question

Here is how I would do it. I am including two example models. The first one (batchingbyparts.fsm) is still using the Combiner, but I have added some other Fixed Resources and logic. The second one (batchingbyparts-processflow.fsm) replaces the Combiner with a Processor and is using Process Flow logic for moving all the flowitems. It also utilizes the Processor's logic to handle process time and using an Operator during that time. As far as I can tell, neither one contradicts what you described in your comment, but let me know if I am mistaken.

First Example Model:

I have Q_A, Q_B, and Q_C set a label named "batchSize" on the first item of a batch in the Queues' OnEndCollecting trigger. The value set on batchSize will reflect a lower batch size if a batch is released early with fewer flowitems than the set batch size of a Queue.

Next, I added an intermediate Queue between the three Queues and the Combiner and named it LoadingQueue. This Queue uses its OnEntry trigger to set its batch size based on the batchSize label on the first item of a batch. I also use the OnEndCollecting trigger to set the Combiner's component list to reflect the current batch size.

I added a Source called PalletSource that only ever creates one pallet when the model first starts and places it in the PalletQueue Queue. PalletQueue is the first input port on the Combiner so the pallet will always be the container flowitem for the Combiner. When the combined pallet and boxes are separated later in the model, the pallet will always come back to PalletQueue and await the next batch to enter the Combiner.

In the Combiner, I set the Combine Mode to be Pack since you indicated you wanted to use an Operator for process time, mostly because having one item (the combined items) for process time with an Operator is easier to handle. I also set the process time to "Use Operator(s) for Process".

Finally, the added Separator is set up with a Process Time of 0 so no additional delay time is introduced. And then I just made sure the output ports were set up to work with the Separator's default Send To Port logic in the Flow tab - container (pallet) goes to output port 1 and contents (boxes) go to output port 2.

Second Example Model:

I removed any logic from Q_A, Q_B, and Q_C except the batching logic. I have the Process Flow use an Event-Triggered Source to create a token every time one of the Queues triggers their OnEndCollecting trigger. Labels are set on the token to keep track of the batch size and originating Queue. The token is then pushed to a local list.

Also in the Process Flow I have a Schedule Source that creates one token when the model starts, which in turn creates a pallet in the PalletQueue Queue. That token then pulls from the list and uses an Assign Labels activity to transfer information from the pulled token - the batch size and queue of the current batch. It also goes through a Sub Flow to create an array named flowitems, which contains references to all the boxes in the batch. That label is used to move the referenced flowitems from their Queue to the pallet, then the pallet is moved to the LoadingQueue Queue.

The combined pallet and boxes goes through the Processor normally (without needing Process Flow), where an Operator is used during process time. When the Processor is finished processing, a Wait for Event activity is triggered in the Process Flow so the "pallet" token will continue with the rest of its Process Flow activities. Those activities move the flowitems from off the pallet and into the Q_Debug Queue, move the pallet to PalletQueue, and remove the flowitems label/array so it can be used again with the next batch.