fix: add video agent failure diagnostics

[devhims]

A production comparison fetched all four transcripts successfully, but three analysis operations hit the research deadline and final synthesis was canceled. The response only reported generic transcript timeouts, making provider failures difficult to distinguish from application cancellation.

This adds separate fetch and analysis events, analyst queue duration and remaining research budget, and an event for every model attempt, including SDK retries. Run, video, and model-call identifiers connect analyst attempts. Cancellation is logged even when a provider call never settles, and late completion cannot overwrite that outcome. Errors retain allowlisted status, retryability, recognized error types, and provider request IDs when available; prompts, transcripts, credentials, and raw error bodies are excluded.

This changes diagnostics only. The response schema, model, retry limits, and 60-second deadline are unchanged. These logs do not expose Cloudflare's internal queue versus inference duration, and do not establish an infrastructure incident by themselves.

Validation:

• Platform TypeScript build passed.
• All 364 unit tests across 52 files passed.
• Regression coverage includes a hanging provider call canceled by the research deadline, late completion, safe error logging, and a retryable 503 followed by success through the production model wrapper.

No production deployment has been performed for this change.

[vercel]

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

Project	Deployment	Actions	Updated
video2ctx-web	Ready	Preview	Sep 5, 2026 9:28pm UTC