Yardline

Warehouse systems know what's inside the building, and transport systems know what's on the road. The yard, where a trailer can sit for a day between the two, is usually a spreadsheet and a guard's memory. When a driver arrives, somebody walks the rows looking for an empty to swap. When a hot load is late at the dock, somebody radios around to find which spot it's in.

Every hour of that has a price. Carriers allow a window of free time at the dock, usually a couple of hours, and bill detention after that. Ocean containers carry their own per-diem charges once they leave the terminal. Those charges are often accepted without dispute, because the site has no timestamped record of when the trailer actually arrived.

Each gate lane gets a compact GPU edge device connected to the site's existing IP cameras over RTSP. Detection models find the trailer, the tractor and the container. Text recognition then reads trailer numbers, plates, carrier codes and ISO container IDs. Container check digits are validated mathematically, so a misread is caught before it becomes a record.

Readings from several frames and several cameras are combined into a single gate event, with a confidence score and the image crops that support it. A low-confidence read doesn't guess. It asks the gate guard to confirm with one tap on a tablet, which is far quicker than typing the whole thing.

A cloud-only yard system has one failure mode that matters: the WAN goes down and the gate stops. Yardline is built the other way round. The site is the source of truth. Every gate event, move and door assignment is written to a local append-only event log first, and the yard state is rebuilt from that log.

When the connection is up, events stream to the central system and to any other sites in real time. When it drops, the gate, the yard map and the jockey's handheld keep working off the local log, and everything syncs, in order, when the link comes back. Because events are idempotent, nothing is double-counted.

A live yard map pushed over WebSockets, showing every trailer's spot, status, carrier and dwell clock. It is colour-coded by how close each trailer is to detention, so the next trailer to cost money is always the most visible one. Yard jockeys get move instructions on a handheld and confirm them with a scan, which keeps the map honest.

The data also keeps a record of when each trailer actually arrived and left, with images. That turns a detention invoice from something the site pays into something it can check.

Gate recognition and the offline-first event backbone are working end to end in the lab against recorded gate footage, and the yard map and jockey handheld are in active development. The next milestone is a pilot at a single distribution centre gate, running alongside the existing process so every read can be checked against the guard's log.

After that comes dock-door scheduling: using the live yard state and inbound appointments to suggest which trailer goes to which door next, rather than leaving it to whoever picks up the radio.