Four Migration Windows, 24,000 Moves — Who Leaves and Where They Land

Every number here comes from before/after top-200 hero-power snapshots of four migration windows in July–August 2026 (groups 293–420, 421–548, 549–676 and 677–804). That’s 24,068 movers in total. For direction, hub and alliance analysis we only used the two windows where we could match accounts by uid (10,436 movers). The full per-window mover lists are in each migration report.

After every migration window, alliance chat fills up with the same kind of gossip: “a whole alliance just landed next door”, “that server got gutted this time”. We wrote a report for each window with per-server inflow and outflow tables. Lay four of those tables side by side, though, and a few patterns jump out that you’d never see looking at one window.

This post is about those patterns: who actually moves, where people pile up, and whether that pile-up is individual choice or an alliance decision. At the end there’s a section on a conclusion we very nearly published that turned out to be an artifact of our own matching method.

One in four moves

In all four windows, roughly one in four players in the top 200 changed servers. That’s a lot. On my own server the ranking screen looks noticeably different after every window, and it turns out that’s just normal.

Note the gap between the last two numeric columns: the share of power that moved is always 1–2 points lower than the share of people. Movers are a bit weaker than average.

It’s the middle that moves, not the top

Divide the average mover’s hero power by the group average and you get 0.91–0.96. Median movers sit around 175–194M hero power, and the top 1% of movers account for almost exactly 1.8% of all moved power in every window. This isn’t a few whales flipping the map — it’s the middle of each server reshuffling en masse.

I expected the opposite. Migration seats are ranked by hero power, so the strongest players get the most choice, and I assumed migration would be a top-end game. But the top end is usually already settled: running an alliance, holding the server government. They have the least reason to leave.

Hubs are built by alliances

Call the 13 servers with the most arrivals in each window (the top 10% of the group) “hubs”. Look at how they filled up and the answer is obvious.

In 293–420 the biggest hub took in 123 players, 53 of them from a single alliance. The second took 122, and 87 of those were one alliance. In 677–804 the top hub’s 78 arrivals included 44 from one alliance. A hub is less “the server everyone agreed was good” and more the server one big alliance moved into wholesale.

Moving as an alliance is the default overall, too. In the two uid-matched windows only 18–19% of movers travelled a route with no one else from their alliance; 38–45% moved in packs of ten or more, and every window had 9–20 routes where 30+ people from one alliance went together.

Pack size has nothing to do with strength, by the way. Median hero power for solo movers, packs of 2–9, 10–29 and 30+ was 191–197M in 293–420 and 181–186M in 677–804 — basically flat. People following their alliance aren’t weaker players being dragged along.

Strong individuals avoid the hubs

So where did the strongest movers go? We pulled out the top 10% of movers by hero power.

In 293–420, 22.3% of all movers landed on a hub, but only 13.8% of the top 10% did. In 677–804 it was 18.3% vs 16.5%. The top-decile movers spread across 91–97 of the 128 servers.

I’d treat the “why” with care, but from an R4’s chair it makes sense. A strong solo player gains little by walking into a hub that a big alliance already runs. Picking a server where you can be the main force — or one where a few friends already are — is the more natural move.

Servers bleed in clumps and gain in trickles

Per-server net flow (in minus out) is strikingly lopsided.

A small set of 16–20 servers lost big, and the power they lost was spread across about 40 others. 677–804 is the extreme case: only two servers netted 50+ players, and the biggest gain was +52. One side collapses, the other side gets a slice each.

Routes are narrower than you’d think, too. Between 128 servers there are 16,256 possible routes; only 1,229–1,339 (about 8%) were used. And among routes where someone went A→B, only 16% also had someone going B→A. Servers rarely trade players with each other — flows mostly go one way.

Older servers or newer ones?

You hear “everyone migrates to the older (lower-numbered) servers” a lot. The data doesn’t back it up. The share moving to a lower-numbered server was 43.9%, 49.6%, 49.6% and 52.4% across the four windows — basically a coin flip — and weighting by power doesn’t change that.

There is a mild locality effect, though. Split each group into its lower and upper half by server number, and 59% (293–420) and 64% (677–804) of movers stayed within their own half. Random choice would give about 50%. Servers in the same half opened a few weeks apart, so season progress and hero unlocks line up — that’s a plausible reason to stay close.

The conclusion we almost got wrong

The middle two windows (421–548 and 549–676) matched accounts before and after by player name, because back then we didn’t store uids. Put all four side by side and those two look odd:

• They used 2,580 and 2,783 distinct routes — double the uid windows (1,229–1,339).
• The correlation between a server’s inflow and outflow was 0.89. In the uid windows it was −0.23 and −0.07. In other words, the name-matched windows say busy receivers were also busy senders — the exact opposite of the clean hub/donor split in the uid windows.
• Only 22–23% moved in packs of ten or more, about half the uid-window rate (38–45%).

My first draft said “in these two groups people moved as individuals and servers traded players back and forth.” But name matching counts someone as a mover whenever a different account with the same name exists on another server. Fake moves like that land on random routes (so flows look scattered) and appear in both directions evenly (so inflow and outflow rise together). That is precisely the fingerprint those two windows have.

So the direction, hub and pack analysis in this post uses only the two uid windows, and the 27–28% move rate for the name windows is probably a bit inflated. From 677–804 onward every report is matched 100% by uid.

Using this when you plan a move

Hubs already have an owner. A server with huge inflow is usually where one big alliance moved wholesale. Walking in alone is close to joining under that alliance. Unless that’s the plan, don’t pick a target by arrival numbers.

A server that just bled out also has room. The top net-loss servers lost 10B+ hero power in one go. Rough for whoever stayed — but from the outside it’s a server with less competition, which is probably part of why strong movers scattered instead of piling into hubs.

If you’re moving as an alliance, line up the route in advance. Every window had 9–20 routes with 30+ people from one alliance. Seats are limited per tier, and if another alliance targets the same server in the same window, only part of your group may get through. Compare target servers’ top end in the migration finder before you commit.

Caveats

The sample is each server’s top 200 by hero power only. Moves below that are invisible, and players hovering around rank 200 right after the window may be missed or miscounted. Alliance tags are the tag after the move, so anyone who switched alliance while migrating is counted as a solo mover. And all four windows are Season 6 servers (293–804). Smaller 64- and 32-server groups of newer servers may behave differently. We’ll update these numbers as windows from other ranges come in.

Unofficial fan-site content · no partnership or sponsorship with the developer · observational analysis from collected data.

By: lastwarhub.com