_tokens
+        joined = "\n".join(chunks)
+        for i in range(12):
+            assert f"\\section{{Section {i}}}" in joined
+
+    def test_single_oversized_section_splits_by_paragraphs(
+        self, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+        body = "\n\n".join(
+            f"Paragraph {i} contains several words of body text for testing." for i in range(40)
+        )
+        doc = f"\\section{{Huge}}\n{body}"
+        max_tokens = 60
+        chunks = chunker.split_tex_into_chunks(doc, max_tokens=max_tokens, overlap=0)
+        assert len(chunks) > 1
+        for chunk in chunks:
+            assert chunker.estimate_tokens(chunk) <= max_tokens
+
+    def test_document_without_sections_still_chunks(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+        doc = "\n\n".join(
+            f"Plain paragraph number {i} with a handful of ordinary words." for i in range(40)
+        )
+        chunks = chunker.split_tex_into_chunks(doc, max_tokens=80, overlap=0)
+        assert len(chunks) > 1
+        for chunk in chunks:
+            assert chunker.estimate_tokens(chunk) <= 80
+
+    def test_overlap_connects_adjacent_chunks(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+        chunks = chunker.split_tex_into_chunks(SECTION_DOC, max_tokens=120, overlap=50)
+        assert len(chunks) > 1
+        # The second chunk must open with text taken from the tail of the first.
+        assert chunks[1][:40] in chunks[0]
+
+    def test_degenerate_single_sentence_cannot_be_split(
+        self, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        # One huge paragraph without any sentence-ending punctuation cannot be
+        # split below max_tokens by the current strategy: the blob is emitted
+        # oversized (known limitation), possibly next to a blank trailing chunk.
+        chunker = _cl100k_chunker(monkeypatch)
+        doc = "a" * 6000
+        chunks = chunker.split_tex_into_chunks(doc, max_tokens=50, overlap=0)
+        assert chunks
+        assert chunker.estimate_tokens(chunks[0]) > 50
+        assert "".join(chunks).strip() == doc
+
+    def test_tiny_max_tokens_never_hangs_or_crashes(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+        chunks = chunker.split_tex_into_chunks(SECTION_DOC, max_tokens=1, overlap=0)
+        assert chunks  # produces something instead of failing
+
+
+class TestGetOverlapText:
+    def test_shorter_chunk_returned_whole(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+        assert chunker._get_overlap_text("tiny", 100) == "tiny"
+
+    def test_tail_overlap_is_suffix_of_previous_chunk(
+        self, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        chunker = _cl100k_chunker(monkeypatch)
+        text = " ".join(f"word{i}" for i in range(200))
+        overlap = chunker._get_overlap_text(text, 10)
+        assert overlap
+        assert overlap in text
+        assert text.endswith(overlap)