Fix CommonSense meditation desync: host-authoritative sync

[sviyh]

Summary

• Fixes desync caused by CommonSense's JobDriver_MeditationTick_CommonSensePatch postfix evaluating joy.CurLevel >= 0.98f differently on host vs client due to floating point drift
• Instead of disabling the feature, replaces it with a host-authoritative version: host evaluates the joy/psyfocus thresholds and syncs the EndJobWith decision to all clients via RegisterSyncMethod
• The original postfix is canceled in MP via a Harmony prefix, and a new postfix on JobDriver_Meditate.MeditationTick runs only on the host

Root cause

Observed in desync log (map 19): CommonSense ended a reign/meditation job on the host (joy hit 0.98), triggering PickUpAndHaul's CheckIfPawnShouldUnloadInventory which created a new job and consumed extra RNG state. The client's joy was slightly below threshold, so the job continued — permanent RNG divergence from that tick forward.

Test plan

• Load MP game with CommonSense active, verify meditation economy still works (pawns stop meditating when joy is full)
• Verify no desync occurs during extended play with meditating/reigning pawns
• Verify single-player is unaffected (original CommonSense postfix runs normally)

🤖 Generated with Claude Code

[Zetrith]

Some issues:

• Floating point drift isn't an issue AFAIK, is this an AI hallucination?
• SyncMethods only execute from code called in the interface, a call during ticking won't be replicated

[sviyh]

It might be. I'll dig a bit more.

Somehow it fixed my problem. The code overall made sense to me but it might be due to the lack of understanding of the codebase on my side.

[sviyh]

Desync-1177.zip
Here are traces. So we can see that common sense on host decided to finish meditation while it didn't on local.

[SokyranTheDragon]

So we can see that common sense on host decided to finish meditation while it didn't on local.

You don't know that. You're just guessing from the logs. For all you know, it could have been the Pickup And Haul call that caused the desync. But it doesn't appear in the client traces since it didn't trigger anything that tracing keeps track of.

Or perhaps something else happened beforehand that desynced (but isn't tracked), and this is the first place that the tracing realized that something has happened.

And as @Zetrith pointed out, this is not how synced methods work. You prevent execution for non-host, and the synced method won't get synced since it happens during simulation. So it'll only get called for the host.

[sviyh]

Oh, I see that the fix doesn't make any sense now - so I took it down. I am investigating if that was the growth moment that triggered while the pawn is in outpost from vanilla outposts expanded.

Thanks to both for explaining and pointing out interface syncing specifics