[shashank03-dev]

Production analysis can wait on Gemini for nearly two minutes before failing with HTTP 503. The browser displays timer-driven progress capped at 94%, including misleading completed-compilation labels, without a request deadline.

• Retry HTTP 500/502/503/504, network errors, and request timeouts up to three times per model, with exponential backoff and jitter. Each request has a 15-second timeout; primary and Pro-to-Flash fallback share a 50-second budget.
• Add a 65-second browser deadline covering frame preparation, API requests, and response reads. Preserve manual cancellation and retain extracted frames for retry.
• Label the animation as estimated progress and waiting for a model response, rather than claiming compilation/audit completion.
• Preserve existing failed-analysis quota rollback.

Validation: unit tests cover temporary recovery, exhaustion, permanent errors, network failure, abort deadlines, fallback, browser cancellation, stalled operations, and waiting-screen labels. Typecheck, lint, and the full unit suite are run locally. The initial server-fix preview built successfully on Vercel; updated preview/CI checks run on this head. Local builds are blocked by an environment-specific ENOENT uv_resident_set_memory error; graphify is not installed.

Supabase configuration is unchanged. A live authenticated AI analysis has not been run for this branch.

[vercel]

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

Project	Deployment	Actions	Updated
motion-code	Ready	Preview	Nov 6, 2026 8:12am UTC