fix: recover video metadata and retain it for follow-ups

[devhims]

Video view-count follow-ups could lose previously fetched metadata and receive a YouTube bot challenge wrapped in HTTP 200. That response bypassed processor failover and could be cached as usable metadata.

This change retries bot-challenged metadata on the alternate processor, rejects challenged cache entries, and retains good stale metadata when refresh fails. If neither processor nor the cache provides usable metadata, the API returns 503 UNAVAILABLE. Private and age-restricted responses remain valid.

Follow-up runs now restore bounded video metadata from persisted provider evidence along the selected conversation ancestry. The model receives view counts, source citations, and explicit freshness information; current-run metadata supersedes historical snapshots. Reusing this evidence adds no provider usage charge. View counts and freshness also appear before potentially truncated descriptions.

No migration is required. This improves resilience but does not resolve the underlying processor egress bot challenge.

Validation:

• TypeScript build and 519 unit tests passed.
• 30 local Worker integration tests and 16 processor tests passed.
• The metadata restoration integration test passed again after the final query and isolation changes.
• Generated OpenAPI consistency and offline Mintlify validation, links, and accessibility checks passed, with existing theme contrast advisories.
• git diff --check passed.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
video2ctx-web	Ready	Preview	Sep 11, 2026 2:52pm UTC