A beginning with the full system in view.

Individual Thing records with provenance, canonical pages, persistent anonymous history and saves, a ranked discovery pool, cross-category diversity penalties, linked rabbit holes with automatic cross-source matches, automated catalog intake for Gutenberg books, Met, Cleveland, Art Institute of Chicago and Smithsonian objects, Library of Congress free-to-use collections, NASA images, and attributed Wikipedia previews, and an owner-only operations view. Seen Things are excluded from organic recommendations.

Distinct known-item and similarity feedback, quarantined suggestions, visitor-level hiding for unsafe-content reports with automatic holds after three browser identifiers, data export, and history deletion. An initial Boredom Rate definition is recorded in the project documentation.

This curated collection exercises the first workflows. Source-validated catalog entries enter discovery automatically; safety holds stay outside recommendations. It is intentionally labeled as development data. Exhaustion shows an honest end state and preserves the Seen Ledger.

A scheduled importer independent of site visits, semantic entity resolution, stronger automated source safety checks, expanded moderation administration, deeper semantic graph expansion, and recommendation evaluation across 100, 1,000 and 10,000 discoveries. Production volume targets remain hundreds of thousands of approved Things.

Verified passkey registration and cross-device account migration, account recovery, abuse resistance, operational monitoring, tested backup restoration, and sustained load testing. Public search indexing remains disabled until launch readiness is verified.