Our OAuth callback endpoint (GET https://crm.ativochat.com.br/api/google-calendar-callback, project "ativochat") is not consistently reachable when Google redirects the browser back to it after OAuth consent (a real top-level GET navigation from accounts.google.com).

Evidence:

1. Zero entries in Vercel dashboard Logs for these attempts, even with an explicit console.log as the first line of the handler.
2. 2. The DB value we persist on success never updates for these attempts.
   3. 3. Direct curl requests and same-origin fetch/XHR calls to the exact same endpoint work perfectly and log correctly every time.
   4. We tried switching the redirect_uri to the project's native ativochat.vercel.app domain instead of our custom domain crm.ativochat.com.br — it worked once, but then failed again on a subsequent identical attempt, so the issue appears intermittent rather than 100% tied to the custom domain.
3. Already ruled out: Vercel Deployment Protection (excludes production custom domains), Passport (unconfigured, Enterprise-only), Firewall (0 denied/challenged requests), Functions region settings, and domain configuration for crm.ativochat.com.br (shows "Valid Configuration", no redirect).

Could anyone from the team or community help me understand what happens at the edge/CDN level to these external GET requests that never appear in the dashboard Logs? Example failed-attempt window: August 14, 2026 between 22:55:50 and 22:57:05 (America/Sao_Paulo), immediately after a successful call to /api/google-calendar-connect at 22:55:50.

Project is on the Hobby plan, so direct support case submission isn't available for Compute/Functions — posting here as an alternative. Happy to provide more details or reproduce live if that helps.