Friday afternoon a man works backwards through four jobsites in his head, lands on a number, and that number is what you pay him.

Everything downstream was left alone on purpose — approvals, payroll and the job's costs read the same hours figure they always did, except now it comes off two recorded instants instead of a recollection. What a man writes down on Friday about last Tuesday is a guess, and you are the one paying it.

The shop lands eleven grand over on labour and nobody can say whether framing or the roof steel is where it went.

Asking a man to file his own hours against a stage guarantees the quickest option every time, so he is never asked — the job is the only question put to him, and only when his crew genuinely has more than one running.

A crew logs four days onto a build that was offered to them and quietly handed to somebody else, and the first anyone knows is a labour line on the wrong job.

An offered job showing up as somewhere to clock in would let hours pile onto work nobody agreed to take, and you would find out from a cost report. Accepting it on his own phone means he chose it, and nobody can say the job went to the wrong number.

The schedule advances at lunchtime and eight hours of framing turn up on the roof steel line, where nobody ordered them.

Reading the stage at clock-out — or worse, at report time — makes yesterday's hours change meaning every time the schedule moves, and a labour number that shifts on its own is a number nobody defends in a meeting. The cost of freezing it is that you cannot re-cut history later. That is the trade, and it is worth making.

Six weeks on, an hour gets questioned and all anyone has is two people's memory of where the truck was parked that morning.

Recording both ends means an odd one — a clock-in half a county from the site — stands out on its own, while every ordinary day passes with nobody asked to prove they were there. The trade is that a man who wanted to cheat it still can; what he cannot do is cheat it invisibly.

One tap on a single bar of signal, nothing happens, he taps again — and a fortnight later payroll is sixteen hours heavy with nothing to say where they came from.

The second tap is not met with an error a man on a roof would have to interpret; he is handed back the shift he already has, so the screen tells him the truth and there is nothing left to reconcile. The database carries a standing index on open shifts per person, so the question am I on the clock is one lookup rather than a scan somebody could talk themselves out of trusting.

A man leaves his phone charging in the truck at seven, and by the time anyone notices, his morning has to be reconstructed by two people who disagree about it.

The entry lands on whatever job the leader himself is clocked into, because a crew works one site at a time and he is standing on it — an exception that costs two taps is one that happens in here, rather than by text message where nothing is kept.

The missing Tuesday surfaces on payday, in front of the man it costs, with nothing left to check it against.

Deliberately a roster and not a report — the leader gets the smallest thing that lets him correct this morning's mistake while it is still this morning's, which is the only point where a fix costs a conversation instead of trust.

Somebody drives home still clocked in on Thursday and the nineteen-hour day turns up when payroll is already running.

Twelve hours is counted from the clock-in itself rather than the crew's planned finish, so a four-in-the-morning start is not flagged for being early. Who hears it is a setting rather than a hard-coded chain — it goes to the owners and project managers unless the company points it at the leader standing on the same slab, who can settle it with one question to the man himself.

Every second Thursday somebody stands at your elbow asking what he is owed, and you put down what you were doing to go and count it for him.

He is reading the office's own entries rather than a copy of them, so payday starts from one sheet instead of two — and a man who can see a day missing from his own week chases it while the crew that worked it still remembers.

Two hours columns, and by March the payroll report and the P&L disagree by a hundred and forty hours with nobody able to say which is the real one.

A second hours field would have meant touching every report that reads the first one, and every one of those is a place two numbers could quietly diverge. Deriving into the existing column costs you the ability to hold a typed figure and a punched figure side by side; the column says which one it is holding, and that was judged enough.

A disputed Thursday, and nobody can say whether that eight hours came off a clock, off a leader's phone, or off a sub's own paperwork.

One universal hours form would have been less code, and it would have thrown away the single distinction that makes a labour dispute answerable. Keeping the source on every row costs a column and buys you an answer to where did this eight hours come from that nobody has to reconstruct.