Three of them produce something. The fourth is what makes the other three correctable. Each is specified, measured and owned separately, because a fault in one looks identical to a fault in another until there is somewhere between them to take a reading.

Three readings each. Twelve in total, which is the whole diagnostic.
Outbound, paid distribution, content placement, events, and reactivation of the contact database the company already owns and stopped using two years ago.
What the market finds when it goes looking on its own. Search results, and the generated answers that increasingly sit in front of them. This subsystem exists because attention created and never recovered is attention given to whoever ranks.
Everything between an interested person and a qualified opportunity. Site, offer, form, routing, the speed of the first human reply, and the follow-up sequence that either exists or does not.
The readings that make the other three correctable. Tracking, the data model, the definitions behind every field, attribution, and the report somebody actually reads.

Run either one alone for long enough and it produces something that looks like progress and does not compound.
A demand programme with no answer layer creates attention that cannot be recovered. Somebody reads the post, keeps the idea, goes looking three weeks later, and finds a competitor's page inside the answer. The spend produced the interest and handed it over.
An answer layer with no demand programme gets cited on questions nobody is asking yet. Coverage improves, citations appear, and the pipeline does not change, because being the answer to an unasked question is a position rather than a result.
Built together, the demand programme creates the question and the answer layer catches it coming back. Neither one is worth running without the other, which is the whole reason they are specified as separate subsystems and measured apart.
Somebody reads the post, keeps the idea, goes looking three weeks later, and finds a competitor's page inside the answer. The spend produced the interest and handed it over.
Coverage improves, citations appear, and the pipeline does not change, because being the answer to an unasked question is a position rather than a result.
Six lifecycle stages, the two that get worse when a machine is left alone with them, and the four layers of a stack.