Testing and quality assurance¶

Repository quality checks¶

uv sync --locked --group dev
uv run python project_tools/check_all.py --scope all
uv run pytest -k cli_pipeline --maxfail=1
uv run pytest -k contract
uv run ruff check .
uv run ruff format --check .
uv run ty check
uv run pytest --cov --cov-report=term-missing
uv run python project_tools/package_coverage.py
uv run python project_tools/update_cli_help.py --check

check_all.py checks the market-intel root repository and its scripts. Three historical submodules have been retired; each owner repository runs its own checks. This repository must not scan adjacent directories to run tests on behalf of those owners.

check_all.py measures every Python production module under src/, including files that tests never import, and enforces a line-coverage floor for each package. Missing source files or a new package without a configured floor fail the gate. These baselines were measured on 2026-10-12 using the complete synthetic suite and locked dependencies. The existing five floors are unchanged; added packages have modest regression headroom. Commentary and publication coverage remain low: their initial floors prevent deterioration and do not establish sufficient behavioral validation.

Production package	Measured baseline	Minimum
`a_share_analysis`	81.14%	70%
`a_share_daily`	76.82%	65%
`daily_messenger`	79.13%	65%
`market_intel_commentary`	15.11%	10%
`market_intel_config`	100.00%	90%
`market_intel_publication`	38.33%	35%
`ops_common`	64.27%	50%
`research_contracts`	79.74%	70%
`style_replica_bridge`	31.34%	25%
`tushare_jobs`	44.51%	35%

Coverage data defaults to .coverage. Set COVERAGE_FILE to store it outside the repository; the collector and package gate read the same file. Synthetic test outputs should also use an external temporary directory. Publication-boundary tests reject outputs inside the owner checkout or another source repository. If another .git marker affects the default temporary root, choose a fresh external test directory with TMPDIR or pytest --basetemp; preserve the marker and safety guard. Pytest clears the selected --basetemp directory, so never point it at existing data, source, or production directories.

Pull requests and pushes to main run check_all.py --scope all --strict-tools, build the Python package, and check the public boundary. GitHub Actions provides the complete quality result; the same gate can be run locally before submission.

Before deleting a merged branch, verify its pull-request state with the cleanup command. It is a dry run by default; pass --yes to delete:

uv run python project_tools/cleanup_merged_branches.py \
  --repo runchengxie/quant-intel-platform \
  --branch fix/example \
  --dry-run
uv run python project_tools/cleanup_merged_branches.py \
  --repo runchengxie/quant-intel-platform \
  --branch fix/example \
  --yes

GitHub Actions does not access live market data, Feishu, or brokerage services. Production data access and deployment checks must follow their separate runbooks. Production data scheduling belongs to Deploy and is outside this quality gate.

Owner-boundary tests¶

Tests in this repository should focus on:

ETL, theme scoring, market news, and fallback behavior
Fail-closed validation of artifact dates, structure, and hashes
Consumption of owner-produced DailyWatch20, D11-H5, and style-factor artifacts
Report rendering, charts, dashboards, and audience isolation
Feishu idempotency, delivery receipts, and failure recovery
Scheduling and recovery logic that must not install or repair research-side timers
Deploy's legacy bounded rollback bridge refresh_daily_watch20.sh restores official artifacts only through the installed daily-watch20-owner CLI. Deploy also owns the tests for the scheduled producer refresh_daily_watch20_quant_research.sh.

Algorithm tests for minute factors, Hermite, rolling training, OOS, ablation, AI ranking, and shadow runs belong to quant-research; market-intel does not duplicate them.

Deployment and scheduler validation¶

Hermes, systemd, and Windows Task Scheduler scripts, templates, and host smoke procedures belong to private quant-intel-deploy. ensure_hermes_gateway.sh, setup_cron.sh, and scripts/windows/test_scheduled_tasks.ps1 are not supplied here and are not this repository's test commands.

This repository validates public CLIs, artifact contracts, and offline fixtures. Host checks, one-off scheduled tasks, and real delivery validation follow Deploy's runbooks and separate authorization. Do not change the host clock or stop production services to test them.

Core test categories¶

Area	Focus
ETL / feeds	RSS and API parsing, retries, simulation, and fallback
scoring / digest	Theme scores, recommendations, rendering, and snapshots
contracts	JSON, artifact, and receipt fields and date constraints
A-share reports	Morning/evening structure, chart paths, stale-chart protection, and missing-data boundaries
DailyWatch20 consumer	Official-artifact validation, raw completeness, and delivery receipts
delivery	`lark-cli`/webhook, audience isolation, and idempotency
recovery	Freshness DAG, retry budgets, windows, and owner boundaries
Dashboard	Payloads, status panels, and HTML output
TuShare compatibility	Report-side export, storage, and window handling

Use the current uv run pytest output for test counts; do not hard-code a count in documentation.

Quality requirements¶

Ruff lint and formatting pass
ty check covers the root project's src/
pytest and contract tests pass
CLI help and docs/cli-reference.md are synchronized
Shell, PowerShell, and JavaScript scripts pass static checks where tools are available
For cross-repository features that depend on owner changes, record the provider pull request and corresponding owner commit before integration

Pull requests created through the GitHub connector cannot replace local uv, systemd, or live-provider validation. Large boundary migrations should remain Draft until the required local quality gates and deployment smoke tests have real results.