Hello Vercel team, The Vercel Help assistant directed us to Community because CDN / Edge Network cases are not available through the support form on our Hobby plan. We want to keep hosting on Vercel and have postponed adding a custom domain. Could a Vercel staff member review this regional connectivity issue and confirm any supported mitigation for managed vercel.app domains? Affected production: https://coach-os-lac.vercel.app/android.html The same production deployment is also available at https://coach-os-git-main-silvernil.vercel.app/android.html. On 16 November 2026, approximately 08:50–09:15 UTC, read-only tests from a Windows machine on a Russian network used curl bound to the physical Ethernet interface with proxies disabled. We privately confirmed the bound connection has a different public egress from the VPN route. We did not change system VPN, DNS, routes or firewall settings. With hostname/SNI and HTTPS certificate validation preserved: | Vercel ingress IPv4 | Production domain | Main alias | | --- | --- | --- | | 216.198.79.195 | HTTP 200 | HTTP 200 | | 216.198.79.131 | TCP connect timeout | TCP connect timeout | | 216.198.79.3 | TCP connect timeout | TCP connect timeout | The main alias initially resolved to .195. The complete 24,156,900-byte APK downloaded in 1.54 seconds and matched its expected SHA256. All ten native API CORS preflights returned 204; the Android origin https://localhost was allowed. Protected unauthenticated POSTs correctly returned 401. After DNS answers changed, ordinary Ethernet-bound HTTPS to the main alias also timed out. Windows DNS then returned .3; Google DoH returned .67 for both names; a Kazakhstan VPS resolver still returned .195. We did not directly test .67 or identify the exact upstream filtering/routing party. Changing aliases therefore did not provide a durable fix, and the candidate Android build switching aliases was not published. From Kazakhstan, the production page and API preflight return 200 and 204. Supabase Auth settings and a zero-row REST health check also return 200 from the affected direct Windows connection. Password, SSO and trusted-IP deployment protection are disabled. No customer data, authenticated live actions, paid API calls or account mutations were used in these tests. Reproduction (replace ETHERNET_IPV4 with the local physical interface address): powershell curl.exe --noproxy '*' --interface ETHERNET_IPV4 --connect-timeout 5 --max-time 8 -v https://coach-os-lac.vercel.app/android.html curl.exe --noproxy '*' --interface ETHERNET_IPV4 --resolve coach-os-lac.vercel.app:443:216.198.79.195 --connect-timeout 5 --max-time 8 -v https://coach-os-lac.vercel.app/android.html The legacy now.sh name redirects to vercel.app. We understand Functions Static IPs cover outbound connectivity and do not fix the incoming path. We are not proposing hardcoded shared IPs in Android, disabled TLS validation, or requiring users to enable VPN. Is there a known regional routing issue affecting these ingress VIPs? Is any supported Vercel-side mitigation available while retaining managed vercel.app domains? If not, please confirm that limitation and advise which additional non-sensitive diagnostics would help the relevant network team investigate. The problem is still reproducible and should not be marked resolved based on the successful .195 control alone.