lumen
lumen is a fictional saas company of about 90 engineers across eight teams. the product is a collaborative document editor with a sign-up funnel, admin back-office, billing, and an ml-powered auto-summarisation feature. below is the topology — eight teams laid out with their interaction lines already drawn.
your job is to classify each team by type: stream-aligned, platform, enabling, or complicated-subsystem. click a node to read what the team does, pick a type from the side panel, repeat for all eight, then check the topology. the interaction lines are clues — look at who talks to whom, and in what mode.
classify lumen's teams
read each team's role, then pick the type that fits. a correct topology will have a mix: several stream-aligned, a small number of platforms, one or two specialists, and a temporary coach or two.
what to look for
- the majority should be stream-aligned. lumen has four of them: sign-up, editor, billing, admin — each owning a user-visible flow with its own metrics. if you counted fewer than four, you probably misclassified a stream-aligned team as a platform.
- platforms are serving many. both platform and search have incoming x-as-a-service arrows from multiple consumers. that's the tell — platforms have portfolios. if nobody's consuming you, you're not a platform.
- complicated-subsystems are specialist islands. ml-core publishes a single api (scoring) to one consumer (editor). the rest of the org doesn't reach into it, and the team doesn't spread out. that's the shape.
- enabling teams show up on a timer. cloud-coach is connected via facilitating (dashed line) to one team, for one project. the interaction mode is the giveaway. facilitating + time-bound + coaching vocabulary = enabling.
the reverse conway reading
now look at the topology as an architecture diagram, not an org chart. the platform team's services are consumed by four stream-aligned teams — so lumen's production architecture has a platform layer with a well-defined api, consumed by four independent vertical slices. ml-core lives as a distinct service with one consumer. search is a shared infrastructure. cloud-coach leaves no runtime trace — it's pure knowledge transfer.
that architecture was chosen by this team graph, not by the code. if lumen's leadership tried to ship a different architecture — say, merging search into the editor service — they'd be fighting Conway's law. the reverse manoeuvre says: first change the teams. then the code follows.
next: the common ways teams misapply the framework — and what each failure mode looks like up close.