Half a Million Light-Years Saved: Eleven Months of EF-Map Usage Data

EF-Map has been counting anonymous usage since the end of August 2025. No accounts, no tracking of who you are, just counters: a route was calculated, a panel was opened, a theme was switched. All of it is public on the stats page, and I recently added a performance section to it, which got me curious about the rest of the archive. So I had an agent go digging through the full 335 days of it to see if there was anything worth writing up. Turns out the numbers tell a story I did not expect them to tell: you can watch the game changing underneath the site.

The rise and fall of route planning

EF-Map exists because getting around EVE Frontier used to be genuinely hard. You had a jump drive with a fuel cost, you died if you got it wrong, and so people planned. Lifetime the site has calculated 85,361 point-to-point routes against 95,692 page loads, which is not far off one route per visit. In October 2025 alone it was 21,823 routes, more routes than page loads that month. People were planning constantly, recalculating, comparing options.

Then the game changed underneath the site, twice. Travel got easier over the spring, so people could just set off and wing it rather than planning every jump. And in the current cycle ships no longer have jump drives at all, so there is very little for a route planner to do. July 2026: 674 routes. That is a 97% fall from the peak. The site itself did not die with it, July still saw 7,716 page loads, they are just doing different things now: solar system view is at its highest ever (1,612 views in July), and since panel tracking started in mid July the killboard, activity tracker and smart assembly panels are getting the attention routing used to get.

What replaced jump planning is yeeting, which is the game's method of flinging yourself between stars and hoping. I added a Yeet Planner for it, and since 17 July it has run 496 plans, of which 463 found a workable course, a 93% hit rate. Two weeks of yeet plans is already most of a month of route calculations at the current rate, so the traffic did not vanish, it changed shape.

The engine got about five times faster, and you can see it

The other thing hiding in the routing data is our own engineering, because every calculation reports how long it took. In November 2025 the average route took 3.8 seconds to calculate. That month the router was mostly running an A* implementation in JavaScript and people were asking a lot of it. In December Dijkstra became the default and the average dropped to under a second. In February 2026 the whole thing was rewritten in Rust compiled to WebAssembly, and since March the average has sat between 0.7 and 1.0 seconds, on bigger and uglier route requests than the early engine ever handled.

I will hedge this one properly: average calculation time depends on what routes people ask for, and a long exploration-mode route costs far more than a short hop, so this is not a clean benchmark. But a 5x drop that lines up exactly with two engine changes is not an accident either. Across the 77,185 timed calculations the site has done about 34 hours of cumulative pathfinding, all of it client-side in your browser, which still amuses me. Your machine does the work, I just serve the map.

Half a million light-years saved

My favourite number in the whole archive. The scout optimizer takes a set of systems you want to visit and reorders them into a shorter trip. It has only been run 555 times, but the average optimization saved 952 light-years, and the total across all of them is 528,456 light-years of travel that never had to happen. The average scout run it was fed contained about 98 systems, so these were not toy inputs, they were people planning serious survey work.

Smart gate routing tells the opposite story, another feature the game changed out from under me. Routes through player-built smart gates averaged 4.1 gate hops and peaked in November 2025 at around 2,100 gate-aware routes in the month. Then the game hid gate location data, and you can watch the feature die in the counters: down to a trickle by February, effectively zero since May. The code still works. There is just nothing for it to see.

One more routing habit worth recording: when given the choice, 88% of routes were optimized for fuel, 10% for fewest jumps, and about 1% used exploration mode. Frontier players are misers, which anyone who has paid for jump fuel will understand. And sharing follows the same pragmatism. The site has generated 1,034 shareable route links, each opened about 3.7 times, but in the same period people hit copy-route-to-clipboard 17,359 times. Given the option of a nicely unfurling link or pasting waypoints straight into the game, players pick the clipboard 17 to 1.

My biggest user was an iframe

Lifetime, cinematic mode has 31,530 entries, which would make it one of the most used features on the site. It is not. A partner site had the cinematic view embedded on their page for a few months, and every visitor they got counted as an entry.

The embed came down at the end of January and the counter fell off a cliff, 3,662 entries in January, then 13 in February. Real human usage of cinematic mode since then runs somewhere between 13 and 122 entries a month. I am leaving the lifetime number on the stats page because it is honest in its own way, but if you ever see me quote 31,000 cinematic sessions, this paragraph is the correction.

Spikes, cycles and one black hole

The two busiest days in the site's history are 25 October 2025 (2,955 page loads and 4,151 routes) and 11 December 2025 (2,701 and 3,966). Both line up with cycle activity, and there is a repeating pattern where an end-of-cycle event produces a spike as everyone simultaneously checks how long it would take them to reach the event location. A map site's traffic graph is mostly a graph of the game's calendar.

Not every spike is the game though. 15 July 2026 shows 705 page loads against a baseline of about 200, and that one was me posting a link to the solar system view of a black hole system in the Anthropic subreddit, of all places. There is also a two-day stretch in January, the 19th and 20th, showing over 2,100 routes a day on only about 800 loads, which is someone or something hammering the route planner far harder than a human browses. I have no idea who. If it was you, I would genuinely love to know what you were doing.

The hardware you bring

The audience skews exactly the way you would expect a spaceship game's audience to skew. Around 15% of sampled page loads come from 4K or better displays, roughly half are 1080p, and about a quarter of machines report 17 or more CPU cores. The largest known memory bucket is 32GB or more. Browsers are 81% Chromium, with Firefox second at around 10%. And out of about 27,000 loads that reported graphics capability, only 346 had no WebGL at all, which is a relief because without WebGL the site is basically a very slow way to look at a loading screen.

The features nobody used

Fair is fair, the archive also grades my own decisions, though the first one here is really a lesson in reading your own counters. The guided tour shows as manually started exactly once in its lifetime, which looks damning until you know it opened automatically for everyone on their first visit, so nobody ever needed the button. It is disabled now, but only because it taught jump-drive travel and jump drives no longer exist, not for lack of use. The Ctrl+K command palette is a fairer cop: 10 opens and 6 commands ever executed, and I can tell you with confidence it has exactly one user, the person who asked for it. I do not even use it myself. And the shoutbox has had 4 posts since counting started on 17 July, though in its defence it costs nothing to run, it piggybacks on the Cloudflare durable object that already powers the live events feed, so it can stay.

The number I cannot give you

The obvious question for any site is how long people actually stay, and I would love to answer it, but I found out this week that I cannot, at least not historically. The session-end tracking was bound to both the page-close event and the tab-switch event with no guard, so every time someone alt-tabbed, it recorded a session ending. The archive holds 410,298 session-end events against 95,584 page loads, a 4.29x overcount, and the mean recorded session length worked out to around 20 hours, which I briefly found flattering and then found suspicious. The bug is fixed as of yesterday, and the distortion was worst for exactly the players I most want to understand, the second-monitor people who alt-tab between the map and the game all day. Clean session data starts now. The new performance section is in the same boat, two days old with a few dozen samples, so that gets its own write-up some months from now.

Everything above is on the public stats page, updating daily, and it will keep accumulating whatever the game decides to become next cycle. If you can explain any of the oddities I cannot, especially the January route-hammerer, let me know and I will update this post.