Case record / Commerce marketplace
Commerce Operations Console Blueprint
Commerce Operations Console Blueprint: a conceptual commerce marketplace blueprint covering Marketplace Development, Amazon Clone and Admin Dashboard on Next.js, Node.js and PostgreSQL. No client outcome is claimed until verified.
Conceptual reference only; no client result is claimed.

Content updated
Evidence status: Blueprint. Public outcome evidence is not attached to this record. This is a planning example, not a completed client project or a measured commercial result. See the disclaimer.
Context / delivery read
Marketplace operations break at the seller and refund edges.
Seller onboarding, catalogue quality review, order states and refunds are modelled before loyalty or personalisation, because those edges decide whether the first orders can be fulfilled and supported.
Recorded services
Workflow and deliverables
Seller onboarding and SKU quality review
Order, cancellation and refund states
Operator console for disputes and payouts
Catalogue rules before merchandising features

Commerce Operations Console Blueprint looks at a multi-vendor store from the back office outward: the decisions an operator makes every day about sellers, listings, orders, returns and payouts. It is offered as a planning reference for early marketplace teams. It does not describe a trading store, and no sales, conversion or seller numbers are claimed.
The storefront in this plan is deliberately plain. Personalised recommendations, loyalty points and seller advertising wait until real orders exist to justify them.
Letting a seller in
A seller applies with business details, tax information and the categories they want to sell in. Bank details are collected through Stripe's own onboarding for connected accounts, which keeps them off the platform's servers. The operator approves each category separately rather than the seller as a whole, so a shop cleared to sell books is not automatically allowed to sell supplements or electronics.
Listings worth reviewing
A listing needs a title, images, price, stock, a shipping profile and whatever fields its category demands, such as sizes for clothing. Listings from new sellers go to a review queue; established sellers publish straight away and are spot-checked. Each rejection carries a reason from a fixed list plus a short note, so the seller knows exactly what to fix rather than guessing and resubmitting.
One basket, several sellers
A shopper pays once, and checkout splits the basket into one order per seller. Each seller order moves on its own: paid, accepted by the seller, shipped with tracking, delivered, completed. If a seller does not accept an order within the time the operator sets, it is cancelled and refunded automatically, so a shopper is never left waiting on an unresponsive shop.
Returns as a written decision tree
Three questions settle most returns. Is the item still inside the return window? Is its category returnable at all? Is the reason a seller fault, such as damage or a wrong item, or a change of mind? The answers decide who pays for return shipping and whether the refund waits until the item is back with the seller. Writing the tree down before the build keeps most return arguments out of the support inbox, because shopper, seller and operator can all see the same rule.
Paying sellers from a ledger
Money from each seller order is held until the return window closes and is then released minus commission. Every movement, whether a sale, a commission, a refund or an adjustment, is written as a row in a ledger table, and the seller's statement is built from those rows. A refund that arrives after a payout becomes a negative balance against the next payout instead of a request to send money back. Because the statement can be rebuilt from the ledger at any time, it can also be checked line by line against Stripe's records.
The console, queue by queue
The admin dashboard is organised as queues, each with an owning role: seller applications, listing reviews, flagged orders, return disputes and payouts on hold. Reports stay close to decisions: order value, orders by state, return rate by category and by seller, and commission earned. Suspending a seller hides their listings immediately but lets open orders finish, so shoppers who have already paid are not caught in the middle.
What the first release leaves out
Seller advertising, loyalty schemes, a recommendation engine, multiple currencies and platform-run fulfilment are all listed with the signal that would justify each, for example enough repeat shoppers to make loyalty worth funding.
The scoping document here is the list of admin decisions itself. Screens are drawn from that list, not the other way round, and an estimate built on it is far less likely to drift.
