I noticed a Houdini behavior:

Say I'm at Frame 30. When I change anything in the DOP network (like friction of the particles), Houdini will automatically re-run the simulation for one frame (not from Frame 1 to Frame 30, but from cached Frame 29 to Frame 30).

It's very annoying and useless. I can't think of one case where I'd like to simulate 1 frame in the middle of simulation.

Disabling simulation by clicking the "brain" icon stops this. But now when I like to actually re-run the whole simulation, I need to jump to frame 1, enable simulation, then run to Frame 30.

Is there a way to simply tell Houdini to never re-run simulation when it's not at Frame 1?

When any parameter inside a DOP simulation is changed, or when any referenced external node is changed, the cache of the simulation is marked as invalid. If this parameter is turned on, the next time the playbar is moved to reach a simulation time of 0, the cache will be cleared and the first simulation timestep will be recalculated. If the cache is invalidated while at a simulation time of 0, the initial state is recalculated immediately. If the current time is beyond time 0, then the most recent timestep will be recooked, and the cache beyond the current time will be cleared. But all prior timesteps will be left untouched other than to be marked invalid. If this parameter is off, the cache is marked as invalid in exactly the same way, but the cache is never cleared automatically. To recook a simulation in this mode, the Recook Simulation button on this parameter dialog or above the viewport must be used.

This Enable Automatic Resimulation option is on the container itself.

Mind blown! Thank you very much