Simplify source setup and focus the dark workspace on AEO

[erickkkyt]

Outcome

New local BeatGEO users can install and start the product without installing Docker or PostgreSQL. A pinned native PostgreSQL package manages a dedicated loopback database and preserves data across restarts. Existing Docker/manual installations keep their configuration and data.

The monitoring workflow now exposes storage/worker health, single-question collection, observed-answer denominators and freshness, bounded website checks, and editable content drafts with revision-specific approval before implementation and retesting. These features reuse the existing account, database and pg-boss queue. The source edition is focused on AEO; desktop clients/installers and commercial SEO datasets are deferred. No upstream endpoints or results are invented.

The workspace, login, setup, settings and report previews now share a dark theme. Opportunities provides separate recommendations, actions/retests, observed-answer and website-check views. Latest collection failures stay visible; no-answer charts no longer imply an active job. EN/中文 labels, empty states and refresh guidance are clarified.

Implementation and migration

• Fixed native dependency version and exact platform install-script permissions, explicitly authorized by the owner. Other pnpm supply-chain controls remain unchanged.
• Migrations 0025/0026 add worker heartbeats, persisted site audits and draft review fields. Applied only to disposable/isolated BeatGEO databases during validation.
• Local launcher starts its owned database and applies migrations before application services, respects workspace configuration over ambient database variables, and stops owned services gracefully.
• Updated five README translations, setup guidance, metric definitions and acceptance report. Original third-party notices are retained.

Validation

• 1,478 unit tests; 20 browser contracts; 17 script tests; 46 PostgreSQL integration tests passed.
• Native persistence, rollback, eight concurrent connections and duplicate-instance rejection passed.
• Fresh isolated installation and UI registration, 12-question image-business starter set, worker readiness, failure display, draft/edit/approval/implementation/retest queue, EN/中文 and 390px layout checked.
• Restart retained account/session, brand, questions, audit and reviewed action; worker became ready again.
• One real ChatGPT consumer-UI collection via BeatBrowser completed after about five minutes: one submission, 11,180 answer characters and 18 visible citation links. Private account evidence is not committed.
• Workspace type checks, lint, web production build and 1,451-package license check passed.

The dark workspace groups visibility, share of voice, citations, answers and query fan-out under Answer Engine Insights with persistent common filters. It bundles Inter and Geist Mono locally, defines Chinese fallbacks, uses a 240px rail and compact controls, removes nested AEO frames, and renders unanswered questions as compact rows. Navigation structure and dimensions were checked against the signed-in Profound DOM and computed styles; pixel parity is not claimed.

Latest UI validation: 570 web unit tests, web typecheck, workspace lint and production build passed. Browser checks covered five analysis routes, retained date filters, sidebar selection, settings boundaries, EN/中文, 390px layout and virtualized empty-row scrolling. Synthetic fixtures were used; populated performance reports, PNG exports and printed output still need visual acceptance. Reference observations and implementation boundaries: docs/beat-geo/profound-ui-reference-2026-10-17.md.

Limits

Native installation was exercised on macOS arm64 only. Paid BeatAPI generation, other AI platforms, cross-day unattended operation and sleep recovery still need acceptance. Website-check UI/storage and failure behavior passed, but local DNS mapped public test domains to reserved addresses; the scanner correctly refused those destinations and reported unknown, so successful public-network crawling is not claimed. This is a development MVP improvement, not complete Profound parity.

Detailed evidence: docs/beat-geo/acceptance-2026-10-16.md.