THE SHORT ANSWER
Audit a mapping by following representative values from source to destination. Check field meaning, data type, units, blank handling and identity preservation. Save the transformation and expected result so another operator can explain the output without reconstructing the original implementation.
Document meaning before column names
Two columns with similar names may represent different facts. A source price might include tax, use minor units or apply only to one market. A group ID may describe a supplier collection rather than a variant family. Ask what each value means in the selling system.
Create a mapping record with source field, destination field, transformation, fallback and owner. Mark unresolved assumptions explicitly. An empty owner or an unexplained constant is a useful audit finding, even if the export currently passes a basic parser.
Review transformations and empty values
Inspect trimming, case changes, numeric conversion, currency formatting and text assembly. Pay particular attention to identifiers that look numeric. A conversion that removes leading zeros can break identity while leaving every row syntactically valid.
Define null, empty and zero separately. Decide when a fallback is appropriate and when missing information must remain an exception. Avoid silently replacing an absent price, material or identifier with a plausible-looking value that nobody supplied.
Build a representative mapping sample
Include a simple product, a multi-option variant, a discounted product, an unavailable item and a record with optional fields missing. Add source-specific edge cases such as accented text, long model references and embedded delimiters.
For each sample, show the original values and expected output. Review the result against the destination’s current contract, not just yesterday’s file. A working historical mapping can become inappropriate when a source or receiving interface changes.
Turn findings into controlled revisions
Prioritize identity and commercial-fact issues before cosmetic formatting. Preview the affected scope of a rule change and compare it with the last usable release. Keep mapping revisions attached to release records so a regression can be traced quickly.
After correcting the mapping, rerun the sample and inspect a broader output slice where necessary. Then monitor the next scheduled import. A one-off manual edit does not fix a recurring transformation, and the next run may reintroduce the original problem.
Write a mapping contract before adding rules
A mapping contract records how a value moves between the source and a specific target. Include the source field, target field, operation, unknown-value behavior and a small example. Name the person responsible for approving a merchant-wide default. This makes a rule inspectable when someone else needs to change it.
Start with the least surprising operation that preserves meaning. A direct copy may be enough for a factual title. An explicit lookup may be needed for an availability value. A merchant setting can supply verified seller context. Keep a missing product fact unresolved until the authoritative source provides it; a formatting rule cannot establish that fact.
| Source value | Operation | Exception to test |
|---|---|---|
| Item ID 000042 | Copy as a string | Spreadsheet or serializer removes leading zeros |
| Known product title | Trim surrounding whitespace | Variant selection disappears from the title |
| Availability code | Use a reviewed target-value lookup | Source introduces an unrecognized status |
| Amount and currency | Serialize using the target’s money format | Currency is missing or amount belongs to a different market |
| Verified merchant name | Apply an approved merchant setting | Marketplace offer has a different seller |
| Missing material | Create a source-evidence task | A rule inserts an unsupported material |
Keep a small regression pack for every mapping change
Save input and expected output for difficult records, not just the first clean row. Include a quoted comma, an embedded newline, a Unicode product name, a leading-zero ID, an empty optional value and a variant with a different image. Add a case whenever a real mapping failure is found. The pack becomes a practical history of the mistakes the workflow must stop repeating.
When you rename a source column or change a transformation, run the pack against the new mapping and compare changed values. Review unexpected differences before applying the rule to the full source. Keep the rule version alongside the source snapshot and release so an operator can explain which transformation created a value.
Source snapshot: supplier-2026-10-31
Mapping version: catalog-v3
Input item ID: "000042"
Expected item ID: "000042"
Changed rule: availability lookup
Expected behavior for an unknown code: create an exception
Reviewer: assigned workspace editorILLUSTRATIVE WORKED EXAMPLE
Example: fix an availability transformation without changing identity
Illustrative example: a source price is stored as 3499 minor units. The audit finds that the export emits 3499 USD instead of 34.99 USD. The team corrects the documented unit conversion, checks several prices and verifies that the next recurring run uses the revised mapping.
Your practical checklist
- Record each field’s meaning, type and owner.
- Review units, fallbacks and blank behavior.
- Preserve identifiers as text.
- Save expected output for representative source cases.
- Attach mapping revisions to releases and retest scheduled runs.
Common questions
Can matching column names automate the whole audit?
They can suggest a mapping, but they do not prove semantic equivalence, correct units or appropriate source ownership.
How many sample records are enough?
Choose cases that exercise your actual source and transformation risks. A small deliberate sample is more informative than many identical ordinary rows.
What is the difference between mapping and enrichment?
Mapping transfers or transforms known source values for a target. Enrichment adds a missing or more useful value, which needs its own evidence and review. Keep those operations visible separately.
Sources & scope
Primary references below show their latest recorded check date. Examples are illustrative, not customer results. Use the current destination documentation for implementation decisions.