We do not put a label on every station. Extrusion, ageing, sandblasting and the intermediate warehouse run on a route card that travels with the load; the label appears at the last finishing step, before shipping. This is the map of where we tag, the mistake we made about it, and the number we still cannot publish.

Every plant starting this project asks the same question first: barcode or QR? We asked it first too, in January 2023, the week the MES went live and everyone wanted a code on everything.

The symbol turned out to be the smallest decision in the project — an afternoon. The decision that has cost us real time, and is still not finished, is a different one: where on this line does a physical tag exist at all.

Here is where we tag, what we do not tag, and what it cost.

Where a label actually exists in this plant

Stage How the load is tracked Label on the profile?
Extrusion Route card travels with the bundle No
Ageing Route card No
Sandblasting Route card No
Intermediate warehouse (WIP) Route card moves with the material No
Anodising Route card, then converted to finished-goods identity Yes — last step
Powder coating Same Yes — last step
Packing and shipping Finished-goods label Yes — before dispatch

Four processes, no label. The label is not created at the press where the batch is born; it is created at the end, when the material becomes shippable goods. Sandblasting does not print one. Packing does.

Not a compromise we apologise for — a decision.

Why the middle of the plant carries a card instead

Ask what a label must survive mid-process: an ageing oven, a blasting cabinet, oil mist, bundling and un-bundling, forklift handling, cutting to length. Each is a chance for a sticker to come off, and every reprint is somebody walking to a printer with dirty gloves.

So the middle of the plant runs on a route card that travels with the load, written when the load is created and staying with it through ageing, blasting and the warehouse. Not elegant technology. What it has going for it is attachment: it belongs to the bundle the moment the bundle comes into existence, which is the only moment when attaching anything is cheap.

Finished goods are the opposite: packed, dry, about to be handed to somebody else — a clean surface where a printed label is applied once, read by our dispatch and then by the buyer.

Tag where the tag will survive, not where the process is most interesting.

The mistake: we assumed the scan was the answer

In EP02 of this series we published a number we could not deliver: that tracing a batch takes about a minute. It does not.

What actually happens: a buyer calls with a problem, our salesperson enters the batch, and the MES returns the billet lot, the press, the die, the shift, the recorded billet temperature and ram speed, and every inspection record on that batch. That lookup is one scan — seconds. Then the work starts: confirming the affected scope, quarantining it, writing the answer and sending it. That part takes us within one shift — up to eight hours on the days it happens.

We corrected the article rather than the claim. EP02 now says in its first lines that one minute is the target and one shift is today, and it carries a quarterly commitment to publish the measured closure time whether it improves or not.

The correction was uncomfortable, and cheap next to the assumption behind it. For a long stretch we behaved as though traceability were a printing problem: the scan worked, the screen filled with records, therefore the plant was traceable. It was not. The scan was returning a page, the page had holes in it, and we were still discussing label stock. That assumption is paid for every time a complaint needs a real answer — most of a working day of somebody's time, spent on the steps after the scan.

What breaks is the join, not the code

Two links decide whether any of this works:

  1. Surface-treatment batch ↔ extrusion batch. Anodising and coating run as their own batches; each has to point back at the extrusion batch it was run on.
  2. Inspection record ↔ batch. Every measurement has to attach to the batch it was taken on — not to a shift, not to a day.

Both are supposed to happen automatically. Both still need a human in some cases, and we will not publish that share until it is measured rather than estimated.

A code can only point at a record. If the record is not attached to the batch, the code points at a page with a hole in it, and the answer gets assembled by phone calls — which is what we were trying to escape in 2023.

The six questions that decide 1D or 2D

Ask us which symbol, and we do not answer with a symbol. We ask six questions, and the first four have nothing to do with the technology.

Question What it pushes you toward
What are the operator's hands doing? Gloves, grease, material overhead A laser-read 1D code at arm's length wins where nobody can stop to frame a phone
What must survive the read? Grit, heat, oil, a scuff across half the label A 2D code carries error correction; a scuffed 1D code is usually just gone
How much must the code carry? Batch number only, or batch + spec + date + machine + shift? A short string fits 1D; a full identity lets capacity decide
Whose system reads it? Buyer's goods-in scanner, forwarder, inspector Not negotiable: if customers expect a standard 1D with application identifiers, your preference ends there
Who scans it outside the plant? Customer, forwarder, engineer with a phone Every phone reads 2D with no app — the argument for 2D on outgoing goods
What printer, and who reloads it at 2 a.m.? The symbol is only as good as print quality and the person changing the ribbon

Notice the order: hands, environment, data, other people's systems, outside readers, printer — and only then the symbol. That is why "QR holds more data" settles nothing in a plant: capacity only matters once you know what the code must carry, and on most extrusion lines that is a batch number and a spec.

The order we would do it again

  1. Decide where the tag physically exists. As few places as possible, as late as the customer will accept.
  2. Make the joins complete. Surface-treatment batch to extrusion batch; inspection record to batch. No exceptions, no manual step.
  3. Then choose the symbol — by then it is mostly about other people's scanners and printers.
  4. Then measure. Mis-reads per station per month, closure time per complaint, published whether they improve or not.

We did step 3 before step 2 in places. That is why the scan is fast and the answer is slow.

What still does not work

  • Four stages carry no label. If a route card is lost or illegible, that batch is rebuilt by hand from the MES and from memory. Rarely is not never.
  • The joins are not at 100%. Those two links come first.
  • We have no measured mis-read rate to publish. We can tell you which scanner reads which label; we cannot yet put a monthly number on it, and a guess dressed as a measurement is worse than none.
  • The route card is still the record for the middle of the plant. Until it is digital, the honest description is "labelled at the end, carded in the middle" — not "fully tagged".

Research note

This is a single-plant case observation, not a benchmark. It suggests that in extrusion — where the work unit is split and re-bundled several times before it becomes goods — the symbol is a downstream decision, and the dominant variable is how completely process batches are linked. Testing it on a larger sample takes two numbers: mis-reads per station over at least a quarter, and measured closure time per complaint. We publish the second quarterly; the first we have not measured, and we would rather say so than put an estimate in a table.

Question for you

If you run a plant: where in your process does the physical tag first exist, and how many operations happen before it?

If the answer is "at the end, because that is where we ship from", you are running our design, with our gap in the middle. If it is "at the first operation, on the billet", tell us how the label survives the oven — that is the problem we have not solved.

Community · Built in the open

Join the discussion

We publish this series the way open-source projects publish code: in public, with the numbers, mistakes included. Ask a question, challenge one of our figures, or tell us how you solved the same problem in your plant. Every question gets answered by the person who ran the project — and the suggestions we adopt are marked Adopted and kept on this page.

0 contributions 0 adopted into how we work Replies within 2 business days

No questions yet — be the first to open one. The people who run the plant read every submission.

Ask a question or send a suggestion

No sales pitch in return. We only use your contact details to reply to this thread.

Interested in Our Products?

Contact us today for a professional quotation tailored to your requirements.

Get a Free Quote