A customer asked me last week why their box "arrived" on the 2nd and was still not collectable on the 6th. Nothing had gone wrong with the data. The vessel had arrived — at the anchorage. It berthed four days later. Both facts were in the tracking feed, and the dashboard showed one of them.
That gap is not a bug in anyone's system. It is what happens when a forecast and a measurement are displayed in the same colour.
Three claims that look identical on screen
Every milestone in a carrier feed carries a status, and there are essentially three.
An estimated event is a model output. Nobody observed anything; a system predicted a time from a schedule, a position and a historical average. It changes without anything happening in the physical world.
A planned event is a commitment. A terminal has allocated a window, or a carrier has published a rotation. It is a statement of intent by someone with the power to act on it, which is a stronger claim than an estimate and a weaker one than a fact.
An actual event is an observation. The gate went up, the crane moved, the pilot boarded. It is the only one of the three that cannot be revised, only corrected.
Most dashboards flatten all three into a date. Your planning system then treats a model output and a gate scan as the same kind of thing, and in a stable market it gets away with it, because the estimate and the actual converge to within a day.
They are not converging right now.
What the congestion is doing to the estimate
Berth waits at Shanghai and Ningbo have been running to ten days after Typhoon Saudel and the four named storms before it. Roughly 3.92m TEU — about 11% of the world fleet — has been sitting in queues, and more than half of that in North Asia.
An ETA model built on historical port stay does badly in exactly this condition, because the history it was trained on does not contain a ten-day anchorage wait. So the estimate updates a day at a time, always a day or two behind the reality, and your ETA drifts rather than jumps. Drift is the worst failure mode for planning — a single large correction gets escalated, twelve small ones get absorbed until it is too late to react.
And the global figures underneath say the same thing. July schedule reliability was 56.4%, and late vessels arrived 6.06 days late on average. If nearly half of arrivals miss their window by the better part of a week, an ETA is a probability distribution, not a date.
The one milestone worth wiring an alert to
Not vessel arrival. Not discharge. Availability — the moment the container is released, cleared of holds and appointable.
It is the only milestone that maps to something you can act on, and it is the one your demurrage clock is racing. A box discharged on Monday and made available on Thursday has burned three days of a five-day allowance while every dashboard in the chain showed a green tick against "arrived".
The practical version of this is short. Alert on availability rather than arrival. Ask your carrier which of their milestones are actuals versus estimates and get it in writing, because it varies by carrier and sometimes by service. And when you quote a delivery date to a customer, quote from the availability forecast plus your inland lead time, not from vessel ETA plus a habit.
One honest limitation: availability is not published cleanly by every carrier, and where it is not, you are reconstructing it from discharge plus hold status plus appointment feeds. That reconstruction is imperfect and you should say so internally rather than presenting it as a hard number.
The tracking view separates estimated from actual milestones rather than merging them, and the port pages show the current call picture at the terminal your box is sitting in — which is usually the faster way to find out whether a drifting ETA is about the ship or about the queue.
If you change one thing this month, make it the alert. Everything else on this list is a conversation with a carrier; that one is a checkbox.
