A use case such as placing an order touches several things: it loads objects from a repository, asks the domain to change them, saves the result and reports back. A service function holds that orchestration, so web handlers, command-line tools and tests all call the same code.

from dataclasses import dataclass

@dataclass
class Product:
    sku: str
    stock: int

    def restock(self, qty):
        if qty <= 0:
            raise ValueError(qty)
        self.stock += qty

def restock(repo, sku, qty):
    product = repo[sku]
    product.restock(qty)
    repo[sku] = product
    return product.stock

repo = {"MUG": Product("MUG", 1)}
print(restock(repo, "MUG", 4))

The service takes its repository as an argument and returns plain values. It does not parse requests, print or open connections; those belong to its callers. The rules live on the domain objects, so the service reads like a short list of steps.

Test a service with a fake repository built in the test. If the service works with that, the seam is real.

Write Product.reserve(qty) raising OutOfStock when qty is more than the stock. Write place_order(repo, sku, qty) that raises ValueError for a quantity below 1, UnknownProduct when repo.get(sku) is None, and otherwise reserves, saves with repo.add(product) and returns the remaining stock.

The explanation opens here once every check passes.