It tiesTies
| Control | Per ledger | Imported | Difference | Records |
|---|---|---|---|---|
| Loans and advances · USD | 5,625,000 | 5,625,000 | 0 | 5 of 5 |
The only state anything is calculated on.
Guide · Import
Tie the tape to the ledger before you measure anything
An expected credit loss over a population that does not tie to the general ledger is a calculation over the wrong population, however careful the arithmetic. It is also the first thing an auditor tests. The required difference is zero, and that is not pedantry — the reason is below.
Why the tolerance is zero and not small
A tolerance on this check is a tolerance for missing exposures, and the amounts that go missing are not random. Rows drop out because of a character in a borrower’s name, a date that would not parse, a currency nobody expected, a facility booked under a product code the mapping has never seen.
Those correlate with exactly the exposures that are unusual — and unusual exposures are where the loss is. A half-per-cent tolerance does not lose half a per cent of the risk; it loses the restructured facility, the foreign-currency loan and the one with the odd name, which between them can carry far more than half a per cent of the allowance.
One exception, and it is bounded: rounding at the minor-unit boundary. A difference below one minor unit per record cannot be a missing exposure, so that much is allowed and nothing else is.
Four ways a population fails to tie
Each of these is run below against the same ledger control total of 5,625,000, and the diagnosis is the one the tool produces.
Rows the reader refusedDoes not tie
| Control | Per ledger | Imported | Difference | Records |
|---|---|---|---|---|
| Loans and advances · USD | 5,625,000 | 5,500,000 | -125,000 | 4 of 5 |
What the tool says: The difference equals the total of the rows that failed to import. Correct those records and the population will tie.
The difference equals the rejected rows exactly, which is the easy case: correct those records and the population ties. Nothing was dropped silently — every row is accepted, rejected with a reason, or accepted with a note.
Short, with nothing rejectedDoes not tie
| Control | Per ledger | Imported | Difference | Records |
|---|---|---|---|---|
| Loans and advances · USD | 5,625,000 | 5,500,000 | -125,000 | 4 of 5 |
What the tool says: The import is short and no rows were rejected, so the extract itself is missing exposures, or the control total is on a different basis — net of allowance, or a wider ledger scope.
The dangerous one. Nothing failed, so nothing drew attention to itself — the extract is missing exposures the ledger has, or the control total is on a different basis. Net of allowance and a wider ledger scope are the two usual culprits, and neither is a data fault.
More than the ledgerDoes not tie
| Control | Per ledger | Imported | Difference | Records |
|---|---|---|---|---|
| Loans and advances · USD | 5,625,000 | 6,075,000 | 450,000 | 6 of 5 |
What the tool says: The import holds 1 more record than the ledger. Check for duplicated rows or an overlapping extract.
A facility counted twice, or two extracts that overlap. Easy to miss because the number looks prudent — a larger population produces a larger allowance, and nobody queries an allowance that is too big until the auditor does.
The one that ties and is still wrong
A currency no control total covers. Every line ties. The total difference is zero. The population is still wrong, and a check that only compared the totals it was handed would report this as clean — which is what most reconciliations do.
900,000 euros nobody asked aboutDoes not tie
| Control | Per ledger | Imported | Difference | Records |
|---|---|---|---|---|
| Loans and advances · USD | 5,625,000 | 5,625,000 | 0 | 5 of 5 |
Uncontrolled: EUR — present in the import and covered by no control total.
Reported as an uncontrolled currency rather than as a difference, because it is not one: it is a part of the population that nothing is checking. Supply a control total for it, or find out why the extract has a currency the ledger scope does not.
What to ask your source system for
A gross carrying amount per ledger account, per currency, and a record count. The count is the part people leave out and it is what catches the duplicate: two extracts that overlap can sum to something plausible while holding a facility twice, and only the count says so.
And ask what basis the total is on. “Gross” means different things in different systems, and a control total taken net of allowance will be short by roughly the allowance — which is the figure being measured, so the reconciliation quietly becomes circular.