Context mapping patterns

Contexts do not stand alone, and the arrows between them carry more information than the boxes. The patterns below are Evans’s, and using the names exactly is worth more than it looks — because several of them describe the same arrow and differ only in a political fact that a generic “depends on” would hide.

The landscapes section is about drawing the map. This page is about what the arrows mean.

The nine

Partnership

Two contexts succeed or fail together. The teams coordinate planning, changes go in together, and neither can unilaterally break the other.

Admits: mutual dependence, and a real coordination cost. Choose it when: the two models genuinely have to evolve in the same conversation. In the seed catalog, Risk appetite and Rating are partners — appetite changes prices and prices change appetite. Cost: high. Two teams that cannot release independently. Do not choose it for convenience.

Shared kernel

Two contexts share a subset of the model — code, schema, or both — and both may change it.

Admits: that neither side can be made downstream of the other without lying. Choose it when: duplicating the shared part is worse than the coordination cost. In the seed catalog, Party identity is a shared kernel between Party and Policy lifecycle. Cost: the highest on the map. Every change to the kernel needs both parties. Choose it knowingly and revisit it.

Customer/supplier

Upstream and downstream, where downstream is a customer — it can ask for changes and expect to get them, and its needs enter the upstream team’s planning.

Admits: that the downstream team has real negotiating power. Choose it when: the two teams share an incentive structure — same director, same budget, same objectives. Cost: moderate. Upstream carries a genuine obligation.

Conformist

Upstream and downstream, where downstream has no negotiating power. It takes the upstream model as given and adapts to it.

Admits: powerlessness, honestly. This is the value of the pattern name. Choose it when: it is already true — a vendor package, another company’s API, a system whose team will not change for you. In the seed catalog, Billing is conformist because the SaaS vendor will not change its model and we will not fork it. Cost: low to run, and it lets an alien model into your thinking. Combine with an anticorruption layer if the downstream context is one you care about.

Anticorruption layer

Downstream builds a translation layer so the upstream model cannot leak into it.

Admits: that the upstream model is unsuitable and the downstream one is worth protecting. Choose it when: downstream is core, or legacy vocabulary would do real damage. In the seed catalog, Claims should receive a cover snapshot through such a layer rather than reading live policy rows. Cost: real. The layer is code, it needs an owner, and it grows. When it grows business rules it has quietly become a third context.

Open host service

Upstream publishes a general-purpose interface for many consumers, rather than a bespoke integration each.

Admits: that there are enough downstream contexts to make one interface cheaper than N. Choose it when: three or more consumers want overlapping things. Product catalogue in the seed catalog serves five contexts this way. Cost: upstream must keep a stable public model, which constrains its own refactoring.

Published language

The exchange uses a well-documented shared format — a schema, a standard — that both sides commit to.

Admits: that the interchange format is an asset in its own right, with more than two readers. Choose it when: the payload is consumed by parties you do not control, or has to be explicable to an auditor years later. Rating publishes its price breakdown this way in the seed catalog, because the actuaries and the regulator read it too. Cost: schema governance, versioning, deprecation.

Often paired with open host service: the service is the how, the language is the what.

Separate ways

The two contexts do not integrate. Each does its own thing, including duplicating some data.

Admits: that integration costs more than duplication. Choose it when: the overlap is small and the coupling would be expensive. Genuinely underused — teams integrate reflexively. Cost: duplication, and the reconciliation conversation once a year.

Big ball of mud

No boundary. Models bleed, dependencies are cyclic, nobody can say what is inside.

Admits: reality. This is not a pattern you choose — it is one you record, so that everything around it can be defended. Choose it when: never. Record it when true, wrap it in an anticorruption layer, and do not let it grow.

Choosing honestly

The failure mode with this vocabulary is aspiration: every arrow gets labelled customer/supplier because conformist feels like an admission of defeat.

It is an admission, and that is the point. A map full of aspirational patterns tells you nothing. A map that says “we conform here, we have no leverage, and our core context is exposed to a vendor’s model” tells you exactly where to spend money.

Two questions settle most of them:

Can the downstream team ask for a change and get it? No → conformist. Yes → customer/supplier.

Would we notice within a week if the upstream model changed shape? No → there is probably an anticorruption layer, or there should be.

The map is organisational

The most useful reading of a context map is not technical. Concentrations of conformist arrows show where the organisation has no leverage. Shared kernels show where two teams cannot move independently. A partnership between contexts owned by directors who do not speak is a risk that will surface as a missed date.

None of that is visible on a systems diagram, and all of it is visible here.