Open the reader · Inspect import coverage
The Vani collections span books, quotations, articles and media across several editorial websites. A useful reading interface needs to connect those materials without losing the source, edition or context of a passage. It also needs to distinguish material already stored from material that still depends on the original website.
The reader combines a TypeScript/Next.js interface with a versioned corpus and local search index. Source records preserve canonical titles, original URLs, available revision identifiers, language, edition and import dates. Different editions remain distinct.
| Concern | Design choice |
|---|---|
| Interrupted imports | Bounded page jobs, persisted progress and a rotating refresh cursor |
| Temporary source failures | Retain the last good page and record the failed refresh |
| Renamed or missing pages | Preserve aliases; require repeated missing-page confirmation before removal |
| Search behavior | Separate stored-index search from explicit live-source search |
| Source fidelity | Preserve attribution and edition metadata; sanitize HTML before rendering |
| Integrity | Verify page hashes, identity, indexed text and search postings |
| Reader continuity | Browser-local bookmarks, preferences and saved reading position |
| Coverage | Publish discovered/imported counts, pending work, failures and unknown whole-site totals |
The interface supports book/chapter/verse navigation, source-linked citations, quotation context, native media controls and keyboard/mobile interactions. Browser-local saved state does not synchronize across devices.
The Azure deployment profile uses a standalone Node application behind a dedicated Cloudflare Tunnel. Releases include versioned corpus assets and an application health endpoint. The documented release procedure builds and checks a release before switching the active application, with a previous release available for rollback.
An earlier Worker deployment is retained as a separate fallback. Source refresh, repository synchronization and production deployment are distinct operations.
The 20 October 2026 endpoint check reported:
| Collection | Imported source pages |
|---|---|
| Bhagavad-gītā selection | 680 |
| Selected supporting pages | 39 |
| Śrī Īśopaniṣad | 21 |
| Nectar of Instruction | 13 |
| Śrīmad-Bhāgavatam, Canto 1 | 799 |
| Reported unique imported source pages | 1,551 |
The collection counts sum to 1,552, while the global import count and the source-by-source breakdown both report 1,551. Collection counts are therefore not treated as disjoint unique-page totals. The public snapshot reported zero failed and pending jobs for the selection. The endpoint check confirms reported metadata consistency, not an independent comparison of every page with its source.
Earlier recorded browser checks covered navigation, citations, saved-state persistence, quotation context, native audio playback and mobile keyboard behavior. They are historical representative checks, not a fresh full-site acceptance run.
The current corpus/index model is bounded. Importing an entire large archive would require a scaling plan for indexing and storage. The UI is English; preserving original scripts does not establish a complete multilingual interface. Audio and video can remain hosted upstream even when their page text is imported.
This work demonstrates the connection between frontend behavior, ingestion reliability, source attribution and deployment operations.