Spoken Archive Engine v2.0
The v2.0 release marks the production-domain milestone for the Spoken Archive Engine and confirms that the public frontend, production API, and evidence-aware retrieval interface are working together as a deployed archival research product.
Production milestone
The Spoken Archive Engine is now available at its production domain:
www.spokenarchiveengine.com
What v2.0 includes
- Production frontend deployed to S3 and CloudFront.
- Production domain configured and working.
- API-backed archive search against the current public collection.
- Grounded answer generation from retrieved transcript evidence.
- Support level display: strong, moderate, or limited archive support.
- Theme display for retrieved answer context.
- Evidence cards with cleaned episode titles and transcript chunk references.
- Independent non-affiliation disclaimer.
Backend retrieval improvements
Phase 2F introduced material retrieval-quality improvements, including a larger retrieval candidate pool, evidence diversity selection, nearby same-episode deduplication, composite support-level calibration, retrieval expansion rules, and a broad-synthesis multi-profile retrieval path.
These changes make broad archive questions more useful while preserving caution around completeness. Broad-query answers should be read as grounded snapshots from retrieved passages, not exhaustive surveys of the full archive.
Current public collection
The current public collection covers Down to Brass Tacks episodes from 2024/5 to early April 2026.
The broader source archive spans 2021–2026 and includes roughly 1,177 episodes. Full-collection expansion remains ongoing.
Design direction
The v2.0 interface continues the product direction toward a sober archival research tool rather than a chatbot, AI demo, or generic semantic search page. The interface uses the language of evidence, support levels, retrieved passages, and public record.
Known limitations
- The current public queryable collection is partial, not the full 2021–2026 source archive.
- Support level reflects selected archive evidence, not independent factual verification.
- Caller claims are part of the historical record and should not be treated as verified facts.
- Broad answers are grounded snapshots, not complete archive-wide surveys.