What happens after someone drops in a YouTube link?
Read this left to right. The blue lane shows the person. The orange lane shows the durable work. The green lane shows what comes back to them.
person and UI processing result search projection
The story of one video
1. Someone wants an answer
2. The system does durable work
3. They get useful evidence back
Why a queue?Model calls are slow and billable. A page request should finish quickly, and a restart should not lose paid work.
Important rule:Partial evidence stays partial. The system avoids a polished summary that silently omits half the video.
Search can fail safely.D1 pages still work because the index is a convenient copy, not the only record.
the queue is just a D1 table, with leases
only the failed range gets smaller
Select any card above
Tell me why this piece exists.
The whiteboard carries the big picture. This panel follows the selected card and connects it to the code that owns it.
Starting point
A person arrives with a question
The product starts with a practical need: understand a video, reuse its strongest ideas, or find evidence later. The viewer stays server-rendered because the useful work is in the data, not a client application.