[pelikhan]

Company-scale workload planning needs a fast way to exercise real ingestion without waiting for thousands of agent executions. This adds configurable synthetic company scenarios and a side-effect-free dashboard calculator for PR-driven AW volume.

Approach

• Stream completed-AW-shaped evidence directly into the existing hash-verified ingestion pipeline, then measure PostgreSQL retention and uncached HTTP queries. Agent execution, completion callbacks, collection subprocesses, webhooks, and GitHub quotas are deliberately outside this benchmark.
• Provide small, medium, and large presets, JSON scenarios, and CLI overrides for repository/developer counts, PR rate, AW fan-out, duration, AIC per run, seed, and timestamp distribution. The large preset runs daily on main at 06:37 UTC through the existing full-validation workflow, with its outcome included in the aggregate daily report. Manual configuration remains available; the small smoke case joins existing service-backed CI.
• Add Simulators > Pull request using Dashboard Language queries and existing controls, with worker/server parity coverage and experimental-page gating. Repository count distributes activity rather than multiplying PR volume.

Requested scenario

The large preset models 10,000 repositories, 10,000 developers, 10 PRs/developer/day, five AW runs/PR, and 100 AIC/run: 100,000 PRs, 500,000 AW records, and 50 million modeled AIC per day.

The full 500,000-record run completed successfully. Production retention remains unchanged: 200,000 runs are retained, 300,000 are evicted, and query provenance correctly reports partial coverage. Reports distinguish generated volume, retained evidence, modeled credits, and measured timings; no capacity or billing guarantee is inferred. Disposable enterprise PostgreSQL instances need additional lock-table headroom, configured in the reusable stress workflow.

POSTGRES_URL='postgres://postgres@127.0.0.1:5432/cao_test?sslmode=disable' \
  npm run test:stress:company -- --preset large

Validation

• Full-size real PostgreSQL run, multi-day burst and zero-volume cases, replay/cancellation checks, and focused race-enabled Go tests.
• 403 dashboard unit suites passed: 3,887 tests passed, one skipped.
• Dashboard browser regressions covered all 309 cases; four contention-related timeouts passed on a bounded-worker rerun. Four additional populated full-site desktop/mobile checks passed and captured screenshots.
• Root/dashboard lint, typechecks, Go vet/format checks, dashboard/corpus validation, focused workflow/launcher/build contracts, workflow compilation, and documentation build passed.
• Daily scheduling, large-preset selection, main-revision gating, reusable-workflow permissions, and aggregate report coverage are covered by 24 focused passing contract tests.
• Lighthouse completed all scenarios but the CFO full-data desktop CLS budget failed at 0.118 versus 0.1 for cold and PWA runs. An isolated build of unchanged HEAD reproduces both failures identically; this PR does not change that baseline issue.

Desktop and mobile screenshots are attached in a PR comment.

[pelikhan]

Populated full-site Pull request calculator at desktop (1440px) and mobile (390px) widths, using the default 10,000-developer workload. These are screenshots of the rendered dashboard, not mocked page content.

[github-actions]

GitHub Actions lint failures

Failed scanners: actionlint

Workflow run: https://github.com/githubnext/gh-cao/actions/runs/41065495475

Generated .lock.yml files were excluded.

[github-actions]

Dashboard dead views

Dead element views (0):
Dead dashboard pages (0):
Analyzed 8 element renderers and 70 dashboard pages across 1 dashboard.json file(s) from dashboard/site/src/components/ui-elements.js.

⚠️ This reachability report is informational and does not block the pull request.
🔎 View the workflow run and report artifact.

[github-actions]

GitHub Actions lint failures

Failed scanners: actionlint

Workflow run: https://github.com/githubnext/gh-cao/actions/runs/41065802253

Generated .lock.yml files were excluded.

[pelikhan]

Full-site calculator regression evidence after separating company repository parameters from the database simulator. Both desktop (1440px) and mobile (390px) Playwright scenarios passed against the populated synthetic Pages deployment.