Interactive Schools Ltd, which builds and hosts websites for educational institutions, encountered a troubling support case when one of its client schools reported that certain parents and staff members were unable to load the site from home connections, though campus access worked without issue. The school's domain sat behind Cloudflare infrastructure shared with other school sites that had no reported problems, making the diagnosis difficult.

Ciprian Lucanu, Software Development Team Lead at Interactive Schools Ltd, explains the challenge: "'It doesn't load for some people' is a hard report to act on. The cause could be the website, the CDN, DNS, the visitor's device, their home router or their internet provider."

Packet captures reveal network-level interference

Working with Cloudflare's support team, the investigation began with Wireshark packet captures collected by school staff across four connection types: the campus network, a residential cable connection, a T-Mobile mobile connection, and a commercial VPN service. Captures included both the affected school's domain and a second school's domain on the same Cloudflare IP addresses, serving as a control.

Two distinct blocking mechanisms emerged from the data. On T-Mobile, TLS connections were being intercepted: after the TCP handshake with Cloudflare completed normally, the carrier injected a 256-byte response composed entirely of 0xFF bytes followed by three TCP resets when the browser sent its TLS ClientHello with the school's hostname in the SNI field. The browser treated this garbage data as a broken server response and disconnected with a TLS protocol_version alert.

The injected packets were identifiable by their TTL of 255, compared to legitimate Cloudflare responses with TTLs in the 40s after traversing multiple routers. Plain HTTP requests to the same hostname received a forged 302 redirect to a T-Mobile security warning page, identifying the filter. Notably, the control domain on the same Cloudflare IP loaded normally on the same connection, indicating the filter matched the SNI hostname rather than the destination IP address.

The VPN service employed a different blocking method: its DNS resolver returned NXDOMAIN for the school's domain and added an additional record, explanation.invalid TXT "Proton DNSBL/1b", marking the response as coming from a blocklist. The browser never attempted a connection.

Both security products blocking the same domain pointed to a shared reputation feed. Investigation of public domain blocklists found the domain listed on SURBL's ABUSE list (127.0.0.64 from multi.surbl.org), while the control domain was not listed. SURBL's lookup page confirmed the listing, and the school began the removal process.

The limits of external measurement

The school's natural follow-up question—which internet providers block the domain?—proved difficult to answer through conventional means. ISP filtering typically operates inside provider networks: in DNS resolvers, customer gateways, or in-path equipment like the T-Mobile interceptor. Querying those resolvers from outside fails because most restrict access to non-customers. VPN exit nodes offer no help, as they reside in data centers rather than on residential ISP networks and apply their own filtering. Public measurement data from OONI contained no US measurements of this domain.

Answering the question required vantage points inside each provider's network on actual home connections—precisely what RIPE Atlas provides.

Measuring across 21 US ISPs with RIPE Atlas

Interactive Schools Ltd had no existing RIPE Atlas credits, so the team contacted the RIPE Atlas program, explained the investigation, and requested a small grant. The RIPE Atlas team credited the account with 5,000 credits within one day, enabling the next phase of work.

Probe selection strategy

The team mapped the 21 largest US residential providers to their ASNs, accounting for multiple networks within single companies: Charter's former Time Warner Cable and Bright House networks, Altice's Optimum and Suddenlink, Verizon's wireless ASNs, Frontier's legacy networks, and Astound's RCN, Grande and Wave networks. For each provider, connected US probes were selected using the public probe API with these criteria:

  • status=1, country_code=US, asn_v4 in the provider's ASN list
  • required the system-ipv4-works tag
  • excluded probes tagged datacentre and anchors
  • preferred probes their hosts had tagged home

Coverage varied significantly across providers. Comcast offered over 200 eligible probes, AT&T over 160, and Verizon Fios over 140. Mediacom and Cable One each had exactly one. To minimize costs, two probes per provider were selected, totaling 40 probes.

Measurement design

Three one-off measurements ran on each probe to keep costs minimal:

  1. DNS: an A query for the school's www hostname with use_probe_resolver: true, sending the query through the probe's network-assigned resolver, typically the ISP's, to detect DNS-level filtering like the VPN case
  2. TLS (sslcert): a TLS handshake with the school's www hostname, with hostname set to send the SNI field and resolve_on_probe: true, to detect SNI-based interference like the T-Mobile case; probes receiving the site's real certificate showed no interference
  3. Control TLS: the same sslcert measurement against the control school's hostname on the same Cloudflare IPs, distinguishing domain-specific blocking from complete Cloudflare unreachability

Results were classified by script: DNS answers containing only Cloudflare addresses were marked "OK", while NXDOMAIN or other addresses indicated blocking. TLS handshakes returning certificates from Cloudflare addresses were "OK"; alerts or resets indicated blocking.

Cost and results

The DNS measurement cost 20 credits per probe (800 for 40 probes). Including a re-run described below, the investigation consumed slightly under half of the 5,000 granted credits.

Remarkably, the blocking was not reproduced on any of the 44 usable probes across 21 providers. Every DNS answer returned real Cloudflare addresses, and every TLS handshake returned the site's legitimate certificate. Of 48 probes scheduled, 45 returned results, with 44 usable.

The two T-Mobile probes in the initial run returned no results. Since T-Mobile was the only provider where blocking had been observed in packet captures, the team re-ran all three measurements on eight different T-Mobile probes. All eight reached the site normally.

Key findings

  • TTL and IP ID values quickly identify injected packets. A "server reply" with TTL 255, IP ID 0 and no DF flag arriving moments after a ClientHello almost certainly originated from path equipment, not the actual server
  • Always measure a control hostname on identical infrastructure. This transformed "it's slow / broken" into "this specific hostname is being filtered" and allowed false positives from broken probes to be discarded
  • RIPE Atlas answers questions impossible to resolve from outside a network. No other affordable method existed to obtain DNS and TLS results from inside 21 US residential providers in a single afternoon; use_probe_resolver and sslcert with explicit hostnames directly mapped to the two filtering methods observed in packet captures
  • Coverage remains uneven, requiring planning around gaps. Large cable and telco networks have excellent coverage, while several regional providers have only one to five probes, limiting conclusions about their filtering practices
  • Credits represent the primary barrier for new users. Starting with zero credits, the request process proved quick and welcoming. For organizations like Interactive Schools Ltd that don't operate networks and aren't RIPE NCC members, the grant made measurement feasible

Acknowledgments

The investigation credits the RIPE Atlas team for the credit grant and rapid response, and the probe hosts on US home networks whose probes conducted the actual measurements. The work demonstrates a practical but less common application of the platform: helping schools and their web providers determine whether "the website is down" reflects an actual website problem. The school's team received thanks for their patience and extensive assistance in gathering packet captures across various locations and conditions.