A transshipment hub is described as a place. Shippers and carriers talk about it that way: the box goes via the hub. But what is actually being bought is a connection. A container arrives on a mainline vessel and has to be on a feeder that leaves within a window. The hub's value is the probability that the connection is made, and that probability is a scheduling outcome, not a geographic one.
The arithmetic of a missed connection is unforgiving. If the mainline arrives 18 hours late and the feeder holds for 6, the box waits for the next loop. On a weekly feeder that is 6 to 7 days of dwell, a yard slot occupied for a week, a customer promise missed by a week, and a second handling charge. One late vessel does not miss one connection. It misses every connection on board that was tighter than 18 hours.
Three plans, no connection
The port plans berths. The carrier plans the loop. The terminal plans the yard. Each plan is optimized on its own terms, and each is reasonable. The connection is the thing none of them plans, because it exists only where the 3 meet. It is the joint of the system, and joints are where physical networks fail.
The failure shows up as a schedule that is correct on paper and wrong on the water. The feeder timetable was set at the start of the season against the mainline's published schedule. The mainline's actual arrival is a distribution, not a time, and on some routes the spread of that distribution is wider than the connection window it was supposed to fit inside.
The departure is a decision
The corrective is to treat each feeder departure as a decision rather than a timetable entry. Hold or go. The inputs are known or estimable at the time the decision is made: how many boxes are connecting from the inbound vessel, what its arrival distribution looks like now rather than at season start, what each of those boxes is committed to onward, and what a delay costs the boxes already loaded and the ports further along the loop.
Sometimes the answer is to hold 4 hours, because 60 boxes are connecting and the inbound is showing an 80 percent chance of berthing within the window. Sometimes the answer is to sail on time and roll 8 boxes to the next loop, because holding would push the feeder's next 2 port calls into congestion and the cost lands on 300 boxes instead of 8. The rule is different on every departure, which is why it cannot be set once in a schedule. It has to be computed, and the computation has to see the inbound ETA as it moves.
Experienced feeder planners make this call by phone every day, and often make it well. The point is not that the decision is new. It is that the inputs are now estimable well enough to compute it, and to compute it for every departure rather than the ones that happen to get a call.
What the hub is selling
A hub that plans its connections can sell them: a connection probability, by service pair, that a carrier or a shipper can price against. A hub that plans berths, loops, and yards separately can only sell a location and hope the timetable holds. In a normal year the difference is a few points of connection reliability. In a disrupted one, when every inbound arrival distribution widens at once, it is the difference between a hub that reroutes and a hub that fills up.