Review Moderation & Content Quality

Browse the synthetic campsite catalog, filter by region and facilities, open a campsite and write a review - pick a language, give a score, and submit. The result view shows the verdict, the reason and the band. Published reviews appear immediately on the campsite page; held or rejected ones never reach the public listing.

Open the inspector to see the live per-review pipeline trace on the uniform trace-row contract - each stage's provenance, cost, confidence and band. A Cloud/OSS toggle and cost meter cover the metered stages, which fail closed to the $0 OSS path at the budget cap. Recorded OSS-vs-Cloud divergences show the borderline reviews where the two verdicts differ, including one honest "not captured" case. Everything is labelled synthetic and simulated; OSS inference is $0.

Operators see the dashboard and the held + rejected queue in priority order, each item carrying its verdict badge, language, confidence, the AI reason and the topic. A viewer credential is read-only; an owner can approve, hold or reject, and every action is logged to the audit trail.

The owner-only Configure section tunes a two-layer rule set - an immutable baseline (toggle or re-band only) plus an owner overlay (add or delete keyword rules) - alongside the confidence thresholds, the topic taxonomy and the reviewer roster, with a one-click reset to baseline. Re-banding a rule downgrades a band live (auto-publish → review, auto-reject → auto-hold). The project-page walkthrough video narrates the full tour.