[tae898]

Part of #19 (remaining tests, item 3 of the latest status comment, plus the formatting failure on docs/antora.yml). Ported from humemai/arcadedb-embedded-python (the test files and humemai/arcadedb-embedded-python#445, #388, #347, #383 where they were first added), with import arcadedb_embedded as arcadedb changed to import arcadedb.

• tests/test_sql_constraints_and_maintenance.py: the SQL forms the docs give for constraints, type scans, isolation level and admin statements.
• tests/test_collection_equality_known_issues.py: SQL = on a list or map written from Python (SQL = on a list or map compares elements with Java equals, so [1L, 2L] is not equal to [1, 2] (CONTAINSALL, GROUP BY and Cypher say it is) arcadedb#10448). The fixed probe handles both engines. The docstring now says the wheel comes from the arcadedata/arcadedb image, not from an engine source tree.
• tests/test_guide_graph_analytics.py and tests/test_guide_timeseries.py: every guide example, run and asserted. In humemai each statement also had to appear verbatim in docs/guide/*.md. The guide pages are not in this repository yet (they are tracker item 4, as AsciiDoc), so doc() returns the statement without checking a page, and the module docstring says so; the check comes back with the Antora pages.
• tests/test_equivalence.py and tests/test_equivalence_ladybug.py: ArcadeDB against DuckDB (SQL), NetworkX (graph algorithms), LadybugDB (openCypher) and NumPy (vector search). The ladybug half is left out of collection on Windows in tests/conftest.py, as in humemai.
• Test extra: duckdb>=1.4.2, ladybug>=0.21.2, networkx>=3.4 (the floors humemai uses), and the same three in the CI test install line of test-python-bindings.yml. All three have wheels for the four CI platforms on Python 3.10 to 3.14 (checked on PyPI), and the dependency-floor audit's uv pip compile --resolution lowest-direct resolves for every classifier. There is no Python 3.15 leg on main, so the cp315 exclusion humemai has is not needed.
• docs/antora.yml: the "Pretty format YAML" pre-commit hook fails on it on main. The hook's own rewrite is committed (version: ~ becomes version:, the nav list indent changes). Formatting only: both files parse to the same data.

Not ported from the humemai-only list: test_check_version.py (PR #43), test_wheel_no_studio.py (goes with #20), test_download_data_dbpedia.py (the DBpedia part of examples/download_data.py is not in this repository), and the benchmark tests listed as not planned in the tracker (test_bench_exec.py, test_render_ab_tables.py, test_jvm_payload.py).

Checked locally: pre-commit on the changed files (black, isort, pretty-format-yaml, actionlint, the name check), Bandit as CI runs it on src and tests (clean), and the six new files plus test_check_test_skips.py and test_no_embedded_name.py pinned to cores 0-3: 85 passed, none skipped. The wheel for that run was not a fresh build of this branch: it was the Python sources of this branch on top of the jars and JRE of an earlier 26.11.1-SNAPSHOT build, since a Docker wheel build is not needed for tests that only use SQL, openCypher and the reference engines. CI runs them on freshly built wheels on all platforms.

🤖 Generated with Claude Code