You won the argument and lost the claim, because you could not prove it

The commercial manager already knew they were right before anyone sat down. The sequence on the whiteboard matched what the supervisor had been saying for weeks: the instruction arrived late, the drawing the gang built from was not the drawing the office said was live. The other side stopped arguing about the engineering. Then someone asked for the issued set as it stood on the morning the work started, and the file could not do what the room had just done.

You can win the argument and still lose the claim. That is the ordinary commercial fate of a job that can narrate a technical truth and cannot reconstruct the issued set, the transmittal and the contemporaneous record that would make that truth into evidence. The industry already knows how to argue. It is much worse at proving.

Winning the argument is cheap

A technical argument is cheap to win because it happens while the people are still there.

The coordinator can point at the clash that was never closed. The package manager can walk the grid. The supervisor can hold up last week's plot. The commercial lead can recite the verbal hold that never became a numbered instruction. In the room, those artefacts still have witnesses. Memory is shared. The wall that came down is still in the programme. Nobody needs a transmittal number to know what happened.

That is why teams walk out of those meetings expecting the money to follow. The team was right. The delay was real. The change was instructed. The site built what it was given. If the other side is reasonable, the claim will follow the argument. Reasonableness is not the test. Reconstruction is.

A claim is not a conversation. It is a demand that a dated, attributable, complete record be produced for a fact the other party is entitled to deny. The fact may be a delay, a variation, a disruption, a late instruction, a superseded issue used at the workface. The argument can be correct on every engineering point and still fail, because the thing being tested is not the engineering. What is being tested is whether the project can show which containers went out, who received them, and in which published state, on the morning the work started.

What the claim actually asks for

The other side does not have to disprove your memory. They have to ask for the file.

They will ask which issued set was current on the morning the work started, and whether that is the set the gang had. They will ask for the transmittal that named the containers, the revisions, the purpose and the recipients. They will ask for the contemporaneous record of the instruction that qualified those containers — hold, omit, invert, proceed in advance of the revision — captured then, not assembled later from inboxes and interviews.

Those three artefacts are the proof. Not a narrative. Not a reconstructed timeline drawn on a whiteboard six months after practical completion. Not a pack the commercial team built because the first search failed.

Projects lose here in a typical pattern. Nobody needs a named dispute to recognise it.

A change is discussed in a coordination meeting. Someone says proceed. The published container is not yet updated, or it is updated and the hut set is not. A zip leaves the office. A mailbox acknowledges. The installing company is two companies further on. The pour happens. Months later the delay is obvious, the cost is real, and the argument in the room is still easy to win. The issued set cannot be reconstructed as a single live instruction. The transmittal proves a send, not receipt at the workface. The contemporaneous record is a thread whose attachments no longer open, or a personal mailbox whose owner has left.

That pattern is how most jobs actually keep information. It becomes expensive only when someone asks you to prove it.

The issued set that cannot be reconstructed

An issued set is a bounded act: these containers, these revisions, this purpose, this list of recipients, this moment. ISO 19650's published state exists because “the file is on the disk” is not the same fact as “the file is authorised for this use.” A folder named Issued is storage. A claim needs the act.

What most projects can produce is weaker than that. A register with a green cell. A screenshot of Rev C in a portal at 16:40. A zip whose filename has mutated. Three copies of the same sheet number, two of them still looking live: the published PDF, the plot on the plywood table, the crop in a group chat. Each copy is locally true. Together they are not an issued set. They are a disagreement the project never recorded as a disagreement.

A claim that depends on “we were working to C” fails if the workface copy is B and nobody withdrew B. A claim that depends on “C was late” fails if you cannot show which C, sent to which named recipients, against which previous revision, on which date. Being right that the sequence was wrong does not survive if you cannot name the containers that constituted the sequence.

The unofficial set in the trailer is often the only artefact that can say what was actually built from, and that is not a side issue. If that set was never tied to the published container — same identity, same revision, same withdrawal — the office can win the meeting with a screenshot and still lose the claim, because the claim is about the instruction that authorised the work, not the file that was stored.

The transmittal that is not evidence

A transmittal is supposed to turn storage into an instruction. Number, date, containers, revisions, purpose, named recipients. That is the act a tribunal, an adjudicator or a contract administrator will ask you to reconstruct. Most transmittals in circulation are not that act. They are a cover sheet for a send.

Issue is not receipt. Receipt is not understanding. A download by the main contractor's document-control mailbox is not proof the installing company had the pack. An acknowledgement tick is not proof anyone knew what changed. A transmittal issued at 16:55 for work that starts at 07:00 is issued. It is not received in time. Dating the send does not date the use.

When the claim arrives, the project reaches for the transmittal log because the log is what the system can print. The log will often show that something left. It will rarely show that the named person at the installing company had the exact revisions before the work started, or that the previous revision was taken off the wall. A green issued cell is administratively cheap. It is evidentially thin. The other side does not have to call it a lie. They only have to say it does not prove the fact you need.

If you cannot reconstruct the act — not the mood of the meeting, the act — you do not have issue as evidence. You have a timestamp.

The contemporaneous record that is an inbox

The third failure is the one that looks most like competence while the job is busy.

The instruction that changed the works usually did not enter the issued container. It arrived as mail. “Proceed as discussed.” “Hold the riser until Rev D.” “Don’t wait for the revision.” A marked-up PDF attached to a thread whose subject line has already mutated. A confirmation of a verbal instruction that never became a numbered document. A forward to a subcontractor at 18:40 because the upload will happen in the morning.

By morning the work has moved. The CDE still shows yesterday's published revision. The sentence people acted on lives in a mailbox, sometimes in a screenshot of a message that started on a phone.

An inbox is a personal communications tool. It does not become a project-owned container with a state, a revision, and a named recipient list the project can still retrieve when the sender has left. Attachments disappear in ordinary ways. Links expire. Archives strip binaries. Leavers take custody with them. A reconstructed pack assembled months later, with gaps in the chain and files that no longer open, is a different artefact from a record captured at the time.

The contract forms assume an instruction can be found. NEC compensation events, JCT architect’s instructions, FIDIC Engineer’s instructions, and the ordinary confirmation of verbal instruction all presume a dated, attributable, complete record. They do not presume that the record is a thread someone will one day search for by remembering a word in the subject line. Those texts have owners. They assume the instruction is still findable. They do not assume you will rebuild it from a subject line.

In adjudication the timetable is short. You do not have six weeks to interview leavers and restore a personal archive. Entitlement that was real on site can fail for want of a complete contemporaneous file. The other side does not have to win the engineering. They only have to show that the proof is a reconstruction.

The technical win does not transfer

Winning the argument is a fact about the room: the sequence, the drawing, the instruction, the delay. Losing the claim is a fact about the file: the issued set that cannot be bounded, the transmittal that cannot carry receipt, the contemporaneous record that was never a project-owned container. The second fact does not care that the first was correct. Being right does not earn you a courtesy on the proof; the claim can only spend the file.

The labour of the reconstruction is then paid as commercial time, as a variation that will not settle, as a delay that cannot be particularised, as a file the project never owned. None of that appears as a line item called “records.” It appears as a claim that dies on its feet.

If you want to know whether last month's delay would survive, do not start with the narrative. Pick one event the commercial manager already believes they won. Ask document control for the issued set as a single bounded act — containers, revisions, purpose, recipients, time. Ask for the transmittal that can show receipt at the installing company, not only a send from the office. Ask for the instruction that qualified those containers, captured then, sitting in a project-owned container rather than a leaver's inbox. Give yourselves an afternoon, and do not ring anyone who was in the original meeting. If those three artefacts cannot be produced from the system, the next claim will fail the same way this one did.

Closing that gap is not free. It means treating issue, receipt and withdrawal as named duties during the job — usually a document controller's, with the package manager responsible for the workface copy — and it means putting verbal holds and “proceed as discussed” into a numbered instruction the same week, not when the compensation event is particularised. Principal contractors and clients are the ones who can make that a condition of the appointment, and they will have to pay for the hours while the job is live. Until they do, the commercial manager will keep winning the room and paying for the silence afterwards.