Your mate says they only changed one tiny thing. You load the shared world and somehow your base is three hours behind, their inventory is from Tuesday, and the save folder now contains `world_final_FINAL_reallyfinal`. That is save desync: two or more versions of the same game world fighting for the right to exist.
Learning how to prevent save desync is less about becoming the group’s unpaid IT department and more about giving one world one clear history. Do that, and you stop gambling a 100-hour run every time someone launches the game from a different PC.
What save desync actually is
A desync happens when players, devices, cloud services, or mods disagree about which save file is current. Sometimes the game notices and throws an error. More often, it does something much ruder: it loads an older state without making a fuss.
This is especially common in games where the host owns the world file. A co-op session may feel shared while everyone is online, but the actual save can still live on one person’s machine. If another player gets a copy, plays it separately, then puts it back in the shared folder, you now have two legitimate-looking timelines. Neither knows the other exists.
Normal consumer cloud sync can make this worse. It is designed to move files around, not understand that a Valheim world database, mod config, and player data may need to change together. It sees files. Your game sees a fragile little ecosystem held together by hope and autosaves.
The rule that prevents most disasters: one writer at a time
The cleanest way to avoid save desync is simple: only one person should be allowed to edit a shared world at any moment. That person is the active host, and everyone else treats the save as read-only until the handoff is complete.
That sounds obvious, right up until Dave boots an old local copy because the group chat said “anyone on tonight?” Now Dave is building a second universe from last week’s save. Both worlds can be fun, but they cannot be merged by wishful thinking.
For casual groups, agree on a tiny routine. The outgoing host exits the game properly, waits for the save to finish writing, then confirms the latest version has uploaded. The incoming host checks they have that latest version before launching. Nobody opens the world while the handoff is happening.
It takes less time than debating whether the missing iron was “definitely in the chest”.
Do not pass worlds around like USB sticks
Copying a save into Discord, a shared drive, or a zip file works in an emergency. It is also how folders called `newest_save_2_USE_THIS_ONE` are born. Manual copies are easy to forget, hard to verify, and dangerously convincing when they are old.
If you must do it manually, use one named source folder and one designated host. Add the date and time to exported copies, keep the original untouched, and never overwrite the only known-good version. This is not elegant. It is disaster management with a high-vis jacket.
Make automatic saves boring again
The best backup system is the one that runs while you are busy losing a boss fight. Automatic versioned backups are your safety net because they preserve more than the latest file. They preserve the moments before somebody installed a questionable mod, force-quit during a save, or accidentally hosted an outdated world.
A proper history lets you compare the timeline in plain English: here is the save from before last night’s session; here is the version from ten minutes before the crash. Instead of asking, “Which copy is right?”, you can restore the last state you know was healthy.
Checkpoint64 is built for exactly this kind of save-folder chaos. It watches supported game saves, checks for changes every 30 seconds, uploads changed files rather than the whole universe, and keeps each version available for a one-click restore. Its co-op handoff uses an enforced lock, so the active player has the write access and everyone else is stopped from accidentally creating a rival timeline. Free plan, actually free. More space can be a one-time purchase. No subscription toll booth waiting at the end of the dungeon.
The trade-off is worth saying out loud: automatic backup is not a magic shield against bad decisions made inside the game. If you save after deleting the castle, that version faithfully contains a deleted castle. Version history is what gives you the exit ramp.
How to prevent save desync before a co-op session
Start each session with a quick pre-flight check. It should be boring enough that everyone actually does it.
First, make sure the intended host has the latest complete save, not merely a folder with a familiar name. Second, confirm the previous host has fully closed the game. Many games write saves on exit, so grabbing files before that point can capture half-finished data. Third, check that all players are using the same game build, mod list, and mod versions. Then launch from the agreed current copy and let the host announce when the world is live.
Mods deserve special suspicion. A missing dependency can look like save corruption. A newer mod version can rewrite data that an older client cannot understand. Before a big modpack update, create a labelled restore point and have the group update together. If someone wants to experiment with a new mod, clone the world first. Your main campaign is not a test server, even if Kevin says it “should be fine”.
Watch out for sync collisions across devices
Playing the same single-player save on a desktop and a laptop creates the same problem in miniature. If both devices are offline, each can make progress independently. When they reconnect, a cloud provider may choose one copy, create a conflict file, or quietly let the newest timestamp win. None of those outcomes knows which version contains your actual progress.
Before switching machines, close the game and wait for the latest save to finish uploading. On the second machine, wait until that version has arrived before launching. Do not leave the game open on both devices just because one is minimised. A minimised game is still an active threat to your timeline.
This matters more with emulators, where save states and battery saves can update at different times. Back up both. Restoring only the `.sav` but not the matching state file can produce a very strange approximation of time travel.
Treat crashes and corruption as forensic events
When a game crashes during a save, resist the urge to relaunch immediately. Relaunching can overwrite evidence with a fresh autosave, which is the gaming equivalent of cleaning a crime scene because it looked untidy.
Make a copy of the current save folder first. Then identify the most recent known-good version and restore it to a separate test location if possible. Load it offline or without experimental mods. Once you know it works, replace the damaged active copy deliberately, not by dragging random folders together until the game stops complaining.
If the world is shared, tell the group what happened before anyone launches their own copy. The fastest route to a recoverable problem is one person making decisions. The fastest route to a branching nightmare is four people “having a go”.
Give your crew a handoff ritual
Co-op saves fail socially before they fail technically. Somebody forgets they were hosting. Somebody assumes the cloud is instant. Somebody starts a solo session in the shared world because they only needed wood. Suddenly the logbook is a conspiracy board.
Set a simple rule: claim the world, play it, close it, release it. Record who last hosted and when. If you are using a save-sharing tool with locks, respect the lock rather than working around it. A lock is not the software being annoying. It is the software preventing two people from writing different endings to the same story.
For creators and modding groups, keep public or viewer-accessible copies read-only. Let people download and explore, but do not let a random fork overwrite the canonical world. The version you film, speedrun, or maintain for the community deserves a protected shelf, not an open tab in somebody’s file browser.
Your saves are not disposable settings files. They are the map, the memories, the questionable architecture, and 200 hours you are not getting back. Put one player in charge at a time, keep a real version history, and make restoring yesterday’s world easier than arguing about who broke it.