We are preparing a database migration for our Hobby project, Recoverly, and need to ensure old HTTP send functions have finished before proceeding.

We have four Node.js deployments with POST /api/send-reminder, each with a confirmed execution limit of ≤300 seconds. The route awaits its database authorization transaction, one email-provider call, and finalization. No application-level background queue or automatic resend was found in these deployments.

Let T be the time when WAF denial has been verified on every old deployment URL and alias.

1. Is T + 300 seconds a guaranteed deadline by which all HTTP requests admitted before T have completed or been terminated?
2. Does maxDuration include the entire interval between passing the WAF and starting execution, including dispatch, cold starts, and any concurrency waiting?
3. If not, what guaranteed maximum additional delay Q applies? Can the platform retry an already-admitted HTTP invocation after T?
4. If no finite bound is guaranteed, which read-only view/API available on Hobby can establish that no pre-T requests remain pending or running, and identify the last execution start?

The Vercel support AI stated that synchronous HTTP requests have no asynchronous holding queue or automatic platform retries. However, the linked documentation did not establish the maximum pre-execution delay or the T + 300s guarantee.

Could a Vercel team member clarify this with official documentation or recommend a supported way to safely drain old deployments?

We are not asking whether an email provider accepted a request; we handle unknown provider outcomes separately and will not automatically resend.

Thank you.