Archive · sealed 30 June 2026

What we captured — and what we didn't

These figures are final. The official servers are gone, so no new traffic can be captured and nothing on this page can move again.

To rebuild Blades before it shut down we had to record how the game talked to its servers. We went out of our way to capture the minimum — only Blades and its arena servers, never the rest of anyone's phone. Here is exactly how, in the open.

153 contributors signed up to help preserve the game. What they routed through us is the whole of what survives.

We routed the minimum, on purpose

A VPN could have sent all of a phone's traffic to us. We deliberately didn't. We reverse-engineered the game's own matchmaking protocol so we could whitelist precisely the servers Blades used — and leave everything else (banking, messaging, browsing, every other app) on the normal connection, untouched.

Routed through us
  • *.blades.bgs.services — login, characters, matchmaking, the asset CDN
  • The AWS GameLift ranges for the 14 regions arena matches could run in
Never touched
  • Everything else on the device bypassed the VPN entirely
  • Non-Blades HTTPS that did route here was passed straight through — we could not decrypt it, only blades.bgs.services

Why arena needed more than one IP

Arena PvP ran on Amazon GameLift. Before a match, the game pinged 14 AWS regions and reported the latencies; GameLift then placed the match on whichever region was fastest and had room, and handed the phone a different server IP for every single match. There was no fixed arena IP to whitelist.

So to reliably capture arena we routed those 14 regions' GameLift ranges — and only those. We confirmed the exact list of 14 three independent ways (the client's built-in endpoint table, its latency-probe targets, and the live matchmaking requests we captured), so we can be confident we were not missing a region and not over-reaching into one we did not need.

The complete IP list was written in plain text into every .conf we handed out — open one and read every line.

Real arena ping, measured by the game itself

Everyone used to argue about arena lag. Here is the answer, now permanent: the actual latencies Blades measured to each region, pulled from 25,772 captured matchmaking requests. Measured from where our contributors were (mostly Europe & North America), so Best is the realistic floor for someone sitting near that region; Typical is the median across all our samples.

RegionBestTypical
Europe (London) eu-west-210 ms70 ms
Europe (Frankfurt) eu-central-112 ms67 ms
US East (Ohio) us-east-214 ms157 ms
South America (São Paulo) sa-east-116 ms264 ms
Europe (Ireland) eu-west-118 ms79 ms
Asia Pacific (Seoul) ap-northeast-220 ms286 ms
US West (N. California) us-west-120 ms201 ms
US West (Oregon) us-west-220 ms188 ms
US East (N. Virginia) us-east-121 ms137 ms
Canada (Central) ca-central-121 ms140 ms
Asia Pacific (Mumbai) ap-south-122 ms187 ms
Asia Pacific (Tokyo) ap-northeast-122 ms282 ms
Asia Pacific (Sydney) ap-southeast-223 ms312 ms
Asia Pacific (Singapore) ap-southeast-123 ms222 ms

When was arena busy?

Inferred from how fast the global top-100 leaderboard gained cups between observations — so it reflects the wider playing population in the game's last months, not just us. Darker = more matches. Times are UTC. We could only sample when a contributor had the leaderboard open, so coverage is patchy and limited to top-100 players — a real signal, not a complete census.

0
6
12
18
Mon
Tue
Wed
Thu
Fri
Sat
Sun
less
more