Ninety-four fields called Shipment
Shipment has 94 fields. It sits at the centre of a freight forwarder's codebase, the company that quotes a price for moving other firms' goods, collects the crates, consolidates them into containers and clears them through customs at the other end. Sales added quotedPrice and chargeableWeight. The warehouse added palletCount, dockSlot and measuredGrossWeight. Customs added tariffCodes, declaredValue and netWeight. One status field carries values from all of them: QUOTED, ON_DOCK, HELD_BY_CUSTOMS, CLEARED.
Every team says "shipment", and every team means something else. To sales, a shipment is a priced promise: a route, an estimated weight and a price valid for two weeks. To the warehouse, it is pallets on a dock, measured and labelled. To customs, it is a declaration: goods classified line by line, with a value and a country of origin. The word sounds the same in every meeting, so the code treats them as one thing.
A number that looks right
The first cost was quiet. Operations needed a truck-planning report and found three weight fields side by side. chargeableWeight was always filled in, from the moment of the quote, so the report used it. Chargeable weight is the greater of the actual and the volumetric weight, which is what the quote's price is based on, and for light, bulky goods it is far above what the crates weigh. For months the truck plans recorded loads much heavier than anything the trucks carried. Every number in the report was a real weight; it was the wrong one for the question.
The second cost was loud. Part of one client's goods was going into a bonded warehouse, to be declared later on a separate declaration, so customs added a parentShipmentId and turned one Shipment row into two copies. The dock screen, which summed pallets across rows, showed twice the pallets, and the invoicing job billed the client twice. Each team's change was reasonable inside its own meaning of the word. The class was wrong because it tried to be right about every team at once, so every change needed every team's review.
A boundary where the meaning changes
The fix is not a better Shipment. It is to stop having one. A Bounded Context is an explicit boundary in the software, usually a module or a service with its own code and its own data, inside which one model holds and each of its terms has one meaning. The forwarder draws three contexts, Quoting, Warehouse and Customs, each with its own model of the goods in its own words. Quoting has a Quote, with a route, a price and the chargeable weight. The Warehouse keeps the word Shipment, now meaning only pallets on a dock with a measured weight. Customs has a Declaration, with tariff lines and the weights the declaration asks for, and can split one in two without touching the Warehouse's pallets. The truck report now reads the Warehouse's measured weight, the only weight that context has.
Inside each context, the tactical pieces (Aggregate, Entity, Value Object) belong to that context: the Customs Declaration aggregate is not passed to Quoting code as an object. How the business itself divides, and how the shared vocabulary of a context is built with the people who know the work, is the subject of Subdomains & ubiquitous language.
The boundary is about the model, not about deployment. A context can be a package inside one deployable application or one or more services; what matters is that code outside it reaches into its model only through a relationship the teams chose. Microservices vs monolith weighs the deployment choice.
Each context owns its own model of the same goods. Here, between contexts only data in an agreed shape crosses, and each of those arrows is a relationship the context map names.
Where to draw the line
Too few boundaries bring back the 94-field class; too many turn every business operation into a conversation between contexts. A word that needs a different definition depending on who is speaking is the first hint, and the forwarder had one in "shipment". Ownership is another: a context works best with one team accountable for its model, because a model two teams change for their own reasons drifts back into the shared class. So is the pace of change. Customs rules move when the authorities change tariff codes or declaration forms, quoting moves whenever sales launches a pricing scheme, and apart, each ships on its own schedule. If two candidate contexts could do nothing useful without calling each other on every request, they are probably one.
Meeting at the edge
The contexts still work together: a booked quote has to reach the Warehouse, measured goods have to reach Customs. Each of those crossings is a relationship between two models with a name, from sharing a small piece of the model to one side adopting the other's, and Context map: how bounded contexts relate walks through each kind. When the receiving side translates the foreign model on the way in, that code is an Anti-Corruption Layer; what the forwarder must not do again is pass one model object across the edge without having chosen to, which is how the 94-field class began.
Split the forwarder on a Grid
In the Bounded Context Challenge, draw at least two contexts, each connected to the aggregates and entities that belong to it, and let the language change across the boundary on purpose: the same goods under a different name in each context, as Quote and Declaration are here, or one name whose definition changes. Label the edge between the contexts with how they meet. The last thing on the Grid should be a sentence saying why one model would not serve both, naming the word that split.