APPAREL RETURN DIAGNOSTIC

Find the return pattern a raw count can hide.

Load one supported sales-events CSV and one supported return-events CSV. This workbench compares ordinary returned-unit counts with exact 90-day SKU cohort rates, while keeping refunds, handling costs and recoveries separate.

Your files stay in this browser. Uploaded bytes are not sent to a server, logged, attached to analytics, or fetched from another service. Original inputs appear only in the ZIP you explicitly download.

Set the cohort window

The interval is inclusive at start and exclusive at end: [start, end). Version 1 requires exactly 90 days and USD.

Start (UTC)End (UTC, exactly 90 days later)Currency

Choose the two canonical CSVs

Sales events CSVNo file selectedReturn events CSVNo file selected

Canonical columns and limits

sales.csv

event_id,order_id,line_id,sku,occurred_at,currency,quantity,gross_amount,unit_cost,reversal_of

returns.csv

event_id,return_id,order_id,line_id,sku,occurred_at,currency,quantity,refund_amount,reason,handling_cost,recovery_amount,reversal_of

UTF-8 CSV; 2 MiB per file; 2,000 combined events; 100 cohort SKUs; 40,000 combined cells; 4,096 bytes per cell. This is a documented format—not a claim that arbitrary Shopify exports work.

Choose both files and an interval.

SOURCE-LINKED RESULT

What the cohort changes

Refunds are cash returned, not “lost profit.” Handling cost is only an explicitly supplied processing cost. Recovery is only explicitly supplied realized value such as resale proceeds or carrier reimbursement. Product unit cost is never counted as destroyed value; missing cost fields remain unknown and excluded.

SKU patterns: ordinary count vs cohort rate

SKU	Sold	Returned	Rate	Count rank → rate rank	Sample	Sources

Reason patterns

Reason	Returned	SKUs	Sample	Sources

Validation and interpretation

I reviewed this result and its source fingerprints.

What this does not claim

It does not read a store account, infer missing sales, identify profit loss, promise savings, or assert what a merchant’s native report contains. Out-of-cohort and orphan returns are visibly excluded instead of being assigned an invented denominator. Low samples are marked. The comparison here is strictly between ordinary aggregate counts and the computed SKU/reason/cohort views.