veritas-kanban/docs/testing/scheduled-qa-gates.md

Scheduled QA Gates

Review date: 2026-08-24

The scheduled QA workflow runs heavier browser and load-test coverage outside the fast pull-request path. Ordinary pull requests stay limited to lint, typecheck, build, production dependency audit, security gates, and test-scope recording. Workspace tests, coverage, desktop artifacts, and Docker contracts run only at scheduled, manual, or ci:full milestones.

The 2026-06-04 audit found the workflow failing before job creation because job-level env used the runner.temp context. GitHub does not expose the runner context until a job is running, so manual dispatch returned a parse error instead of producing logs. The workflow now writes VERITAS_DATA_DIR from $RUNNER_TEMP during job setup.

Playwright, Mantine QA, coverage, desktop artifacts, load profiles, and Docker contracts are deliberate milestone gates. Apply ci:full to an integration or release candidate, or use the documented scheduled/manual dispatch. Do not add them to every ordinary pull request merely to duplicate the final milestone.

Workflow

Workflow file:

.github/workflows/scheduled-qa.yml

Triggers:

• Pull request carrying ci:full, rerun on each synchronized candidate head.
• Weekly schedule: Monday at 08:17 UTC.
• Manual dispatch: workflow_dispatch.

Manual dispatch input:

Input	Values	Default	Effect
load_profile	smoke, full	smoke	Chooses either k6 CRUD smoke or all profiles

Playwright Gate

The Playwright job runs:

pnpm test:e2e

The job installs Chromium only. The Playwright config still owns the test matrix, including desktop Chromium and the mobile Chromium project for mobile responsive and offline tests.

Scheduled CI sets PLAYWRIGHT_HTML_REPORT=1, so the run emits both GitHub annotations and an HTML report. This keeps ordinary CI output terse while still preserving enough evidence for diagnosis.

Artifacts:

Path	Contents
playwright-report/	HTML report for the scheduled run
test-results/	Failure screenshots, traces, videos, and logs

Retention: 7 days.

k6 Gate

The k6 job starts the built API server, waits for /api/health, then runs the selected profile through the pinned official k6 Docker image:

grafana/k6:1.7.1

Default scheduled profile:

smoke

Manual full profile:

smoke read-load write-load mixed-load ws-stress v5-remote-mix

Artifacts:

Path	Contents
k6-results/*.json	k6 --summary-export output per script
k6-results/*.log	Full stdout/stderr per script
veritas-server.log	API startup/runtime log for the k6 server

Retention: 7 days.

Thresholds

Thresholds live in the k6 scripts and fail the scheduled job when breached. They are intentionally conservative for the generated CI dataset:

Profile	Main threshold
smoke	All checks must pass
read-load	p95 HTTP duration under 200 ms, errors under 1 percent
write-load	p95 HTTP duration under 500 ms, errors under 1 percent
mixed-load	p95 HTTP duration under 500 ms, errors under 1 percent
ws-stress	WebSocket connection errors under 5 percent
v5-remote-mix	Hot-path p95 budgets from 250 ms to 750 ms by endpoint

The scheduled run uses generated test data and should catch obvious regression signals without pretending to be a production soak test. Release candidates should attach a manual load_profile=full run before claiming performance readiness.

Local Commands

Playwright:

pnpm test:e2e

k6 smoke:

pnpm test:load:smoke

k6 full profile:

pnpm test:load

The local k6 commands expect an API server at http://localhost:3001.