[intern-31] Integration test: convert → chunk → query

[Supermonters]

Closes #369

cc @anhnth24 — em xin gửi PR intern-31 để anh review ạ.

Summary

• Thêm integration test happy path trong crates/core/tests/integration_convert_chunk.rs: file DOCX thật → Converter::convert_path → chunk_markdown → search_corpus.
• Fixture synthetic crates/core/tests/fixtures/heading-hierarchy.docx (Heading 1 Phần I + intro + 3× Heading 2 Mục 1/2/3, mã REF-M1/M2/M3). Không dùng tài liệu khách hàng.
• Không mock converter, không path tuyệt đối (CARGO_MANIFEST_DIR + tests/fixtures/). Không đổi logic production.

Thay đổi file

crates/core/tests/fixtures/heading-hierarchy.docx

DOCX tự soạn, style Word Heading 1/2 (không chỉ Bold). Convert ra:

# Phần I

Giới thiệu phần I.

## Mục 1

Nội dung mục 1. Mã tham chiếu REF-M1.

## Mục 2

Nội dung mục 2. Mã tham chiếu REF-M2.

## Mục 3

Nội dung mục 3. Mã tham chiếu REF-M3.

Golden DOCX sẵn có (gold-006.docx, …) chỉ 1 H1 + 1 H2 → sau chunk thường 1 mảnh, không đủ AC “heading 1 → 2–3 heading 2”.

Tên file mô tả nội dung (heading-hierarchy), không gắn intern31. Trace issue để ở title PR / Closes #369.

crates/core/tests/integration_convert_chunk.rs

Một test: integration_convert_chunk_then_query_happy_path.

Luồng:

1. Converter::new().convert_path — convert thật, format == Docx, markdown có # Phần I.
2. chunk_markdown(md, 2000) — chia theo heading, giữ path "Phần I > Mục 2".
3. build_corpus + search_corpus("REF-M2", 5) — keyword search có sẵn (không viết TF-IDF riêng).

Test nào pass? Assertion nào validate?

Loại (AC issue)	Assert	Ý nghĩa
Count	chunks.len() == 4, corpus.len() khớp	H1 intro + 3 H2
Title	chunks[i].heading = Phần I, Phần I > Mục 1/2/3	hierarchy đúng
Order	chunks[i].index == i	Mục 1 → 2 → 3, không đảo
Retrieval	search_corpus("REF-M2") không rỗng, hits[0] chứa Mục 2	query ra đúng chunk

Vì sao đây là integration test (không nhét vào chunk.rs)?

• Unit test chunk.rs (intern-05) đã khóa heading path trên chuỗi Markdown hardcode.
• intern-31 khóa cả pipeline: Word style Heading → converter DOCX → chunk → search. Mock convert_path sẽ không chứng minh file thật convert đúng.

Test plan

• cargo test -p fileconv-core integration_convert_chunk -- --nocapture
• cargo fmt --all -- --check
• cargo metadata --locked --format-version 1 --no-deps
• python3 scripts/check-dependency-policy.py

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

Bugbot Autofix prepared a fix for the issue found in the latest run.

• ✅ Fixed: Unmanaged test fixture
  • Added crates/core/tests/fixtures/manifest.json for heading-hierarchy.docx and wired that root into make check-fixtures so checksum, unmanaged-file, and secret-canary checks cover it.

Or push these changes by commenting:

@cursor push fe6df63fc6

Preview (fe6df63fc6)

diff --git a/Makefile b/Makefile
--- a/Makefile
+++ b/Makefile
@@ -49,6 +49,7 @@
 	python3 scripts/check-fixtures.py
 	python3 scripts/check-fixtures.py --root app/src-tauri/fixtures/knowledge/v1
 	python3 scripts/check-fixtures.py --root crates/knowledge/fixtures
+	python3 scripts/check-fixtures.py --root crates/core/tests/fixtures
 	python3 scripts/check-fixtures.py --self-test
 
 check-markhand-gates:

diff --git a/crates/core/tests/fixtures/manifest.json b/crates/core/tests/fixtures/manifest.json
new file mode 100644
--- /dev/null
+++ b/crates/core/tests/fixtures/manifest.json
@@ -1,0 +1,15 @@
+{
+  "version": 1,
+  "fixtures": [
+    {
+      "id": "core.docx.heading-hierarchy.v1",
+      "path": "heading-hierarchy.docx",
+      "sha256": "806a293e688a70880f3a62a727fc84085d76cf8862c861dcf259a1819c3aaf25",
+      "kind": "docx",
+      "owner": "markhand-core",
+      "source": "synthetic",
+      "license": "CC0-1.0",
+      "sensitive": false
+    }
+  ]
+}

_{You can send follow-ups to the cloud agent here.}

^{Reviewed by Cursor Bugbot for commit f7771c1. Configure here.}

[cursor]

Unmanaged test fixture

Medium Severity

The new heading-hierarchy.docx under crates/core/tests/fixtures/ has no local manifest.json (id, sha256, license, sensitive: false) and is not wired into make check-fixtures. That skips the repo fixture contract and secret/canary checks used by other component fixture roots such as crates/knowledge/fixtures.

 

^{Reviewed by Cursor Bugbot for commit f7771c1. Configure here.}