Batch quality: trace each issue to a row and rule
Make dataset issues traceable to a batch, row, field and rule. A GO Sport industry reference with fictional quality notes for JSON or CSV.
Mirakl’s GO Sport case describes a marketplace team manually checking incomplete or inconsistent product data from multiple sellers. It reports adoption of Mirakl Catalog Manager, including a validation-rules engine that identifies invalid product data. After several months of use, the account describes greater error visibility and less manual validation work. It includes commentary from GO Sport’s marketplace manager. The publication does not disclose individual error-row structures or the exact rules applied to a particular batch.
Put the batch and rule beside every issue
For a marketplace-data delivery, PandaBuyNow proposes identifying the dataset version, source-record key and affected field in each quality note. Preserve the observed value, source reference, expected format, rule identifier and rule version. Add an outcome such as passed, failed or not evaluated, using definitions agreed with the recipient.
Distinguish a formatting failure from an unavailable source value or an unresolved product match. A rule may check whether a field contains a decimal number without establishing whether that number accurately describes the product. Keep the checked condition explicit so the result remains interpretable.
Test two fictional error rows
Consider invented batch B-07. Record LINE-08 contains width “wide,” while the agreed width field expects a decimal value. Retain that source text and report a numeric-format issue against LINE-08 and the width field. Record LINE-09 contains width 12 but gives no unit. Report the missing unit separately rather than assigning centimetres.
A batch summary can distinguish records checked, records with issues and the total number of issues. One record may contain several issues. These invented identifiers and values describe no GO Sport or PandaBuyNow dataset.
Carry the result into the delivered file version
When preparing a revised data file, retain the earlier issue reference and identify the replacement record and dataset version. Record which rule version was used for the later result. Leave a field unresolved if the available material still does not support the required interpretation. A changed rule should remain visible alongside any changed result.
Before finalizing the structure, review data excerpts containing a format error, a missing unit and a record that could not be evaluated. For a PandaBuyNow discussion, provide representative product links, the required JSON or CSV fields and the recipient’s expected quality-note format. Agree feasible fields, checked conditions, source references and treatment of unresolved rows in the data-delivery scope.