All open work
BUILD 01 / Data QA Workbench

A dirty order export becomes an auditable handoff.

A seven-row synthetic order export is deliberately small enough to inspect by eye, but contains the same classes of failure that break imports and monthly reports: a blank amount, duplicate key, exact duplicate row, casing drift, outer whitespace, and invalid date text.

Demonstration built from synthetic data. It proves the inspection method and export structure; it does not claim a customer result.
01 / INPUT CONTRACT

Synthetic order export

7 data rows · 5 fields · CSV · no personal or customer data

  • record_id is intended to be unique
  • amount should be numeric and required
  • status should use one controlled vocabulary
  • order_date should be an ISO date
02 / METHOD
01

Preserve the source

The original text is not changed. Every issue cites the row or field where it was observed.

02

Profile structure

The audit measures fill rate, uniqueness, dominant type, whitespace, row width, exact duplicates, and the selected key.

03

Separate signal from correction

Potential problems enter an issue register. A human decides whether an apparent outlier or exception is legitimate.

04

Create handoff evidence

The result exports to Markdown for review, CSV for issue tracking, and JSON for a repeatable record.

03 / OBSERVED OUTPUT
6

structural signals

Across missingness, duplicate key, duplicate row, whitespace, casing/type interpretation, and invalid date text.

2

high-severity blockers

The repeated business key and exact duplicate can double-count or overwrite records.

1

missing required value

The blank amount needs recovery, explicit null handling, or exclusion before aggregation.

0

source uploads

The public checker performs the review inside the browser tab.

DECISIONS FROM THE BUILD
  1. Block dashboard refresh until the repeated record_id is resolved.
  2. Preserve the blank amount as an exception; do not convert it silently to zero.
  3. Normalize outer whitespace and status casing only in a derived clean file.
  4. Keep the source register and audit output beside the clean delivery.
05 / RELEASE CHECK

What had to be true.

  • Input can be inspected row by row
  • Every finding is reproducible
  • No correction overwrites the source
  • The audit states what it cannot verify
TRY THE PUBLIC EDITION

Run the method on your own input.

Open the free tool See the complete paid edition