Performance

PoE2 Oracle is a native Windows app written in Rust: one process, with no Electron and no browser inside, whose interface the graphics card draws. This page holds the measurements behind the numbers on the site: PoE2 Oracle next to other price checkers in the game and with the game in the background, its own speed and size, how they were measured and on which PC, and what the numbers don’t tell.

How it draws

• GPUI, the UI framework of the Zed code editor, draws every window of the app with the graphics card, through Direct3D 11, all of them with one device: at feature level 11.1 on the PC below, with 11.0 and 10.1 to fall back on. Its shaders are compiled when the app is built, not when it starts.
• Each window shows its frames through a DXGI flip swap chain of its own, and DirectComposition puts the windows together: that is how the overlays are see-through pixel by pixel.
• DirectWrite renders the text: its glyphs go into an atlas on the graphics card, which draws them from there, with ClearType in opaque windows and in grayscale in transparent ones. The interface’s two fonts are built into the exe.
• It is one process, and there is nothing else to install: one exe with its C runtime built in, no .NET, no Visual C++ Redistributable, no Chromium.
• Two things are drawn by others. The menu of the icon by the clock is Windows’ own. The optional window for signing in to pathofexile.com shows the site in Windows’ own browser component, Microsoft Edge WebView2, which starts only when you click Sign in; its processes end with the window.

In the game

Measured on 27 September 2026 on the PC described under How it was measured, with Path of Exile 2 in front. PoE2 Oracle, POE2 Currency Overlay, Exiled Exchange 2 and PoE Overlay II ran at once, and every minute of measuring read all four. Sidekick wasn’t part of this session.

Real play

Measured from 10:32 to 10:37 (UTC+3), with PoE2 Oracle’s build of commit 786f69b, while the PC’s owner played as he usually does: three minutes, read with no input of ours. Each minute read all four apps, so it was the same play for all of them.

CPU in each minute, in % of one core, and the median of the three:

• In real play, PoE2 Oracle used the least CPU of the four: 0.672% of a core at the median, against 3.225% for POE2 Currency Overlay, 3.700% for PoE Overlay II and 4.228% for Exiled Exchange 2. In every minute it used 4 to 7 times less CPU than any of the others: 401–412 ms a minute, against 1.7–2.9 s.
• The other three used more CPU in each minute than in the one before, by 0.39–0.54 percentage points; PoE2 Oracle stayed at 0.668–0.686%.
• After the game was restarted, PoE Overlay II had its main window open again, on the second monitor.
• Where PoE2 Oracle’s CPU went, by thread, in a minute: the XP overlay’s screen watcher 177–182 ms (64–70 wake-ups a second); its UI thread 115–126 ms (44–58), as the XP and map plates update; GPUI’s vsync thread 38–41 ms (21–24); its keyboard hook, woken by each key pressed in the game, 10–16 ms (26–36); and the NVIDIA driver’s 31 threads about 40 ms (about 155).

The other counters, from the same readings, the median of the three minutes:

• The memory column isn’t a comparison. Windows trims the memory of programs that sit idle, so over a session the numbers in Task Manager’s Memory column change as you play. Exiled Exchange 2 and POE2 Currency Overlay had been running since 08:09–08:10, and since the morning Windows had trimmed Exiled Exchange 2 from 90 to 71 MB and POE2 Currency Overlay from 107 to 41 MB. PoE2 Oracle had been running since 10:20, and by its log the owner checked prices in it six times from 10:20 to 10:34, none of them in a measured minute. The site compares memory a few minutes after each app’s launch: see With the game in the background.

Controlled test

Measured from 09:40 to 09:46 (UTC+3), with PoE2 Oracle’s build of commit 5fc9b59, the character in the hideout and the inventory open. At rest: a minute with no input, begun after 100 seconds without any. Mouse moving: a minute of slow circles over the inventory grid, with real input and no clicks.

• With the mouse moving, PoE2 Oracle used the least CPU of the four: 0.348% of a core, against 0.71% for POE2 Currency Overlay, 1.00% for Exiled Exchange 2 and 1.97% for PoE Overlay II.
• At rest, PoE2 Oracle (0.088%), Exiled Exchange 2 (0.081%) and POE2 Currency Overlay (0.099%) were about even, Exiled Exchange 2 a little the lowest. In the earlier minutes of the same session, the other three measured about the same: Exiled Exchange 2 0.075–0.090% at rest and 1.00% with the mouse moving, POE2 Currency Overlay 0.10% and 0.73%, PoE Overlay II 1.45–1.47% and 2.03%.
• The memory column isn’t a comparison: PoE2 Oracle’s build of 5fc9b59 had been started at 09:39:57 and hadn’t checked a price yet, while the other three had each done six checks earlier in the session. Price checks compares memory before and after the same six checks.
• The build of 5fc9b59 changed the XP overlay’s screen watcher: it reads a frame asynchronously, raw input wakes it, and it looks at the screen at most every 2 seconds while the mouse is at rest and every 50 ms while it moves. Against the build before, PoE2 Oracle used about 8 times less CPU at rest and 16 times less with the mouse moving.

Where PoE2 Oracle’s CPU went, by thread, in a minute:

• At rest, most of what is left is two of the NVIDIA driver’s threads: 36 ms a minute and about 150 wake-ups a second between them.
• The plates still step aside for a tooltip: with the pointer on the life flask, whose tooltip covers the XP bar, the XP plate and its ⚙ were gone within 129 ms, and back within 94 ms after the pointer moved to an empty inventory cell.

Price checks

Measured from 08:37 to 09:25 (UTC+3), with PoE2 Oracle’s build of commit 93296e4, the build before 5fc9b59. Six items were checked, each in all four apps.

Times are from the key press, read off a recording of the screen at 60 frames a second, so to a frame, ±17 ms. Window: the first frame that shows the app’s window. Result: the last change in the window’s area, after which nothing changed for 500 ms.

• PoE2 Oracle priced the currency and the gem without a trade search, and was the first of the four apps to show their result: at 231 and 217 ms.
• The four items that need a trade search took PoE2 Oracle 683–731 ms and POE2 Currency Overlay 638–792 ms; POE2 Currency Overlay was sooner with three of them. Most of the time goes to the trade site: in PoE2 Oracle’s log, about 0.61 s of a 0.64 s search is spent waiting for pathofexile.com’s two answers (see Speed and size).
• POE2 Currency Overlay’s window came first, at 46–48 ms, but with the previous result in it until the new one arrived.
• PoE Overlay II didn’t search the trade site. At 451–659 ms, its window having opened at 435–529 ms, it showed its own estimate of the value, with a confidence of “very low” or “low”, and none for the jewel.
• Exiled Exchange 2 opened its window at 93–114 ms and priced the currency and the gem at 917 and 980 ms. With its Ctrl+Y check, the one used here, it doesn’t search the trade site by itself: for the other four items it showed the filters and waited for Search.
• PoE2 Oracle’s panel fades in: its first frame came at 92–109 ms, and the filters were all drawn by about 200 ms.

The CPU a check cost is what the app’s processes used in the 13.3 seconds from the key press, less what they use at rest in that time. Memory is Task Manager’s Memory before the first check and after the last.

• Over the six checks, PoE2 Oracle’s memory grew the least: by 12.6 MB, against 22.6 MB for Exiled Exchange 2, 35.5 MB for PoE Overlay II and 64.5 MB for POE2 Currency Overlay.
• Its dedicated GPU memory grew from 84 to 130 MB: once the panel has been shown, its window stays.

PoE2 Oracle wasn’t the best at everything in the game. In the controlled test at rest, Exiled Exchange 2 used a little less CPU: 0.081% of a core against 0.088%. Per check, Exiled Exchange 2 used a little less CPU too, 154 ms against 164 at the median, with no trade search for four of the six items. POE2 Currency Overlay showed its window first, with the previous result, and finished three of the four trade searches sooner. And around the price checks PoE2 Oracle held the most dedicated GPU memory of the four, 84 MB before them and 130 MB after, with the build before 5fc9b59; in real play, with the build of 786f69b, its 68.1 MB was less than POE2 Currency Overlay’s 72.3 MB and more than Exiled Exchange 2’s and PoE Overlay II’s. See Caveats.

With the game in the background

Measured on 26 September 2026, in three sessions, on the PC described under How it was measured. Path of Exile 2 was open in the background and in view, and nobody played or checked a price. The other apps ran one at a time, each next to the game and a build of PoE2 Oracle; the build the site shows was measured on its own, in the same state. The site’s comparison shows this memory as memory after launch: each app was measured a few minutes after it started, waiting for a price check.

Each time, in UTC+3, starts a minute of measuring. The four measurements the site shows:

The other counters, from the same readings:

• Memory is in MB of 1024 × 1024 bytes, as in Task Manager. PoE Overlay II was measured 2–3 and 4–5 minutes after it started, with its main window open, and 8 minutes after, with it minimized.
• Memory in Task Manager is the column Task Manager shows by default, Memory: the private working set, the RAM that belongs to a process alone. The site compares this one. For an app of several processes it is their sum; see Caveats for why the working set isn’t summed instead.
• Commit is Task Manager’s Commit size, the process’s private bytes: memory promised to it, in RAM or in the page file, whether it has used it or not.
• Dedicated GPU memory is on the graphics card; shared GPU memory is system RAM the graphics card uses for the process.
• Idle CPU is in percent of one logical core, so 100% is one core fully busy. This PC has 24, so Task Manager’s CPU column shows a 24th of these numbers: 110% there reads about 4.6%.
• PoE2 Oracle’s row is the median of its three windows, at 19:00, 19:01 and 19:03, on its own next to the game, 3.7 to 8.2 minutes after it started. They gave 39.6, 39.5 and 39.6 MB, and 0.128, 0.122 and 0.116% of a core; its dedicated GPU memory was 71.8 MB in the first and 40.1 MB in the other two (see Caveats).
• In those minutes PoE2 Oracle showed three windows over the game: the XP plate, its ⚙ and the map plate, which shows once a map has been played since logging in. Its earlier builds were measured with the first two.

PoE2 Oracle isn’t the lowest in three columns. Idle CPU: Exiled Exchange 2 used less, 0.094% against PoE2 Oracle’s 0.122%; POE2 Currency Overlay’s 0.119% is within the spread of PoE2 Oracle’s own three minutes, 0.116–0.128%; Sidekick (0.145% and 0.398%) and PoE Overlay II used more. Dedicated GPU memory: Exiled Exchange 2 and PoE Overlay II held less, POE2 Currency Overlay and Sidekick more. Shared GPU memory: Exiled Exchange 2, which draws without the graphics card, used none. See Caveats.

The previous build. POE2 Currency Overlay was measured next to the build of 26 September, 15:32, in the same state, with two windows over the game: the XP plate and its ⚙. The median of its two windows, on its own at 17:28 and beside POE2 Currency Overlay at 17:32: 1 process and 55 threads, 37.5 MB in Task Manager, a working set of 88.6 MB, 224.1 MB of commit, 70.5 MB of dedicated and 0.8 MB of shared GPU memory, and 0.20% of a core. On its own at 17:19, while a maximized browser window covered the whole game and it hid its plates, it had 37.4 MB, a working set of 88.6 MB, 277.4 MB of commit, 70.5 and 0.2 MB of GPU memory, and 0.14% of a core; the other apps weren’t measured in that state.

The earlier build. Exiled Exchange 2, Sidekick and PoE Overlay II were measured next to an earlier build of PoE2 Oracle 0.1.0, from before its idle work was cut down. In the same state it had 1 process and 71 threads, 49.9 MB in Task Manager, a working set of 133.1 MB, 277.2 MB of commit, 159.6 MB of dedicated and 1.0 MB of shared GPU memory, and used 0.71% of a core: the median of seven windows, with 0.70–0.76% in each.

Speed and size

PoE2 Oracle’s timings come from the log of its earlier build on the same PC, over 18 price checks on 26 September 2026: 5 in the log of one run and 13 in the next.

• Of the 640 ms, about 610 go to the two requests to pathofexile.com. PoE2 Oracle’s own part is the copy (about 12 ms, most of it the game’s), reading the item (about 6 ms) and reading the answers (about 2 ms).
• Two searches waited 3.4 and 2.8 seconds before they went out, and the rows for the search and for all of it leave them out: the app holds back a third search within five seconds while GGG’s rate limit is one request away.
• Two searches found nothing and so fetched no listings: 274 and 341 ms from the item’s text.
• The log doesn’t note the time from the hotkey to the copy the app sends, nor the moment the panel shows on screen. Those, from the key press to the panel and the result on screen, were recorded on 27 September: see Price checks.
• The build of 786f69b times each check in its log by its own clock, from the key press to the moment it shows the panel and draws the listings, so the screen shows each a frame or so later. Over the six checks the PC’s owner made while playing on 27 September (see Real play), the panel came 36–66 ms after the key press and the listings 628–788 ms after it, 591–743 ms of that on the search and the fetch from pathofexile.com. The site gives these times.

Start-up. The app was ready to price 0.27 s after its process started: its first log line came 36 ms after the process, and “ready to price items” 233 ms after that. The start before, right after installing and with detailed logging on, took 0.38 s from its first log line, 0.14 s of it creating the Direct3D 11 device. By the files’ times, both starts found the trade site’s catalogs saved on disk; a very first start downloads them before it is ready, and neither log holds one.

Size.

• The installer, PoE2-Oracle-Setup-0.1.0.exe: 7,268,475 bytes (6.9 MB).
• The app it installs, poe2-oracle.exe: 28,299,776 bytes (27.0 MB). Installed with its uninstaller, the licenses and the third-party notices, it takes about 27.5 MB. The build of 18:30, whose memory and CPU with the game in the background are measured above, is an exe of 28,505,600 bytes, from an installer of 7,321,898 bytes; the build of 5fc9b59 an exe of 28,535,808 bytes; and the build of 786f69b an exe of 28,691,968 bytes, from an installer of 7,383,172 bytes.
• Its cache, %LOCALAPPDATA%\poe2-oracle\cache, holds about 5.4 MB for the current league: the trade site’s catalogs in English and Russian (3.0 MB), three hours of GGG’s record of the Currency Exchange (2.3 MB) and poe2scout’s prices (0.14 MB).

How it was measured

The PC. Windows 11 Pro 64-bit (build 26200), in Russian. Intel Core i7-13700KF (16 cores, 24 threads), 32 GB of RAM with 15.2 GB free at the start, NVIDIA GeForce RTX 4090 with driver 32.0.16.1714. Two 3840 × 2160 monitors at 200% scaling, the main one at 60 Hz.

The sessions with the game in the background, on 26 September. From 14:14 to 14:45 (UTC+3), Exiled Exchange 2, Sidekick and PoE Overlay II were measured next to the earlier build of PoE2 Oracle, which had been running since 12:07, with 13 price checks and a sign-in behind it. From 17:19 to 17:33, POE2 Currency Overlay was measured next to the build of 15:32, running since 17:03, and that build on its own at 17:19 and 17:28. From 19:00 to 19:05, the build of 18:30 was measured on its own three times; its installer had started it at 18:56.

The game on 26 September. Path of Exile 2 (Steam) ran through all three sessions on the main monitor, in view but not the window in front: in front was a browser on the second monitor. PoE2 Oracle’s plates were on screen over the game in every window but the one at 17:19: the XP plate and its ⚙ in the first two sessions, and in the third the map plate too, as a map had been played since logging in. In the first session the character stood in the hideout in away mode. The game was started anew before the second session, with the owner at the PC: from 17:18 to 17:28 a maximized browser window covered the whole game, which is when the 17:19 window was taken; in the rest of the second session, from 17:28:41, and in all of the third, the game was in view and not covered at all. The game’s log got no new line during any of the measuring minutes.

The session in the game, on 27 September. From 08:37 to 09:25 (UTC+3), PoE2 Oracle’s build of commit 93296e4, POE2 Currency Overlay, Exiled Exchange 2 and PoE Overlay II ran at once, for minutes at rest and with the mouse moving, and for the price checks. The game was in front for every minute measured, the character in the hideout with the inventory open. From 09:40 to 09:46, the minutes at rest and with the mouse moving were measured again with the build of 5fc9b59. Its installer, run silently in the user’s session with ordinary rights, closed the running PoE2 Oracle and put in the new exe in 2.0 seconds; the app was then started as usual, at 09:39:57, and its minute at rest began 2 min 53 s later. The owner took the mouse a few times during the session; those minutes were thrown out and measured again.

Real play, on 27 September. Then the build of commit 786f69b was installed silently too, its installer taking 2.0 seconds, and started at 10:20:09 (UTC+3). From 10:20 the PC’s owner played, and checked prices in PoE2 Oracle himself; from 10:32 to 10:37, three minutes were read with no input of ours, beginning at 10:32:47, 10:35:00 and 10:36:13.

The method, on 26 September.

1. The other apps were started one at a time, as a player starts them: Exiled Exchange 2, PoE Overlay II and POE2 Currency Overlay from their desktop shortcuts (PoE Overlay II’s has Overwolf’s launcher start it), Sidekick from the Start menu shortcut its installer made. Each ran in the user’s session with ordinary rights, and none took the focus.
2. Each was given 90 seconds to start, PoE Overlay II 120, since it starts in steps: Overwolf’s launcher, its client, then its browser processes. Then came a minute of measuring: three readings 20 seconds apart, with CPU counted over the whole minute. The tables give the median of the three readings. Memory moved less than 1.5% between them, but for PoE Overlay II with its window open: 620–689 MB in the first window and 649–675 MB in the second.
3. In the second and third sessions, the state of the desktop was checked from inside the user’s session before the measuring started, before every reading and at the end of each minute: whether the game ran, whether it was in front or minimized, how much of it other windows covered (which window is on top at 144 points across the game’s window), which window was in front, and which windows PoE2 Oracle showed. A minute in which the state had changed would have been measured again; from 17:28:41 on it held in every reading. The 17:19 window came before the check could see other windows over the game: a screenshot at 17:20 and a check at 17:23 found the game covered by the browser, so that window is kept as a state of its own.
4. Each process was read the same way:
   • memory in Task Manager: the performance counter Win32_PerfFormattedData_PerfProc_Process.WorkingSetPrivate;
   • working set and commit: the process’s WorkingSet64 and PrivateMemorySize64;
   • GPU memory: the GPU Process Memory counters, DedicatedUsage and SharedUsage;
   • CPU: the process’s cycle count (QueryProcessCycleTime) at the start and at the end of the minute, checked against its CPU times (GetProcessTimes), which agreed within 0.2 percentage points, 1.2 at PoE Overlay II’s 110%;
   • threads: Get-Process.
5. An app’s processes are every process whose exe is in its folder (for POE2 Currency Overlay, in its folder or its updater’s), and every process those started: this is how Sidekick’s WebView2 processes, which run from Windows’ own WebView2 folder, count. For PoE Overlay II, all of Overwolf’s processes count, as it doesn’t run without Overwolf. Each process was held open for the whole minute, so one that ended or started meanwhile counts too.
6. PoE2 Oracle was read the same way in every window, and on its own: at 14:14 the earlier build, at 17:19 and 17:28 the build of 15:32, and at 19:00, 19:01 and 19:03 the build of 18:30.

The method, on 27 September.

1. Memory and CPU were read from inside the owner’s session, by hidden tasks that open no window, over each app’s process tree: Task Manager’s Memory (WorkingSetPrivate), the working set and private bytes, the GPU Process Memory counters, and the CPU cycle count (QueryProcessCycleTime) over 60 seconds, with three readings 20 seconds apart. PoE2 Oracle’s threads were read over the same 60 seconds.
2. The price checks: six items, each in all four apps, in an order that shifted from item to item, PoE2 Oracle first. The pointer was put on the item with SetCursorPos and then a small real movement of the mouse through the Interception driver; the hotkey went through Interception too: Ctrl, the letter, the letter released after 60 ms and Ctrl after 200 ms. Before each key press, a check made sure the game was in front and the pointer in place.
3. ffmpeg recorded the screen at 60 frames a second (ddagrab with NVENC), each frame stamped with the wall-clock time (setpts=RTCTIME-C); the key press’s time was taken on the same clock, so the times are counted in frames.
4. CPU and memory per check: the process tree’s CPU cycles and private working set just before the key press and 13 seconds after.
5. Each app’s window was closed its own way: PoE2 Oracle’s and PoE Overlay II’s with Esc, which each takes while its window is open; Exiled Exchange 2’s with Esc while its window had the focus; POE2 Currency Overlay’s with a click on its close button, once a check had made sure its window was under the pointer.
6. With the mouse moving: 60 seconds of slow circles over the inventory grid, with real input and no clicks.
7. Real play: the owner played as he usually does, and nothing of ours touched the PC’s input. The three minutes were read as in step 1, all four apps in each, and every reading noted whether the game was in front: it was at every one. His play can’t be repeated, so this is one session.

Caveats

• Task Manager’s Memory and the working set. Memory, the private working set, is a process’s pages in RAM that belong to it alone: the fairest single number. The working set adds the pages it shares with others: DLLs, mapped files, and the memory Chromium’s processes share among themselves. Summed over several processes, it counts the same pages once for each, and so overstates an app of many processes; PoE2 Oracle, with one, has nothing to overstate. Windows’ own DLLs and the graphics driver, shared by every program, are in each working set and in no one’s Memory. Commit is memory promised rather than used: PoE2 Oracle’s 162.6 MB of it came with 39.6 MB in RAM.
• The game in front and in the background aren’t compared with each other. The numbers in the game were taken on 27 September with the game in front, those with the game in the background on 26 September, and none are carried over from one to the other. Memory especially: POE2 Currency Overlay had 132.9 MB with the game in the background, and 41.8 MB in the game before its first check.
• Three builds in the game. The price checks, with the memory and GPU memory around them, were measured with commit 93296e4; the controlled test with 5fc9b59, which changed the XP overlay’s screen watcher; real play with 786f69b, the commit after it, which left the screen watcher as it was.
• Real play is one session. The owner’s play can’t be repeated, and another session would give other numbers. All four apps were read in the same minutes, so they had the same play. The other three used more CPU in each minute than in the one before; three minutes don’t tell whether that would have gone on.
• Times to a frame: ±17 ms. “Window” is the first frame that changed: for PoE2 Oracle, the start of its panel’s fade-in. Exiled Exchange 2’s “result” for the items that need a search is only its filters, as it didn’t search.
• CPU per check covers 13 seconds of rest too, taken off by the measurement at rest; for PoE Overlay II that is about 195 ms, which widens its range.
• PoE2 Oracle’s memory before the checks is that of the instance started at 08:43: 35.1 MB before its first check. An instance started at 08:12 read 14.4 MB at 08:38, as Windows had trimmed its working set, and isn’t compared.
• GPU memory is Windows’ own count per process. Around the price checks, PoE2 Oracle held the most dedicated GPU memory of the four, 84 MB before them and 130 MB after, with commit 93296e4. In real play, with commit 786f69b and its panel hidden, it held 68.1 MB, less than POE2 Currency Overlay’s 72.3 MB; its own log put it at 83.7 MB while the panel was open. With the game in the background, PoE2 Oracle held more than Exiled Exchange 2 and PoE Overlay II, and less than POE2 Currency Overlay and Sidekick; the build of 15:32 held 70.5 MB, the earlier build 159.6 MB, the most of all. The build of 18:30 read 71.8 MB in the first of its three minutes: between 19:01:03 and 19:01:33, about 4.5 minutes after it started, its GPU memory dropped to 40.1 MB and stayed there. Its log noted nothing at the time, and why it dropped wasn’t looked into; the table gives the median of the three minutes. The memory wasn’t broken down by measurement; by GPUI’s code, each window holds GPU memory in proportion to its size in pixels, so the number grows with the screen, and this PC runs 4K at 200%. Exiled Exchange 2 holds none: it turns off hardware acceleration (app.disableHardwareAcceleration()) and draws its overlay with the CPU. Overwolf’s overlay in the game runs inside the game’s own process and isn’t counted for PoE Overlay II.
• Idle CPU with the game in the background. PoE2 Oracle used 0.122% of one core, the median of its three minutes, which ranged from 0.116 to 0.128%: more than Exiled Exchange 2 (0.094%), about as much as POE2 Currency Overlay, whose 0.119% is within that spread, and less than Sidekick (0.145% on its setup page, 0.398% ready to work) and PoE Overlay II. On this 24-thread CPU that is 0.005% of the whole, which Task Manager’s CPU column shows as 0. Its plates stay on screen over the game, while none of the other apps showed anything over it at the time. Split by thread in two of its minutes, about a third of it went to the graphics driver’s threads, a third to Windows’ thread pool, where GPUI runs its timers and background work, a quarter to its UI thread and a tenth to GPUI’s vsync thread. The build of 15:32 used 0.20%, and 0.14% with the game covered and its plates hidden; the earlier build used 0.70–0.76%.
• What was measured with the game in the background: idle. Nobody opened an overlay or checked a price in those sessions; the price checks were measured in the game. POE2 Currency Overlay had its Price check tab open, as its settings kept it; on its Currency tab it asks for live rates every 6 seconds, which wasn’t measured.
• The states. Exiled Exchange 2 opened no window: it waits in the tray and shows its overlay on Shift+Space. Sidekick ran on the settings of an earlier install, whose league had ended. At its first start it opened its setup window, asking for a league, and waited there: the “setup page” row, which anyone whose saved league is last season’s will see. With the current league in its settings, its next start went through its whole start-up in 12 seconds with no window (quiet start was on in those settings): the “ready to work” row, the one to compare. PoE Overlay II, started from its shortcut, opened its main window by itself on the second monitor, without taking the focus: the character, a session summary, the stash, and beside them two animated banner ads, whose goods changed between screenshots 45 seconds apart, and two offers to go Premium to remove them. While that window is open, Overwolf’s browser keeps repainting it: its GPU process used 92–99% of a core and the page’s process another 10–13%. The window was then minimized with its own Minimize command and measured again. Closing it without quitting the app wasn’t tried. POE2 Currency Overlay showed a splash at start, gone by the 30th second, and no first-run window, as its tutorial had been done before. It keeps one window over the game, always on top and letting clicks through, at zero opacity: it can’t be seen until its hotkey makes it opaque. None of the others showed ads; POE2 Currency Overlay’s window has a link to a Ko-fi donation page.
• PoE Overlay II’s Session Recap, in the game. When the game lost the focus, PoE Overlay II opened its Session Recap window in the middle of the screen, 3158 × 1628, with banner ads and an offer of Premium: twice in 30 minutes. It covered the inventory, so one check of PoE2 Oracle and one of POE2 Currency Overlay went for nothing: the item was under that window. Both times the window was closed as its close button closes it.
• Three builds with the game in the background. The site sets the build of 18:30 against apps measured next to earlier builds: POE2 Currency Overlay next to the build of 15:32, the other three next to the earlier build, earlier the same day, with the game started anew between the first and second sessions. The state was the same: the game in view and not in front, PoE2 Oracle’s plates on screen, every app idle.
• The game’s state and the plates matter. When another window covers the game, PoE2 Oracle hides its plates and uses less CPU: the build of 15:32 used 0.14% so, against 0.20% with the game in view. The map plate shows only once a map has been played since logging in, and the build of 18:30 was measured with it: three windows over the game, where the builds before had two. If anything, that counts against the build of 18:30. A minimized game is another state again, which wasn’t measured.
• The first minutes against hours. With the game in the background, the other apps were measured 1.5 to 8.5 minutes after they started; PoE2 Oracle’s earlier build after more than two hours, the build of 15:32 after 16 to 30 minutes and the build of 18:30 after 3.7 to 8.2 minutes. Apps on Electron, CEF and .NET usually grow as their caches, heaps and history of checks fill, so for them these numbers are likely a lower bound. PoE Overlay II’s were the same at 2–3 and at 4–5 minutes: not a start-up spike.
• Exiled Exchange 2 0.16.7 is a build of the fork mttzzz/Exiled-Exchange-2: Kvan7’s Exiled Exchange 2 0.16.3 with an update channel and Russian translations added. The engine is the same, so its memory and CPU shouldn’t differ.
• POE2 Currency Overlay 3.0.7, by POE2 VibeTools (POE2-VibeTools/poe2-currency-overlay), was its latest release, and the installed app’s code matches that release’s source.
• PoE Overlay II Standalone, the version without Overwolf, wasn’t installed on the PC and wasn’t measured.
• One PC, one run of each app. Within a measurement the numbers held steady, but another day, league or cache size would give somewhat different ones.