A forensic accountant rarely gets a bank's own export. What arrives from counsel is what the other side produced: scans of paper statements, photocopies of photocopies, a few months of one account and a single page of another, sometimes a statement printed from a portal and rescanned before it was handed over. The question is not whether that stack is tidy. It is what can be proven from it, and what has to be asked for.
A bank statement carries its own proof. The printed opening balance, plus every credit, minus every debit, equals the printed closing balance, to the cent; where the statement prints a running balance, every line must agree with the line before it. That holds for a scan as much as for a download, because it is a property of the figures, not of the file.
So a scanned page is straightened, its table rules removed, and read at 300 dpi with the larger OCR model. The rows read are then proven against the printed balances of the same page. If they tie, the statement is proven, and every row is cited to the file, page and line it came from. If they do not, the exhibit shows the first row where the running balance breaks and by how much. Pages the reader could not use are listed as unread, never skipped in silence, and a statement with unread pages is reported as read in part rather than reconciled.
Months chain the same way whatever the file type: each statement of an account must open at the previous one's closing balance. A missing month is a gap, a shared day is an overlap, a balance that does not carry over is a break, and each is listed with the two exhibit numbers. A production that skips the month a transfer went out shows up here as a gap, with the dates.
A digital PDF carries a structure: the software that wrote it, its creation and modification dates, its fonts, its save history, and a text layer that can be compared, figure by figure, with what the page prints. Those checks are how a file edited after the bank wrote it shows itself.
A printed and rescanned statement carries none of that. Its producer is the scanner, its dates are the scan's, and there is no text layer to compare with the print. The exhibit does not pretend otherwise. Provenance on such a file is reported as not assessable, and the proof rests on what remains: the arithmetic and the month chain. What can still be measured on a scan is the geometry of the rows themselves, word baselines, letter heights and row edges taken from the scan's own noise, so that a line sitting off its row is listed with its text. Thresholds for that are set on real scanned statements so that a clean page produces no note, and the register says which pages were measured and which had too few rows to test.
The practical consequence for discovery: ask for the digital download whenever one exists. A scan of a bank's PDF adds nothing the PDF does not already prove, and it erases the structure that would have told you whether the PDF was touched.
Every file in a matter is declared with a source when it is added: a bank subpoena production or a witnessed download, the client, the subject, or the opposing party. A bank-sourced file is taken as the bank's own record on the reviewer's declaration, and only the arithmetic and the chain are applied to it. A file from the subject or the opposing party gets the full authenticity section: producer and dates, files written in one sitting, arithmetic that does not hold, debits posted on days the network does not settle, text painted over text. Each note says what broke and what to verify, and never what it proves.
Declaring sources honestly matters more than it looks. A subject's statement declared as bank-sourced is not examined; a bank production declared as the subject's will collect notes that mean nothing. The exhibit prints the declared source of every file in the register, so the declaration itself is part of the record.
Redact inside the PDF, by drawing boxes in the file as it came from the bank. Printing a statement, blacking out lines and rescanning it turns a file with provenance into one without, and the exhibit will say so. The bank's own PDF with a redaction layer keeps its producer, its dates and its text layer, and the checks that depend on them still run.
The exhibit reports what the arithmetic, the calendar and the file itself show, and what to ask the bank for. Whether a statement is altered, what a transfer was for, and who a counterparty is are the examiner's findings, made by reading the page, comparing against a genuine specimen and asking the bank. The exhibit's basis of preparation says so, and its last section lists what was not checked, which on a scanned production is the list worth reading first.
The register and the proof are free on screen, no account needed for the first matter. Start a matter · Read the sample exhibit