Trace a bank spreadsheet row back to its PDF page

Add source references before sorting or merging bank rows. A spreadsheet row number is not a stable transaction identity: it changes when you insert or reorder data. Keep an extraction ID plus the original PDF location.

Assign IDs and source locations before transformations

1. Give each source file a neutral identifier such as DOC-01. Record page number and a local row or visual locator without placing real account details in public examples.

2. Assign a stable extraction ID before transformations. Keep it unchanged when the description or amount is corrected.

3. Preserve source IDs on merged text fragments and record every contributing location when a transaction spans pages.

Fictional example

Fictional row E-0042 maps to DOC-01, page 3, transaction 7. After sorting it appears on worksheet row 12, but E-0042 and the source locator remain unchanged. Its description continues on page 4: add that second location rather than claiming page 3 alone contains the whole transaction.

Make the locator unambiguous

PDF page labels and physical positions may differ; keep both distinct in the source register. For this proposed register, use the physical PDF position counted from 1 as the primary page field; keep a printed label in a separate field. In a fictional file with a cover, physical page 3 might be printed “2”. Record both rather than silently subtracting one from every reference.

Define “transaction 7” as the seventh transaction beginning on that page, excluding headings and continuation lines. For E-0042, record DOC-01/v1; physical page 3; start 7, plus physical page 4; continuation at top. Add a short visual anchor such as the adjacent date and reference. If two payments share a date and amount, their positions still distinguish them. If the location is uncertain, mark it unresolved instead of assigning the nearest row.

A positional index alone establishes neither a source mapping nor a history of approved corrections. Freeze the initial mapping with the extraction snapshot. Rebuilding an index after a new extraction must not silently reuse the old identities. For a small file, a manually maintained source register is sufficient. Have another reader locate E-0042 from the register alone and check its start and continuation; needing your verbal explanation means the locator needs work.

Keep the transaction and its locations in separate tables

For the fictional E-0042, use this small register. The fields are a proposed review convention, not metadata that every extractor supplies.

Field	Fictional value	Why it stays with the row
Source file and version	DOC-01/v1	Identifies the exact preserved document
Extraction snapshot	RUN-A	Distinguishes this extraction from a later attempt
Stable row ID	E-0042	Identifies the transaction within RUN-A
First source location	Physical page 3; transaction start 7	Locates the date and amount
Additional source location	Physical page 4; continuation at top	Locates the rest of the description
Printed page numbers	2 and 3	Helps compare paper numbering without replacing physical positions

Use DOC-01/v1 + RUN-A + E-0042 as the complete working key. “E-0042” alone may recur in another file. Keep one transaction row and two location records linked by that key, rather than duplicating the transaction for each page. A CSV with only one source-page column needs a separate accompanying register or a clearly documented way to retain both locations; do not quietly drop page 4.

Preserve the old value when correcting an extracted cell

Suppose this fictional row was imported as −83.10 EUR but the PDF clearly prints −38.10 EUR. Preserve the raw value, put −38.10 in the reviewed field and record “digits transposed” with the source location. The ID stays unchanged. The corrected total rises by 45.00 EUR because −38.10 − (−83.10) = +45.00. Check that this is the entire change to the transaction total; an unrelated change points to another edit or an incorrect range.

A new extraction from DOC-01/v1 gets its own snapshot, such as RUN-B. Compare its rows against RUN-A with an explicit correspondence table; do not give its 42nd row the old identity merely because the ordinal matches. If the PDF itself changes, create v2 and recheck page positions. A file fingerprint can help distinguish bytes but does not certify who issued the document or that its contents are correct.

Test the complete path after saving and exporting

Find the original amount from a sorted transaction, then find every description fragment from a joined transaction. Finally, inspect a corrected row and its previous value. Every exported transaction should have a usable key and every location record should point to an existing transaction. If either check fails, repair the register before delivering it; do not replace a missing locator with the current worksheet row number. Keep the register with the exact export version so a later correction does not leave readers following obsolete references.

Checks and limits

Test a random row after sorting and a corrected row after saving. An index column is a technical starting point, not proof that the source association is correct. Keep a stable file-version reference if the PDF is replaced. This trace supports review; it does not certify authenticity.

Keep the original file unchanged and record any unresolved item before using the table.

Related checks

Locate the source cell before comparing or correcting its value. Bank PDF copy-paste in the wrong order: diagnose before sorting · Overlapping bank exports: review duplicates without losing payments · Diagnose a bank PDF before choosing a table extraction tool.

BankStatementScan is in preparation. The site offers an access request and a fictional example; it does not currently convert your bank files. See the planned service.

Sources checked on 16 October 2026

• W3C — Consistent PDF page numbering.
• W3C — Model for Tabular Data and Metadata on the Web.