The Scrap Bin Was a Receipt

The part got all the way to assembly before anyone noticed it was wrong.

It had been cut. It had been bent. It had been welded. Every operation performed correctly, by people who were good at their jobs, against a drawing that was no longer true. Nobody found out until a guy on the line went to put it where it went, and it didn't go.

That's the expensive kind of wrong. Not wrong at the laser, where you've burned a sheet. Wrong at assembly — where you've burned the sheet, plus the brake, plus the weld cell, plus the hours, plus the schedule. And now you get to do all of it again.

We ate that more than once. Batches of parts, cut to a revision that had already been superseded, nobody the wiser until the pieces finally met.

Sit with this part, because it took me too long to see it: nothing had gone wrong in the shop. Every station did its job. The laser cut what it was told to cut. The brake bent what it was told to bend. The failure was upstream of all of it, in a place with no machine in it at all.

Somebody had changed the part. And nobody wrote it down.

---

How changes actually happened

We had no change control. That sounds like a policy gap. It isn't. It's a physics problem.

Anyone could change anything. A welder who didn't like the gap he had to fill every morning could ask for it to be closed up, and it would get done.

And here's the thing people miss when I tell this story: he was usually right. The man standing at that fixture eight hours a day knows something the drawing doesn't.

Right about his eight hours.

That's the qualifier, and it's the one nobody attaches. He had no line of sight into what that gap was doing downstream — whether the clearance he was closing up was the clearance assembly needed to get the thing to seat. He wasn't being careless. He was being local, which is all any of us can be from one station. Nobody gave him a way to check, and nobody was checking on his behalf.

So a good change, made in good faith, arrived downstream as a ghost. Inherited by departments that got no input, no notice, and no chance to say wait, we need that gap.

The change was never the problem.

The problem was that his change entered the world as a running revision, quietly reinserted into the system, with no record that it had happened and no explanation of why. Six months on, nobody could tell you who changed it, when, or what it was meant to fix. There was no why. There wasn't even a that — no record a change had occurred at all.

Meanwhile purchasing is buying to a bill of material describing a part that no longer exists. The laser is cutting a nest of the old geometry. And every station downstream keeps running — confidently, competently — on a description of reality that stopped being true weeks ago.

The shop was never out of control. The record was out of control, and the shop was faithfully executing it.

---

The part that changed twice

Here's where it went from expensive to absurd.

The gap gets closed. Weeks later, sometimes months, somebody else — different station, different problem — puts in a request to open that same gap back up. Because from where he stands, he's right too.

Neither man knows the other exists. Neither knows he's in an argument. There's no record of the first change, so there is nothing for the second one to contradict. The drawing just quietly flips back.

That is a conversation between two people who never met, conducted entirely through a part, at the speed of production, with a batch of scrap generated on every swing.

And it can run forever. Nothing in the system ever says this dimension has moved twice and the last two guys disagreed. Nobody is counting, because nobody is writing it down.

A part that oscillates is the loudest signal there is that you have no record. It's two right answers with nowhere to have the fight.

---

The scrap bin was a receipt

That's the line I'd put on the wall of every plant I walk into.

Your scrap bin is not a manufacturing problem. It's a record-keeping failure that finally got expensive enough to see.

Think about what that scrap actually was. It was the moment the physical world and the written world disagreed loudly enough that a human noticed. Everything before that moment — the change, the silence, the stale BOM, the bad nest — was invisible. Free. Costless right up until it wasn't.

So the bin isn't where the waste happened. It's just where the bill came due. The waste happened weeks earlier, in a conversation nobody logged.

And the price scales with how long the record stayed wrong. Catch it at the drawing and it costs you a conversation. Catch it at purchasing and it costs you a PO you can still cancel. At the laser, you've burned material. At the weld cell, material plus three operations. At assembly, all of it plus the schedule plus the promise you made a customer. And in the field, all of that plus your name.

Same error. Six prices. The only variable is how long the record was allowed to lie.

---

What we built

Two moves. Neither is clever. That's exactly why they worked.

Lock the part. We stopped letting state move silently. The part got locked down in Vault, held in its current state, so it could not quietly become something else while everyone kept believing the old thing. Before that, the system of record was a spreadsheet and a set of habits — which is to say, it was nothing. A record anyone can change without leaving a mark is not a record. It's a rumor with a filename.

Open the door, then gate it. This is the part I'm proudest of, and it's the opposite of what people expect from a process guy. We made it easier to ask for a change, not harder. Anyone could submit an ECN. The welder. The assembler. The guy who noticed the fixture fought him every single morning. Anyone. Submit it.

Then it got triaged. Somebody looked at it and answered two questions: does this change actually need to happen — and who else does it touch?

That second question is the one the welder could never have answered from his fixture, and it is the entire reason triage exists. Triage is not a bottleneck. It's the moment downstream gets a vote. Without it, every change is a decision made on behalf of people who never heard about it — and sooner or later, one of them files the opposite request, and you're back to a part that can't make up its mind.

That's the whole trick, and most shops get it exactly backwards. The instinctive response to a scrap problem is to clamp down — make requesting a change painful enough that people stop asking. All that does is drive the change underground. The welder still closes the gap. He just stops telling you.

We went the other way. Ask freely. Approve deliberately. Record always.

The gate was never there to stop changes. It was there to make sure a change couldn't happen without leaving a record — and without the people who'd inherit it ever getting a say.

---

The tax, and who paid it

A real process costs something, and if it doesn't, it isn't real. I said that last time about stage-gate. It's just as true here.

Here's the bill on this one. Once an ECN was approved, a person was responsible for running down everything downstream of it. The BOM. The purchase orders. The drawings. The nests. Every place the old truth was still sitting there, being believed by somebody. A human had to go find them all and correct them, one at a time, on diligence and memory.

It worked. It was also the worst-designed part of the whole system — and I'm the one who designed it.

Because that is a human being used as a propagation engine. Record-keeping — the thing people are worst at and machines are best at — handed to the one component in the building guaranteed to be tired, busy, and interrupted. We solved "the record doesn't update itself" by making a person into the update mechanism. It held. But it held on conscientiousness, and conscientiousness is a hell of a thing to use as a load-bearing wall.

If I'd had AI... the obvious answer is monitoring — watch the floor, spot the scrap pattern, raise a flag. I don't think that's the answer. Anomaly detection just brings you the bad news sooner. You're still making scrap, only less of it. The real fix is that an approved change should propagate itself. Approve the ECN, and every downstream record that depends on that part — BOM, nest, PO, drawing, work instruction — updates, because all of them were reading from the same object to begin with. Nobody runs anything down. And every department that inherits the change is told, automatically, before it costs them anything. The scrap doesn't get caught faster. It never gets made.

---

Your turn: Go look in your scrap bin this week. Pick one thing out of it and trace it backwards — not to the station that made it, but to the decision that made it wrong. Find the moment the record stopped being true, and ask who knew.

Then ask the only question that actually matters: in your shop, right now, could a change happen without anyone writing it down — and without the people who'll inherit it ever hearing about it?

If the answer is yes, that bin fills back up. It's a receipt. You're still running a tab.

And if you can find one dimension that's moved twice, in opposite directions, you don't have a scrap problem. You have two good people who were never introduced.