Dataset Catalog

External imports enter the catalog only after blocking validation passes. Catalog provenance records support level, adapter/version, original relative paths and hashes, preservation/portability, parsed and missing telemetry, synchronization confidence, calibration, chapters, streams, transforms, warnings, and completion state.

The dataset catalog is the device-local, searchable index of research sessions. A session package contains the immutable capture manifest and media; a catalog record points to that package and stores editable organization data such as name, tags, favorite status, rating, collections, and review state.

Phase 4 stores catalog metadata in an atomically written, versioned JSON document (catalog-v2.json) in Application Support. This keeps the implementation identical on iPhone, iPad, and macOS and provides explicit backup-before-migration behavior. It is intentionally isolated behind the DatasetCatalog actor so a future high-volume SQLite adapter can replace it without changing UI or capture code. Schema version 2 is current.

Search covers names, notes, tags, device and camera information, codec, source, and provenance. Filters combine review/completion state, source, provenance, collection, location availability, favorites, integrity state, package size, and date. Sorting is deterministic.

Catalog metadata never rewrites the finalized capture manifest. Package locations are device-local catalog fields, not portable manifest paths. Simulated records carry both simulated provenance and a visible SIMULATED label.

Current limits: the catalog UI demonstrates local catalog workflows and simulated import. Files/share-sheet pickers, ZIP expansion, raw-video metadata extraction, and production background scheduling require later platform adapters; unsupported actions return explicit errors or guidance.