If you use polling you may induce errors. If your error is static, you can poll after the error to see what one(s). Reading may still induce a new error, but if the underlying error is repeatable, the induced errors will likely show as such with a few trials. If the error is dynamic with FETs cycling as you described, it is more challenging.
- You could check for polling induced faults with a scope on ZEDE and DSG for example. If DSG falls when ZEDE is high, it was likely induced by zero delay mode.
- If you are polling at 1s and the fault occurs faster, it is not likely polling
- If it recovers ~ 100ms after the event, it is likely a reset and power up event (see figure 5 in the datasheet)
- If it recovers in <<100 ms, it is likely a failure to see the load detection
The part only allows ~ 100us for DPCKN go get above the threshold before it starts recovery. You can adjust C29 to make it faster, just be sure it keeps any overshoot at protection within abs max limits. R46 is to aid self recovery from a loaded fault condition as described in the datasheet. If placing the pack in the charger to recover is suitable, you do not need R46 If pack- does not move at fault, using the CHGST signal high from a charger for recovery is an option, but not for your schematic. So it gets complicated as you need to control one of those signals. Since DPCKN low is not a condition to startup, you can pull DPCKN up until you want to recover after a fault, then use some mechanism to pull it low such as a certain pack voltage conditon or a signal derived from a pack pushbutton.
Since you hare holding CHGST high at all times, you may consider making R33 and R34 larger to reduce quiescent current. Just don't make them so large that CHGST switches with noise.