The August 19 web-edge record lives on this page only.

What was actually in the room

The summary file describes a Windows 11 desktop with an Intel Core i9-12900K, on a 1 Gbps Ethernet port, with the NordLynx tunnel up for the whole collection. No public IP address, MAC address, gateway address, account identifier or home address was kept. The VPN masked the underlying ISP, so GSL Networks is the observed exit network, not a claim about the household provider.

There was no direct-ISP baseline and no alternate-route comparison. The record cannot say whether NordLynx helped or hurt. It describes this device, this exit, this date and this short window.

The summary file's own started_at and completed_at fields sit about fifteen seconds apart. Treat that as the file's timestamps. Do not read it as a reconstructed clock for every sample. The sample counts below are the part that was retained on purpose: three rounds, ten ICMP echoes and five TLS handshakes per round, per target.

The ICMP column does not separate the games

Look across the ICMP medians. Cloudflare's public control sits at 197.5 ms. Google's control sits at 197 ms. The four publisher web edges sit between 198 ms and 198.5 ms. The spread is about a millisecond and a half, inside a window where the slowest and fastest echoes on those hosts also overlap. From this exit, on this afternoon, an ICMP echo to a game publisher's website did not look different from an ICMP echo to the two public controls.

That is a finding about these hosts, from this tunnel. It is not a finding about Valorant gunfights, Tarkov raids, WoW worlds or Genshin co-op. A status page, a support site and a marketing site answer ICMP from whatever edge received the packet. A match server, a realm, or a co-op host is a different machine, often unpublished, often chosen after you queue. None of those machines were probed.

Thirty replies out of thirty also do not prove a route never loses a packet. A short collection misses rare bursts. A web edge can answer echoes while a game session does something else entirely.

The useful outlier is a TLS handshake, not a raid

TLS medians did not follow the ICMP column. Cloudflare's handshake median was 412.4 ms. Riot's status edge was 414.6 ms. The Tarkov site was 421.9 ms. HoYoverse support was 422.5 ms. Google's DNS host was 507.3 ms. The Blizzard support name, us.battle.net, was 1260.7 ms, while its ICMP median stayed at 198 ms with the rest of the set.

That gap is the point of keeping both columns. A slow TLS setup to a support website can happen while ICMP to the same name looks ordinary. The handshake includes connection setup. The echo does not. Neither number is Home latency or World latency inside World of Warcraft, and neither number is a license to rank a network product. If someone subtracted 198 from 1260 and called the remainder a WoW penalty, they would be mixing two operations and a website that is not a realm.

The same warning applies, more quietly, to the hosts whose TLS median sat near 420 ms. Closer to the ICMP column is still not the same measurement. It still is not a match.

Limits that used to be pasted into four guides

• The official targets are status, support or web edges. They are not match servers and cannot establish in-game ping, desync or play quality.
• NordLynx remained active throughout. GSL Networks is the observed exit, because the tunnel masked the household ISP.
• No direct-ISP baseline and no alternate-route A/B comparison were performed.
• ICMP loss here would mean missing replies from these hosts. It is not game packet loss. In this file, no ICMP replies were missing. TLS time includes connection setup and can move independently of ICMP.
• The charts are rendered from the summary. They are not screenshots of a game HUD.

How a later comparison should be built, including a Windows gateway check that this collection did not perform against the household path, is on the methodology page. Game-specific sheets are on the Valorant, WoW, Genshin and Tarkov guides. Those pages now say, in their own words, that they do not contain this table.

Editorial log

• August 19, 2026: collection retained. Sample counts, host names and the tunnel description were not revised later.
• October 10, 2026: the four game guides still embedded this table and told readers it was one shared study. That was accurate and still too easy to misread as four game tests.
• October 31, 2026, Jay Fan: the table, the environment block and the GSL / NordLynx limits were removed from the four guides and placed here. Guide pages keep a short pointer. No new probes were run. No match ICMP was added.

Corrections: redmedia24@yeah.net. Please quote the sentence and the file, and do not include account identifiers.