A private back office for evidence-backed review sites
The system collects crawl, render, screenshot, and Lighthouse evidence; turns it into configurable scores and comparisons; and keeps an operator between drafting and publication.
695-item v1 checklist substantially complete · final server validation still pending
How the pieces connect
- Crawl, render, screenshot, and Lighthouse
- Evidence records
- Configurable scoring
- Comparison draft
- Operator approval
- Static publication
The job
Review and comparison sites become unreliable when evidence, scoring rules, drafts, approvals, routes, and publication live in different tools.
The hard part
Collect evidence deterministically, make scoring explainable, limit model work to bounded drafts, and require operator approval.
How I built it
- Keep raw evidence separate from interpretations and editorial copy.
- Make scoring formulas and source connectors configurable and auditable.
- Use queues for collection and drafts, but keep publication as an explicit operator action.
- Generate public routes and sitemaps from the same accepted publication state.
What I verified
- Go API, PostgreSQL, NATS workers, React back office, and Astro public layer
- Crawl, rendered-page, screenshot, and Lighthouse evidence
- Scoring, comparisons, alternatives, search, approvals, connectors, and notifications
- 695-item v1 checklist with final production-server validation still open
Current state: The back office and publishing workflow are built. Final validation on the production VDS is still open; publication remains an explicit admin action.