The engine needs one line-level orders file: one row per order line, with a SKU, a date, a quantity, and a sales amount. Costs, fees, refunds, discounts, currency and marketplace columns make the diagnosis sharper, but the four basics are enough to start.
Amazon (Seller Central)
Reports → Payments → Date Range Report (transactional) is the richest option — it carries fees and refunds. An Orders report works too. Amazon's convention of storing fees and refunds as negative numbers is recognised and normalised automatically — you will see a disclosure in the report's assumptions, not corrupted totals.
eBay (Seller Hub)
Seller Hub → Reports → Orders, or the Transaction report if you want fee detail per line.
Shopify
Analytics → Reports → Sales by product variant SKU, or an Orders export. If your theme names columns differently, that is fine — see below.
ERP / accounting systems
Any line-level sales export works: SKU, date, quantity, amount. Filter to a single currency and at most two marketplaces — beyond that the dataset routes to a Custom Autopsy instead of the automated report.
Column names don't need to match
"Item Number", "Qty Shipped", "Order Total" — common export names are recognised automatically, and anything unrecognised takes a minute to map by hand: see fixing column mapping.
Export up to the last 6 months — that is the standard analysis window. Files up to 50 MB each, five files, 100 MB total, CSV or XLSX.