they were on the wall before anybody noticed. eight panels, hung overnight for a space that opened at nine, and every one of them was the right artwork for somewhere else. the artworker who supplied those files took a phone call at ten past six that morning. i still think about that call, because she had not done one thing wrong. she had been asked for the artwork against a code and she had sent the artwork against that code. it simply was not the only item wearing it. three weeks earlier, a late change had come in and a row had been added to the matrix by copying the row above it. the code travelled with the row. whoever did it meant to go back and change it, and then something else landed on them, the way something else lands on all of us about forty times a day. from that moment two different items answered to the same name, and every system we owned believed the name, because believing the name is the only thing a system knows how to do. there is no clean way out of that at ten past six. you cannot reprint eight large-format panels before nine. so you take down what is wrong, you move what can be moved, you explain it to a client who is standing there watching you explain it, and the space opens with gaps in it. the panels were not really the cost. the cost was that a client watched us not know something about our own job. and here is the part i have never got over, which is why i am bothering to write this down. the person who carried it was the artworker. not whoever copied the row, because nobody ever knew who that was, and by then it did not matter. her name was on the files, so it was hers. that is the shape of this particular failure every time: it lands on the person furthest from the decision that caused it, and they are usually the one person in the chain who was doing exactly what they were told. this is what event artwork version control comes down to. naming, whatever the software people tell you. the moment one name can mean two things, nothing downstream can save you, because everything downstream is obedient. it is worse than a wrong file for one reason. a wrong file eventually looks wrong. a duplicated code looks perfect, right up until eight of them are on the wrong wall. so the answer was never that everyone concentrates harder. a job that only holds together while nobody is having an off day is not a job, it is a bet, and you will lose it in the third week when the changes come in faster than the thinking does. the answer is that a code gets issued once, at the moment the item is briefed, by the thing doing the briefing, so there is no step left where a human hand can give the same name to two things. one item, one code, one row, from the moment it exists. and one person owns that document. artworkers put their final links in it. suppliers take artwork out of it. producers stay out of it altogether, which sounds unkind and is not. a source of truth with several hands in it stops being a source of truth. it becomes a negotiation, and it can no longer settle the one argument it exists to settle. we had an artworker on a job last year with a file that did not sit right against the dimensions on her row. most people would put that down to a bad export and ask for another one. she went and checked the code itself, got two rows back, and sent one message. it took her about a minute, and nobody downstream ever knew there had been a minute. she checked because somebody taught her to distrust a code she had not verified. that is most of what we actually do, and it is why nobody ever rings us at ten past six. a code that means two things means nothing.