This page walks the bundled Northstar Parts sample end to end and reports exactly what the workbench produced: twenty-eight findings, six of them blocking, from two twelve-column files of four data rows each. The two source files are generated by the product's own sample module and parse back through its own importer, so you can download them and reproduce every number below.

The two files

The accepted baseline, orders-2026-08.csv, has the header row order_id, customer_id, order_date, region, quantity, unit_price, currency, priority, warehouse_code, promised_date, sales_rep, notes. The candidate, orders-2026-09.csv, has order_id, client_id, order_timestamp, quantity, region, unit_price, currency, fulfillment_center, promised_date, notes, discount_code, source_system. Both are comma delimited and UTF-8. Download them here: accepted baseline CSV, proposed candidate CSV, and a complete review checkpoint you can restore in the workbench to skip straight to the findings.

What the comparison reported

Six findings are blocking. Three baseline columns are absent from the candidate: order_date, priority, and sales_rep. One is a type shift: quantity was observed as integer in August and as number in September, because the candidate carries 2.5. The remaining two blockers belong to the finance workbook consumer, which requires order_date (absent after mappings) and allows only integer for quantity.

Twenty-two findings are marked for review rather than blocking. Two are committed renames: customer_id to client_id, and warehouse_code to fulfillment_center. Four columns moved position, including region from position four to five and notes from twelve to ten. Two blank rates rose: currency by 75.0 percentage points, promised_date by 25.0. Two widths grew: quantity from one character to three, and notes from fifteen to fifty-seven. Three columns were added: order_timestamp, discount_code, and source_system. The last eight are the warehouse label merge refusing unlisted columns, because that consumer was described as intolerant of extras.

How to read the result

The interesting finding is not the missing order_date, which anyone would spot. It is the pair around quantity: a silent integer-to-number shift plus a width increase, in a column that a positional workbook reads without complaint and then rounds. The second is currency going three quarters blank while remaining present, which no column-name diff would report at all.

Method and limits

Every number above comes from four data rows per file. A blank rate of 75.0 percent means three of four sampled rows were blank, not that three quarters of the producer's real month is blank. Type inference is observational in the same way. Use the sample to learn how the findings read, and use your own month of data to make a decision.

Do this in the app

Open the workbench, choose Open sample in the Overview view, then move to Findings. Filter by blocking severity to see the six above, then dispose of each with an outcome and note before freezing in the Packet view.

Related reading