A ledger is an integrity instrument

The Balance That Cannot Tell the Truth

In this synthetic signed ledger, delete one side of a journal entry and the zero-sum check exposes an exact residual. Restore it and add a stipulated fictitious sale, and the check passes while reported profit rises. Then apply four modelled evidence constraints to balanced alternatives to see what this internal arithmetic cannot decide.

In this model, debits are positive and credits are negative. Each declared journal-entry column must add to zero. That rule is the whole machine below, and every amount is an exact integer in an unnamed synthetic currency.

Layer one

Make the books fail. Then make them lie.

Use the buttons in order. The table is rebuilt from the signed postings after every change.

Trial-balance residual

…

absolute signed sum

Accounting equation residual

…

assets − capital − profit

Reported profit

…

…

Computing the ledger…

Each column is one entry. + is a debit, − is a credit. A green grand sum says only that the postings fit the rule.

Layer two

The null space is larger than the mistake

A one-sided omission is easy prey. The sophisticated objection is that this makes the algebra a clerical curiosity. But the same equation that catches the omission also defines every change it cannot catch:

1ᵀΔ = 0

The trial balance does not catalogue fraud. Its null space contains every internally balanced alternative, including legitimate changes, classification errors, and invented events. In this toy, we know which ledger is false because we defined the underlying events. Real evidence has to do that work outside the equation.

Generate balanced alternatives

Move the amount. Four materially different changes are generated from equal debit and credit postings. The known-events ledger is included as the control.

The dispatched alternative uses an exact cost fraction stored with the ledger.

…

Add constraints from outside the ledger

Each selected record tests a different reported field against the known synthetic events.

The customer confirmation and dispatch record are then fabricated to fit the fictitious credit sale. This is a model of corrupted evidence, not a claim that these procedures ordinarily fail.

What the selected evidence can still see

…

…

    The check

    Independent route

    Route A multiplies and sums signed posting columns. Route B separately totals debits and credits. Their live residuals are … and ….

    Exact anchor

    The predeclared special cases are one surviving debit and an equal debit-credit pair. They evaluate to … and ….

    Control on the control

    The checker changes the fictitious credit by one unit. It must reject the entry. …

    Evidence sweep

    The verifier tries every subset of the four records. With all ordinary records, … candidates remain; with the two-record collusive trail, … remains.

    Free choices: the chart of accounts, five base entries, generated amounts, cost fraction, candidate alternatives, and four evidence fields are synthetic. Uncertainty: this is a formal core, not a complete accounting system or an audit plan. The checks prove the displayed arithmetic and the elimination logic. They cannot prove that any real transaction occurred.

    Where the instrument stops

    Bookkeeping integrity is not occurrence, authorization, valuation, classification, or timing

    A zero residual is necessary in this signed toy. It is not evidence that the events happened or were recorded in the right account and period. Real systems also include compound entries, control accounts, closing entries, contra accounts, multiple currencies, rounding, and software controls. The accounting equation is shown separately from the raw signed sum because they are related here, but they are not the same test.

    The error-detecting-code analogy has a boundary

    The zero-sum invariant detects some damaged postings, much like a parity check. A ledger is not normally encoded to correct errors, and the invariant contains none of the economic meaning of its account labels. Calling the null space a catalogue of fraud would be false.

    External evidence is not an oracle

    PCAOB AS 2401 says collusion can cause a properly performing auditor to find false evidence persuasive, that absolute assurance is unattainable, and that procedures effective for error may be ineffective for fraud. The collusion switch makes that limit operable in a tiny declared model. It does not promise what any single audit procedure will detect.

    Sources and method. David Ellerman, On Double-Entry Bookkeeping: The Mathematical Treatment, provides a broader group-of-differences treatment of double entry. The signed matrix here is a narrower pedagogical construction. The audit-evidence limit comes from PCAOB AS 2401, Consideration of Fraud in a Financial Statement Audit. Full sources, choices, and rerun instructions are in research/the-balance-that-cannot-tell-the-truth/README.md.