Any engineer can revise a drawing in an afternoon. Reaching everyone still working from the version before is another matter: the colleague halfway through a calculation, the contractor halfway through a price. Swedish practice has exact standards for the first half of that job, and a filename for the second. A revision is cheap to make and expensive to broadcast.
The rule
Bygghandlingar 90, the eight-volume work that governed Swedish construction documents for three decades, states it as an order: filenames shall remain unchanged regardless of revision code or stage. The logic is airtight. The filename is the document's identity; change it, and the history snaps, and the same drawing lands at a partner firm as a stranger with no past.
The clients all repeat the rule, each in their own words. Håbo forbids it flat out. Nacka explains the reasoning, to keep the chain of history whole. Göteborg sets the ban in capitals and names the usual sins, stage and date. Jönköping lifts the sentence nearly verbatim from the old handbook. A rule that every client in the country has to restate is a rule under siege.
And it loses at home. Håbo's ban on status words in filenames is published as CAD-Kravspecifikation utgåva 1 KLAR.pdf: an edition, the word "done", spaces included. Locum's guideline on how to name files exists in three versions on the same website, told apart by the dates in their filenames. Neither is sloppiness. When the only way to see at a glance which document is current is the name of the file, the name of the file is where the truth goes.
What people actually believe about those names was measured once. A 2009 study at Svenska kraftnät found engineers versioning by appending a number to the filename. Nearly everyone assumed the highest number was the newest version. Just as many assumed it was the approved one. Those are different claims, and a counter cannot carry both.
The asymmetry
Why does a profession this careful run its version control on filenames? Because a change has two costs, and only one of them is anyone's job. Making the revision is billable work with a name attached: an hour, an afternoon. Broadcasting it is the second cost, and it is larger, and it is ownerless. Every discipline downstream must first discover that the change happened, then work out what it means for their own work. Marking a revision is a task in a standard, with a letter, a date and a signature. Noticing the mark depends on someone reopening a document they had no reason to reopen.
The problem is younger than the profession. Before about 1880, design drawings were barely reproduced at all; the men on site were told what to build. John Herschel had shown the Royal Society the chemistry of the cyanotype back in 1842, but only in the late 1870s did sensitised paper reach the market, and an architecture journal describe running drawings through it. The blueprint solved distribution: identical copies, many desks. It also created something nobody wrote down. From the moment fifty desks could hold the same drawing, somebody had to keep fifty desks current, and nobody was appointed.
The apparatus
None of this went unnoticed. Sweden keeps three live standards on the problem: SS 32206 for versions and changes, SS 32207 for the title block, SS 32209 for status codes. Stockholm's technical handbook requires each change letter in three places at once. Statens fastighetsverk specifies which date changes at revision and which never may. When a drawing becomes a relationshandling, the record of what was actually built, the same care runs in reverse: clouds erased, revision rows removed, because from now on the only question the document answers is what is true, not how it got that way.
The international version is ISO 19650, which defines a common data environment: one agreed source of information per project, every document moving through defined states. Its introduction says why with unusual candour. Considerable resources go to correcting mismanaged information and to the uncoordinated efforts of delivery teams. Even the contracts weigh in. Where documents contradict each other, AB 04 ranks them, fifteen categories deep, contract first, drawings fourteenth. Note what the ranking ranks: kinds of document, not versions of one. It decides whether the specification beats the drawing. It has no opinion on whether the drawing in your hand is the current one.
Then there is the detail that gives the game away. Stockholm's handbook, revision date 15 May 2025, requires revision handling per SS 32206:2008, a standard SIS has withdrawn and replaced with SS 32206:2024. The people maintaining the handbook are not the problem; they are the proof. SIS made the change, and publishing it was the cheap part. Reaching every requirements document in the country that still points at the old number is the expensive part, and it happens one attentive reader at a time.
Carmont
The drainage above the railway line at Carmont in Aberdeenshire was built in 2011 and 2012, and it was not built according to the design. A low bund appeared that the designer had never drawn. A connection to an existing catchpit was left out. A catchpit was moved. Geotextile was omitted.
The Rail Accident Investigation Branch found no evidence that any of these changes was referred back to the designer for consideration, and none that as-built drawings were ever submitted to the designer or to Network Rail.
On 12 August 2020, after intense rainfall, the drainage did not do what its designer had intended, and a passenger train derailed. The conductor, Donald Dinnie, the driver, Brett McCullough, and a passenger, Christopher Stuchbury, died.
The investigators looked wider. Of 64 projects they sampled, more than half lacked any trace of a health and safety file. Of eleven drainage projects, five had never entered the asset management system. Simon French, the Chief Inspector of Rail Accidents, addressed the whole industry when the report came out in 2022. Changes during construction are normal and often sensible, he wrote, but the designer has to understand what is proposed, and review implications that can look inconsequential to the team on site.
I hope this example will resonate throughout the UK's construction industry.
A change was made in the ground, and it needed to reach a desk where someone could work out what it meant. Nothing carried it.
The leverage
In ordinary weeks the same failure costs money rather than lives, and Sweden has measured it properly at least once. Josephson and Hammarlund at Chalmers stationed a full-time observer on each of seven building sites for half a year and logged every defect the observers could see: 2,879 of them, 4.4 percent of production cost. About a quarter of the bill traced back to design, and the most common design defect was not a calculation error. It was lack of coordination, resulting in conflicting drawings.
The Americans put the same thing on an invoice. A 2004 study for NIST carries a line item called information verification: what gets spent checking that the version in front of you is the right one, in an industry where multiple versions of the same file live at every partner firm. For architects and engineers in the design phase, it came to 114 million dollars in a year. Trafikverket describes the client's side in a single observation: costs arise when the asset has changed and the as-built documents have not.
The obvious conclusions are the wrong ones. Fewer revisions? A design that stops changing has stopped improving, and the late change is often the one that rescues the project. More discipline? The discipline exists, written down in more places than any one person can read. What is missing is narrower: when a change lands, the ability to answer, the same day, which documents and which priced decisions were resting on the version it replaced. Today that answer is reconstructed from memory and a distribution list. And it is engineering judgment, not file administration: to know what a change touches, you have to know how the design hangs together.
The standards have seen it coming. ISO 19650-4, from 2022, lists who may review information and adds, without ceremony, that an information reviewer can be an artificial intelligence agent or an automated rules-based process. A committee left that door open before most of the industry knew there was a door. The firms that walk through it first will check an entire document set against a change in an afternoon, and the gap between them and everyone else comes down to one question: when the next revision goes out, how long before you know who is still on the old version?
Get news and articles in your inbox.
The leverage sits in the answer time. Yesper reads an entire document set at once, so which documents and decisions rested on the version before is a question with a same-day answer. Book a demo and we'll show you.
Book demo
