Reading a crash report
17 min readASA DevKit as of November 2026 (UE 5.8.1); live game and dedicated server builds of the
When the DevKit dies, the crash window offers little more than "send report". The real information is in three places on your disk: a one-file summary, a folder Unreal writes for every crash, and the log of the session that crashed. Together they nearly always tell you what kind of crash it was and what the editor was doing. That's enough to fix most of them.
I've collected 519 crash reports from my own DevKit over 18 months. The counts below are real, and the advice is what worked on them.
If the editor didn't crash but something misbehaves, start with Reading ARK's logs. If a cook or upload failed, see When a cook or upload fails.
The short version
- Open the newest
.crashstack. It's named<MM.DD-HH.MM.SS>.crashstack, in<DevKit>\Projects\ShooterGame\Saved\Logs\. Notepad is fine. - Read the error line near the top. Access violation, assertion, GPU, out of memory: it puts the crash in one of the five families.
- Find the first real frame. Read the callstack from the top and skip the reporting frames. The first real frame tells you what the engine was busy with: loading, compiling a Blueprint, drawing or game code.
- Read the 30 to 50 log lines before
=== Critical error: ===. After a restart, that's in the newestShooterGame-backup-….log. The asset being loaded, the Blueprint being compiled or the PIE command shows up there. - Do what the family's section says. If the same asset crashes the editor every time you open it, it's almost always the asset, not your PC.
- Crashed in the live game or on a server? The stack has no names. Reproduce it in PIE or on a local dedicated server.
Where the crash files are
| File | Where | What it is |
|---|---|---|
<MM.DD-HH.MM.SS>.crashstack |
<DevKit>\Projects\ShooterGame\Saved\Logs\ |
ARK's own one-file summary, written on every crash. It's UTF-16 text, named with your local date and time. |
crashcallstack.txt |
same | The callstack of the last fatal error only, overwritten each time. Not every crash writes it; the .crashstack files are the complete history. |
ShooterGame-backup-<date-time>.log |
same | The log of the session that crashed, renamed when you restart the editor. Its name is in UTC. It ends with the === Critical error: === block. |
UECC-Windows-<id>_0000\ |
<DevKit>\Projects\ShooterGame\Saved\Crashes\ |
Unreal's crash folder, one per crash. |
.crashstack, crashcallstack.txt, UECC-… |
<Game>\ShooterGame\Saved\Logs\ and \Crashes\ |
The same for the live game, with no symbols (below) |
.crashstack |
<Server>\ShooterGame\Saved\Logs\ |
The same for a dedicated server. |
Inside a UECC-… crash folder:
| File | What's in it |
|---|---|
CrashContext.runtime-xml |
The error again (<ErrorMessage>), the <CrashType> (Crash, Assert, Ensure, GPUCrash), <EngineVersion>, <SecondsSinceStart>, memory figures and the callstack. <UserActivityHint> says which editor panel had focus. |
ShooterGame.log |
A copy of the session log up to the moment of the crash. 507 of my 516 folders have one. Its last lines show what was happening. |
UEMinidump.dmp |
A memory dump for a debugger. Without the matching symbols it's of little use to modders. |
CrashReportClient.ini |
Crash reporter settings. |
Breadcrumbs_RHIThread_0.txt, *.nv-gpudmp |
Only after GPU crashes: the last rendering steps, and an NVIDIA GPU dump. |
Slate.txt |
Only after some UI crashes. |
The crash folder never empties itself. Mine holds 516 folders going back to 2025, with about 1 MB of log each. The crash reporter's settings say to purge after two days, but it doesn't. It's safe to clear with the editor closed. Keep the ones you're still investigating.
Anatomy of a .crashstack
This one is built from two of my crash files, shortened, with names replaced. The top is from a load crash and the footer from a PIE crash.
Assertion failed: [File:…\LinkerLoad.cpp] [Line: 5802] ← the error line
<DevKit>\…\Mods\<YourMod>\Content\<Asset>.uasset: … Serial size mismatch: Got 87, Expected 83 ← the detail
CL: 831037 ← the game build (ASA 812.101), not the engine
0x00007ff95b62408a KERNELBASE.dll!UnknownFunction [] ← reporting frames: skip
0x00007ff8299783f4 ShooterGameEditor-Core.dll!ReportAssert() [...\WindowsPlatformCrashContext.cpp:2137]
0x00007ff85f0cd2fd ShooterGameEditor-CoreUObject.dll!FLinkerLoad::Preload() [...\LinkerLoad.cpp:5802] ← first real frame
0x00007ff85f44f838 ShooterGameEditor-CoreUObject.dll!FAsyncPackage2::PreloadLinkerLoadExports() [...]
...
==========
ASA Version 812.101 (CL: 831037)
Memory Used/Avail: 20696MB/13944MB
Map: <TestMap>
Game Mode: Dedicated Server
TimeSeconds: 200000650.5
Players Connected: 1
FPS: 60.00
==========
<PC name>-PIE-1 PlayerControllerBlueprint_C_0
1. Fatal error! comes first for exceptions and fatal errors. For a failed check, the assertion text comes first
instead, as in this example.
2. The error line has one of five shapes:
| Error line | What it means |
|---|---|
Unhandled Exception: EXCEPTION_ACCESS_VIOLATION reading address 0x… (or writing) |
Code used memory it had no right to: a None, destroyed or garbage pointer. An address near zero (0x0000000000000000, 0x…0010, 0x…00d8) means something was None, and the code read a field at a small offset from it. |
Unhandled Exception: EXCEPTION_STACK_OVERFLOW |
Endless recursion: something calls itself forever. |
Unhandled Exception: 0x… with no name |
In my files, always a GPU crash. The top real frame is ReportGPUCrash. |
Assertion failed: <condition> [File:…\<Something>.cpp] [Line: N] |
An engine sanity check, a check(), failed. The condition and the source file name are your clues. The next line sometimes names the asset. |
Fatal error: [File:…] [Line: N] followed by a message |
A deliberate stop: out of memory, a checked cast to the wrong class, a map that can't be unloaded. |
3. CL: is the changelist, Wildcard's build number. Compare it with the build you think you're running. Say a
server crashes right after an update, and its file shows an old CL: the server wasn't updated.
4. The callstack lists the newest call first. Each frame reads 0x<address> <module>!<Function>() [<source file>:<line>].
The module is the DLL. ShooterGameEditor-CoreUObject.dll is the engine's object system, and
ShooterGameEditor-ShooterGame.dll is ARK's game code. The function is Class::Method. The source path in
brackets is on Wildcard's build machine, not your PC.
5. The footer sits between ========== lines. DevKit builds since about September 2026 add it to crashes during
play. It lists the game version, memory used and free, map, game mode, game time, creature and player counts, the local
player's position, structure count and frame rate. A PIE run shows Dedicated Server as its game mode. Low free memory
or a single-digit FPS is a clue on its own. Game time starts at 200,000,000 s in PIE, and at 100,000,000 s in a live
single-player session.
Finding the first real frame
What you see: the top few frames are the crash reporter doing its job.
What to do: skip every frame named KERNELBASE, ReportAssert, FWindowsErrorOutputDevice::Serialize,
FSentryErrorOutputDevice::Serialize, FOutputDevice::LogfImpl, AssertFailedImplV, FDebug::AssertFailed,
BasicFatalLog or CastLogError.
What it means: the first frame after those names the subsystem.
| First real frame starts with | The editor was… |
|---|---|
FLinkerLoad::, FAsyncPackage2::, LoadPackageInternal |
Loading a package (an asset) from disk. |
FKismetCompilerContext::, FBlueprintCompilationManagerImpl::, FBlueprintEditorUtils:: |
Compiling a Blueprint. |
FBlueprintCompileReinstancer::ReplaceInstancesOfClass |
Compiling a Blueprint while instances of it exist, often with PIE running. |
UEditorEngine::Map_Load |
Opening or switching a level. |
FStaticMeshBuilder::, FMeshDescription::, UStaticMesh::CacheDerivedData |
Building a mesh. |
ReportGPUCrash, D3D12RHI::, FD3D12Viewport::Present |
Drawing: the GPU or its driver. |
OnOutOfMemory |
Allocating memory, and failing. |
APrimalStructure::, APrimalDinoCharacter::, AShooterCharacter:: |
Running ARK game code, in PIE. |
USceneComponent::, UWorld::RemoveActor, FWeakObjectPtr:: |
Moving or destroying actors. |
python311.dll |
Running an editor Python script. |
Then read down a few frames to see who called it. Say a FLinkerLoad crash was called from UEditorEngine::Map_Load:
it was loading something in your level. Called from the Blueprint editor, it was loading something that Blueprint
references.
The session log: what happened just before
What it means: the .crashstack tells you what failed. The log tells you what the editor was doing.
Where to look: open the crashed session's log and jump to the end. That's the newest ShooterGame-backup-….log,
or the ShooterGame.log copy in the crash folder.
[…][319]LogWindows: Error: === Critical error: ===
[…][319]LogWindows: Error: Fatal error!
[…][319]LogWindows: Error: Unhandled Exception: EXCEPTION_ACCESS_VIOLATION writing address 0x0000000000000000
[…][319]LogWindows: Error: CL: 831037
[…][319]LogWindows: Error: [Callstack] 0x… <module>!<Function>() […]
What to do: scroll up 30 to 50 lines from that block. Look for any of these:
- the last asset being opened;
- a
LogBlueprintcompile line; - a
LogSerializationorLogLinkererror naming a package; - the PIE start, a
LogPlayLevelline.
That's why helpers on the Discord ask for this log before the .crashstack. The crash folder's copy is taken at the
moment of the crash, so it usually stops just before the Critical error block.
The five crash families
These counts come from my 519 DevKit .crashstack files. Many came from editor scripts I run, so load and compile
crashes show up more than on a typical modder's PC. But the kinds are the ones everyone hits.
| Family | Crashes | Share |
|---|---|---|
| 1. A None or destroyed object | 147 access violations, plus 8 bad casts and 4 stack overflows | 31% |
| 2. A corrupted asset | 143 (86 + 39 + 18, three assertion types) | 28% |
| 3. The GPU | 52, plus 4 out of video memory | 11% |
| 4. The Blueprint compiler | 43 (32 + 11) | 8% |
| 5. Out of memory | 11 | 2% |
| Others | 107: array index out of bounds (21), map unload leaks, UI and plugin asserts | 21% |
1. A None or destroyed object
What you see: EXCEPTION_ACCESS_VIOLATION reading address 0x… or writing, often with an address near zero. It
has two cousins. One is Fatal error: [File:…\Casts.cpp] [Line: 10] with Cast of <X> to <Y> failed, a checked cast
to the wrong class. The other is EXCEPTION_STACK_OVERFLOW.
Where to look: the first real frame says where. In my files I see UWorld::RemoveActor, FWeakObjectPtr,
UObject::Rename, Niagara emitter loading, APrimalStructure::AllowSnappingWith and Python.
What it means: C++ code, the engine's or ARK's, followed a pointer to an object that was None, destroyed or never loaded. In Blueprints the same mistake gives a harmless "Accessed None" in the log. In native code it kills the process.
What to do:
- Find what was being touched. Read the first real frame and the log's last lines.
- Game code frames during PIE? Then it's likely your Blueprint's doing. Look for a reference that can be None at
that moment: a component destroyed earlier, an actor in the other world, a soft reference that never loaded. In
PIE,
UEDPIE_0is the server world andUEDPIE_1the client. Guard the reference with Is Valid. - Crashes when you open a Blueprint? Open the assets it references first, one at a time, then the Blueprint. Modders report this gets a crashing Blueprint open again. Moving hard class references in a data table to soft references stopped it for good.
- Stack overflow? Something calls itself without end. Look for a function or event that triggers itself, or two actors that keep notifying each other. A known vanilla one: opening the inventory while holding a weapon in PIE (lessons §3).
- Avoid the known editor triggers. Don't compile a Blueprint while PIE runs, or save an asset whose actor is in a running PIE level. Don't switch levels with Blueprint editors open. Don't stop PIE while a widget is open.
2. A corrupted asset
What you see: one of three assertions, all while loading. The first one names the asset:
Assertion failed: [File:…\LinkerLoad.cpp] [Line: 5802]
<DevKit>\Projects\ShooterGame\Mods\<YourMod>\Content\<Asset>.uasset: <Class> /<YourMod>/<Asset>.<Asset>: Serial size mismatch: Got 87, Expected 83
The second is Assertion failed: InPos <= Size [File:…\FileManagerGeneric.cpp]: reading past the end of a file. The
third is Assertion failed: ReaderPos + Num <= ReaderSize [File:…\BufferReader.h] under MeshDescription frames, from
bad mesh source data.
Where to look: just before the crash, the log shows LogSerialization: Error: Invalid boolean encountered while reading archive /<YourMod>/<Asset> - stream is most likely corrupted.
What it means: the file on disk doesn't match what the engine expects. A crash or power cut half-wrote it during a save, or it was damaged, or a different engine build saved it. The good news is that the line names the asset. I've seen it on levels and level Blueprints, geometry collections, a font, a mesh's physics setup, a control rig, an enum and a World Partition cell.
What to do:
- Note the asset path from the second line or the
LogSerializationline. - Put back a good copy. With the editor closed, look in
<DevKit>\Projects\ShooterGame\Saved\Autosaves\, your own backup or version control. Keep backups **outside** your mod's Content folder. A strayX.backup.uassetinside Content is fatal at PIE start (lessons §1). - No good copy? Move the bad
.uassetout of the Content folder, start the editor and rebuild that asset. Other assets that referenced it will need their references fixed. - Everything you open from one mod crashes, but other mods are fine? Verify the DevKit in the Epic launcher. Then purge the derived data cache, the editor's store of processed assets. Modders report that verifying gets the kit opening again. It doesn't repair a broken asset, though.
3. The GPU
What you see: Unhandled Exception: 0x… with the first real frame ReportGPUCrash. Or Fatal error: [File:…\D3D12Util.cpp] with DXGI_ERROR_DEVICE_REMOVED, DXGI_ERROR_DEVICE_HUNG or
DXGI_ERROR_DRIVER_INTERNAL_ERROR.
Where to look: the log shows LogD3D12RHI: Error: GPU crash detected: and video memory lines. The crash folder has
Breadcrumbs_RHIThread_0.txt, and maybe an .nv-gpudmp.
What it means: the graphics driver gave up. It's your PC, not your mod: the driver, video memory, an overclock or an overlay. It happens at launch, a few seconds after opening a map, or after leaving the editor idle.
What to do: most of these fixes come from modders on the Discord.
- Lower the editor's scalability: Settings > Engine Scalability Settings > Low. Turn off Realtime in viewports you aren't using (tutorials §8).
- Close overlays that hook every window, such as MSI Afterburner's stats overlay. That alone fixed a launch crash.
- Update the GPU driver, or roll back one version if a new one started it.
- As a last resort, raise Windows' GPU timeout. One modder went from a crash every ten minutes to none. They raised the GPU timeout (TDR) in the registry and set the NVIDIA power mode to maximum performance. It's a single report, and it's a system setting.
- Don't try forcing DirectX 11. The live game refuses to start with
Unable to launch with RHI 'D3D11' since the project is not configured to support it.
4. The Blueprint compiler
What you see: Assertion failed: Result [File:…\EdGraphNode.h] with the first real frame
FKismetCompilerContext::ReplaceConvertibleDelegates. Or Assertion failed: EntryNodes.Num() == 1 [File:…\BlueprintEditorUtils.cpp] under FFunctionFromNodeHelper::FunctionFromNode.
What it means: one Blueprint has a damaged graph. The first is usually a broken node in delegate or event wiring, like an event dispatcher bind or a Create Event node. The second is a function graph with zero or two entry nodes. The crash comes when that Blueprint compiles or saves. That includes PIE start, the cook, and sometimes opening it.
What to do:
- Find which Blueprint. Check the log's last
LogBlueprintlines before the crash, or the asset you just edited. - If it opens, rebuild the nodes. Delete the nodes you last touched and add them again. For the entry-node assert, find the damaged function and rebuild it. It's often one you just collapsed or pasted.
- If it won't open without crashing, restore it from
Saved\Autosavesor a backup, as in family 2. - An "internal compiler error" right after connecting a pin? Modders report it goes away the same way. Remove and re-add the nodes, or restart the kit.
5. Out of memory
What you see: Fatal error: [File:…\GenericPlatformMemory.cpp] [Line: 273] with OnOutOfMemory. The log says
Ran out of memory allocating N bytes … The paging file is too small for this operation to complete. Out of video
memory shows as D3D12RHI TerminateOnOutOfMemory instead.
What it means: the editor ran out of RAM plus page file. The DevKit's memory creeps up over a session. The first Blueprint editor of a session alone can load tens of gigabytes (lessons §3).
What to do: save often, and restart the editor before it nears the limit. Close other programs. Give Windows a
larger page file on a fast drive. For RAM, my beginner guide video's advice holds: 64 GB is comfortable, 32 workable,
16 painful. The footer's Memory Used/Avail line shows how close you were.
The others
Array index out of bounds(21 crashes): engine code indexing past an array, mostly during load or in editor panels. Note what you clicked. It's usually an editor bug to work around, not your content.Fatal error: [File:…\EditorServer.cpp]inUEditorEngine::Map_Load, with "World Memory Leaks" incrashcallstack.txt. You switched levels while something still held the old one, like an open Blueprint editor or a running PIE. Close Blueprint editors before switching maps.FindFunctionChecked,BlueprintGeneratedClass.cpp: a class is missing a function it should have, or its generated class is broken. Recompile and resave the Blueprint named in the log.- Plugin asserts, such as the AMD frame interpolation one or Houdini Engine: a plugin bug. For Houdini, one modder fixed it by pointing the plugin at the right Houdini version in its settings.
Crashes in the live game or on a server
What you see: a .crashstack where every frame is ArkAscended.exe!UnknownFunction, or
ArkAscendedServer.exe!UnknownFunction on a server.
Fatal error!
Unhandled Exception: 0x00008000
CL: 835078
0x00007ffb7a3b569a KERNELBASE.dll!UnknownFunction []
0x00007ff6d9b015a0 ArkAscended.exe!UnknownFunction []
...
==========
ASA Version 95.11 (CL: 835078)
Game Run ID <id>
Sentry URL: <a link only Wildcard can open>
Memory Used/Avail: 4682MB/36573MB
Map: MainMenu
Game Mode: Single Player
FPS: 2.02
==========
What it means: the shipping game and server come without symbol files. Those are the .pdb files that turn
addresses into function names. The community dates this to about 2025 for the client and mid-2026 for the server. All
80 client and 5 server crash files on my PC look like this. So the stack can't tell you whether your mod was involved.
Where to look: these parts still help.
- The error line: An access violation with a low address is still a None. A readable fatal message sometimes says
it all, like the D3D11 one above. On a server,
Requested mods failed to load on server.means a mod ID in-mods=couldn't be loaded. - The CL and the footer: build, map, game mode, memory, FPS. Use the time to find the matching lines in the
server's
ShooterGame.log. - The log, if there is one. A server always writes one; a client only with
-log(Reading ARK's logs). The last lines before the crash, plus your own[<YourMod>]messages, narrow it down.
What to do: after an ARK update, a mod that suddenly crashes servers often just needs a fresh cook against the new game. Modders report that resubmitting to CurseForge fixed it; see When a cook or upload fails. Otherwise, reproduce the crash where you can watch it.
Reproduce it on a local dedicated server
I go in this order:
Try PIE first. If the crash also happens in PIE, the DevKit gives you a full stack with names. Done.
Then try a local dedicated server with a local cook. PIE runs server and client in one process and hides a lot (Works in PIE, not on a server). Many live-only crashes need a real server.
Show me how
Install the ASA dedicated server on your own PC, from Steam. Then:
- Run a local cook for Windows Client and Windows Server. Use the cooking-pot button in the editor, or
<DevKit>\ModTools\Scripts\CookMod.bat. The zips land in<DevKit>\ModTools\Output\<YourMod>\, as<YourMod>-Windows.zipand a server one. - The mod must already exist on CurseForge. Start the server and your game once with your mod's project ID, so
they download it. Then shut them down. A mod that's only "ready for review" loads with the ID plus
-dev. - Replace the downloaded mod's files with the ones from your zips, except the
.upluginfile. The local cook window's own instructions say so. Modders report that overwriting the.uplugin, or too much else, makes the server download the mod again over your files. Several just replace the.pakfiles. - Start the server with
-log, join it, and reproduce the crash. ItsShooterGame.lognow shows your messages and the order of events up to the crash.
- Run a local cook for Windows Client and Windows Server. Use the cooking-pot button in the editor, or
Narrow it down. There are still no names, so let your own messages do the work. Add Print Message in Shipping Build lines before and after the suspect steps (Reading ARK's logs). Switch off half your features at a time until the crash stops. The last message before the crash marks the spot.
The community has found these behind live-only crashes:
- sparse overrides set incorrectly, causing generic null-pointer crashes on servers;
- a widget mod cooked before a game update, crashing during garbage collection;
- remapped supply crates crashing the NPC zone manager at start-up.
Before you ask for help
Send three things:
- the
.crashstack, or its top 30 lines and the footer; - the 50 log lines before the Critical error block;
- one sentence on what you were doing, like "opening
<YourBlueprint>" or "pressing Play with two clients".
Redact them first. The crash footer contains your PC's name, and server logs contain the admin password. A helper can't do much with the minidump or a screenshot of the crash window.
Watch
No video in my library covers crash reports. This one covers the hardware side: RAM, disk, and launching with
RunEditorSDK.bat so the launcher doesn't think a crashed kit is still running.
- How to Make an ARK ASA Mod (Complete Beginner Guide) (my video, 0:00: hardware advice; 3:28: launching the kit, memory creep and saving often; 8:10: telling a busy editor from a crashed one)
Go deeper
- Reading ARK's logs: every log, the lines that matter and the noise to skip
- When a cook or upload fails: when the crash is the cooker's
- Works in PIE, not on a server
- Modding lessons §3: known PIE crashes
- Testing your mod: PIE, then a local dedicated server, where live-only crashes show up
- Symptom index: start from what you see when you don't know which page covers it
Sources
- My own crash files, read-only on 2026-11-05: 519 DevKit
.crashstackfiles (May 2025 to November 2026), 516 crash folders with 507 log copies, and 80 game-client and 5 dedicated-server crash files. The crash-family counts, the first-frame table and every other count on this page come from those files. Examples are rebuilt with placeholders. - Checked in those files on 2026-11-05: the crash-folder contents and its purge setting; the footer (in DevKit crash
files since about September 2026) and its game-time starting values; the GPU and out-of-memory signatures; the
assets behind corrupted-asset crashes; the D3D11 message (a 2026 game crash file); and the
Map_Loadleak crash, which matches lessons §3. - Modding lessons §1 and §3: backups inside Content, known PIE crashes, memory creep.
- From the community, not yet confirmed by me:
- how to read the first frames (2024 to 2026), and asking for the log before the
.crashstack(May 2026); - opening referenced Blueprints first, and soft references in a data table (May 2026);
- stopping PIE with a widget open (2025);
- verifying the kit and purging the derived data cache (2025 and 2026);
- the GPU fixes (2023 to 2025), including the TDR registry change (April 2024);
- the internal compiler error fix and the Houdini fix (2025);
- when live builds lost their symbols (2025 for the client, July 2026 for the server);
Requested mods failed to load on server(2023), and resubmitting after an update (2025 to 2026);- testing on a local dedicated server,
-devand keeping the.uplugin(2023 to July 2026); - the causes of live-only crashes (2023 to 2025).
- how to read the first frames (2024 to 2026), and asking for the log before the
- Community knowledge from the ARK Modding Community Discord (discussions from October 2023 to July 2026, mostly 2024
to 2026), found with Overseer. Threads used (they open for server members):
- Where the crash files are; ShooterGame.log versus the crashstack (May 2026)
- Open referenced Blueprints first; soft references in a data table (May 2026)
- Verify, DDC purge, the log is written even without a popup (June 2025)
- No symbols in live builds; reproduce on a local dedicated server (June 2025)
- Requested mods failed to load on server (October 2023)
- GPU crashes: overlays and drivers (June 2025)
- Local cook to a test server and client (November 2023)
- Local cook on a dedicated server:
-devand keeping your files (July 2026)
Where this page's information comes from
Applies to: ASA DevKit as of November 2026 (UE 5.8.1); live game and dedicated server builds of the
my own crash reports and crash folders; the modding Discord (see Sources)