Do not overwrite the original export during cleanup

Preserve a baseline before cleanup

Preserve the original result so you can explain changes without repeating extraction or guessing what the tool produced.

ParseMill exports reviewed invoices to Excel with separate summaries, line items, sources and review status. You can import an XLSX or CSV template and save a client setup. Run a real invoice through to the final workbook and check it in the application that receives your work.

A naming convention is useful, but it does not prevent accidental edits by itself. Use the access controls or file-protection practices appropriate to your workspace, and verify the saved baseline when reproducibility matters.

Use a simple file-version convention

The names are examples, not a required software feature. Include source revision and delivery date in a manifest if filenames would become unwieldy. Avoid labels such as final-final that provide no reliable relationship between artifacts.

W3C: Data on the Web Best Practices recommends version indicators and history. Use explicit links between artifacts so a later reviewer can identify which original export underlies a reviewed copy.

Record changes as they happen

1. Save the original export and record its source identity, extraction configuration and creation time. Keep a hash when byte-level identity is useful.
2. Create a reviewed working copy. Log corrections by stable record key, field, old value, new value and source evidence.
3. Keep derived formulas and normalization rules in a separately labeled layer. Record the reviewed input version they use.
4. Before handoff, reopen the delivered artifact and confirm that its version and change record match. Preserve superseded copies according to the organization’s retention rules.

W3C: Provenance model primer models derivation activities between entities. The original, reviewed and analysis files can therefore be recorded as distinct artifacts with explicit relationships.

Avoid losing the origin of a change

A cleanup command that trims text, changes types or fills blanks can affect many rows. Record its rule and scope, not just “cleaned data.” If the operation changes a source string, keep the raw field available so the transformation can be inspected.

Do not overwrite the original file with a corrected export under the same name. If an error is discovered later, the baseline helps determine whether the extraction produced it or a subsequent edit introduced it.

Verify the revision chain

For the fixture, reviewed copy 002 must identify original export 001, and analysis 001 must identify reviewed copy 002. A recipient should be able to reconstruct the sequence without guessing from modification timestamps. The change record should explain unresolved exceptions as well as corrected cells, including the intended use that remains blocked by each exception.

Use this version convention when cleaning ParseMill exports. It keeps manual work reversible and gives your team a clear path from the source document to the final register or analysis.