Runink FACE

Cold Chain and Yard Safety

A container drifts warm overnight and nobody opens the door until morning. What the yard camera already sees, read against the handover record and the paperwork the consignment travels with.

20 May 2024 · 6 min read

Runink FACE · Reactive logistics

This is a Runink FACE scenario, its yard-facing end. It is an illustration of the mechanism, not an account of a deployment. What the software reads here is the paperwork and the images a cold chain already produces — the consignment record, the handover, the photograph taken at the door — against the rule that governs the shipment. What FACE is.

In Short

  • The yard camera is read where FACE runs. A frame arriving from a yard or infrared camera is checked to be an actual image before anything reads it, reduced to a size a model can take, and read by a vision model running where FACE runs, not at an outside service. What comes back is a written observation tied to the frame it was read from.
  • A cue reaches everyone watching the yard. FACE can broadcast a cue — slew that camera, hold those crane moves — to whatever is subscribed to the yard's event stream. The record shows it as exactly that, a request, so a cue is never mistaken for a move already made.

The Reading Has To Arrive First.

The reading that condemns a load is recorded hours before anybody looks at it. The whole problem is the gap between the two.

Where It Goes Wrong

A cooling unit starts to fail on a Tuesday night. The sensor records it. Nobody is watching at that hour, and the data is not looked at until the container is opened at the far end.

By then the question has changed. It is no longer "can we save this load", it is "who pays for it". That is a much more expensive question, and it is the only one left.

The reading was there all along. Nobody was reading it.

The yard has the same shape of problem. Rules about which goods may sit near which are known, written down, and checked by a person who is also doing four other things.

What Happens Instead

It starts with the camera side of the yard. A frame is validated as a real image before a model sees it, scaled down to something a model can take, and read by a vision model where FACE runs — so the footage does not go out to anybody's API to be described. The observation comes back attached to the frame it came from, which is what makes it arguable rather than assertable.

On top of that, a cue can be broadcast to everything watching the yard's event stream. The broadcast asks the people and systems watching to act, and the record shows it as a request, never as a move that already happened. So "is this handled?" always gets a true answer.

Where there is something to act on, it waits as a drafted move, and a named person approves, edits or rejects it with the sign-off kept on the record. Approving is what sends it. The response names each step that ran and any that could not, instead of reporting success, so "approved" and "done" stay two different words.

Who Owns This

Three desks, and what each of them is holding today.

  • The yard supervisor. Today the rules about which goods may sit near which are known, written down, and checked by a person who is also doing four other things. What changes is that the frames the yard cameras already record get read, and what comes back is a written observation attached to the frame it came from — a thing you can open and disagree with, rather than a line in a book.
  • Compliance and risk. Today a segregation finding exists if somebody happened to walk past. What changes is that a cue can be put on the yard event stream for whatever is watching it, and the move that follows is drafted and waits for a named person to approve, edit or reject it. The sign-off is kept with the frame that started it.
  • IT and information security. Today a camera product means asking whose service the yard footage is being sent to in order to be described. What changes is that the frame is validated as a real image and read by a vision model where FACE runs, so the review covers the machines your plan names, not an outside service.

How You Would Know It Worked

Every figure below is yours, not ours. Write down where you stand today, because the baseline is gone for good the moment anything changes.

  • What you write off on chilled and frozen stock. The write-off account in your ledger and the quality rejection log for the same period, with temperature cases separated from every other cause.
  • Hours from the first bad reading to somebody acting. Sample last quarter's temperature events. Note when each was recorded, and when a person first did something about it.
  • How many drifts were caught while the load could still be saved. Count them as a share of all drifts. This is the number the rest of it rests on.
  • Safety findings in the yard. Your own inspection and incident records, counted per quarter, split between rules about keeping goods apart and everything else.
  • What your cameras already record and nobody reads. Count the cameras in the yard, then count how many hours of what they record is ever looked at by a person. This is the gap the vision side of this works in, and it is usually the largest number on the list.

Bring your write-off account and a day of yard camera footage.

Questions A Yard Team Asks

What comes up before anybody talks about a contract.

What does the camera side actually read?
A frame from a yard or infrared camera. It is checked to be a real image before anything reads it, reduced to a size a model can take, and read by a vision model running where FACE runs, not at an outside service. What comes back is a written observation tied to the frame it was read from.
Does the footage leave the site?
Not to an outside service. The frame is read where FACE runs: your own servers or cloud account on Dedicated and Enterprise, Runink’s shared machines on Lite. It is described there, and the observation comes back attached to that frame.
If it sees something, does it stop the crane?
It broadcasts a cue — slew that camera, hold those moves — to whatever is subscribed to the yard event stream. The cue is a request, and the code will not report it as a move that was stopped. What turns a cue into a move is a named person approving the drafted action.
Who signs off an action it drafts?
A named person, whose approval, edit or rejection is kept on the record. Approving is what sends it. The response names each step that ran and any that could not, so approved and done stay two different words.
What happens when the reading is wrong?
Every observation comes back attached to the frame it was read from, so whoever reads it can open the picture and argue with it. Nothing moves on an observation by itself; the move is drafted and waits for a person.
What should we bring to a first conversation?
Your write-off account and a day of yard camera footage. Bring one more count with them: how many cameras are in the yard, and how many hours of what they record is ever looked at by a person. That gap is the space the camera side works in.

Runink: Data You Can Trust. Decisions You Can Defend.

Runink FACE reads the logistics records you already hold, compares them against the rules that govern them, and drafts an action for the person who owns the decision. It runs on your own servers or cloud account, or on Runink's shared machines to start, with no outside AI service.