Nothing is retyped between the Journal and the week.

The contract

One direction, no manual steps, no summary that can drift.

Journal rows are entered once. Weekly aggregation reads them by formula. The scorecard reads the aggregation. The gate reads equity. Authority reads all of it. At no point does a human move a number from one surface to another, and no surface holds a stored copy of a figure another surface computes. The chain runs one way and every link is a function rather than a transcription, which means the state of the whole system is fully determined by the Journal rows plus the formulas — and the formulas do not change week to week. That is a stronger property than it sounds: it means the entire weekly output is reproducible from the row data alone, by anyone, including an auditor who was not present for any of it.

What it eliminates

Transcription error is not reduced. It is made impossible.

In a system where weekly figures are typed into a summary tab, a wrong number has two possible origins: the underlying trade was recorded wrongly, or the summary was transcribed wrongly. Those require different investigations and the second is nearly undetectable, because a transcribed figure is internally plausible and matches nothing it can be checked against. Removing the manual step removes the branch. Investigating a wrong weekly figure becomes a single question with a bounded search space — which row — rather than a hunt across two categories of cause. The difference shows up most sharply under time pressure, when a two-category search is usually abandoned in favour of accepting the figure.

Why this makes the audit real

Reconciliation only means something if the rows determine everything.

The Saturday reconciliation checks rows against the broker and the capture sheet. That check is worth performing precisely because the rows are the only input — verifying them verifies the entire week's output by construction. If any weekly figure could also have been typed independently, reconciling the rows would leave the summary unverified and the audit would be partial in a way nothing flags. The no-retyping contract is therefore not a tidiness rule sitting beside the reconciliation procedure; it is the property that gives the procedure its reach.

The cost the contract imposes

It is genuinely less convenient, and the inconvenience is the mechanism.

There are weeks where a figure is known and typing it would take a moment while correcting the underlying row takes longer — a trade closed in two pieces, an adjustment the sheet does not model cleanly. The contract forbids the shortcut, which means the operator must fix the representation rather than the result. That is more work, and it is the point: the shortcut leaves the week looking correct while the evidence base stays wrong, and every later question asked of that period returns the wrong answer confidently. The shortcut also compounds: a period patched once tends to be patched again, because the underlying representation gap was never closed.

• Two possible error origins collapse to one, and the surviving one is searchable.
• Verifying the rows verifies the outputs by construction, not by inspection.
• The forbidden shortcut fixes the number and leaves the evidence wrong.

How to tell it is being honoured

The check is structural and takes a minute.

Click a weekly figure and see whether it is a formula or a value. Any weekly cell holding a constant is a breach of the contract, regardless of whether the constant happens to be right today — it will not update when the row beneath it is corrected on Saturday, so the correction will silently fail to propagate. This is worth checking after any period of heavy manual editing, because the usual way the contract breaks is not a decision to break it but a formula overwritten during a hurried fix and never restored.

The key idea

The contract is what makes one audit cover the whole system.

Governance stacks tend to accumulate verification points, one per surface, each partial and each easy to skip. This one has a single verification point because it has a single input. Reconcile the rows and every figure above them is verified, not because each was inspected but because none of them could have diverged. That is an unusual property to hold and a fragile one — it survives only as long as nothing is ever typed twice.