Coherence: meaning survives many handoffs

The failure that doesn't look like one

An account manager asks an assistant for a renewal summary. The assistant reads three systems and finds one word with three senses: in the CRM, customer is a segment (mid-market); in billing, a contract tier (enterprise, from a legacy bundle); in the help center, an audience tag. It writes: "Enterprise customer, mid-market segment, renewing in Q4." The renewal email goes out with enterprise language. The summary is stored. Next week another agent builds the quarterly deck from that memory, and "enterprise" is now a fact about the account that no system ever asserted.

Notice what the failure was not. Billing and the CRM were not wrong. The failure is that nothing could say which sense was in force for this handoff, so the assistant blended them, and three hops later the blend looked like knowledge.

What coherence actually is

Coherence is not making customer mean one thing everywhere. It is knowing there are several senses, which one this handoff needs, and who is allowed to change each, and carrying that across every join. Looking a term up does not mean fetching the one true definition. It means retrieving the binding that applies at this join, and when two authorities conflict, recording the conflict instead of inventing a third meaning.

Three depths, named from the work

• The request — prompt, context, harness. Built from nothing each time. Fails loudly, in an afternoon.

• The estate — data engineering and content engineering, peers who rarely share a meeting. Everything the organization already wrote down, modeled, argued over, and settled. Fails quietly, over years.

• The ground — semantic engineering. What the terms mean, and who says so. Not a fourth ring around the outside. The floor. Nothing above it is more settled than it is.

A very good harness over an unsettled customer does not give you a slightly worse answer. It gives you a confident, fast, well-formatted wrong answer, every time. Leverage is symmetric.

The definitions already exist

The CRM, the ERP, the CMS, the DXP, dbt and the warehouse, the vocabularies, the taxonomies, the translation memories: each holds a definition, and each is expert in something the others cannot hold. Every attempt to collapse them into one master model has produced a very expensive fifth place where the definition lives, slightly differently again. The work is to connect the models, not merge them, and to keep the mappings between them alive as each changes. That is the unoccupied job, and it has to be done whether or not anyone buys anything.

A weeklong taste test

Pick customer. Write the four definitions side by side, each in the words its own system uses. Name who may change each one. Publish the mapping somewhere a machine can read it. Point the harness at it. Watch the next twenty retrievals. Some resolve cleanly. Some hit a join where two authorities disagree, and the right behavior is to record the contest and hand it to the owner. Done, for the first stretch, is a small set of contested terms reconciled and bound, not "coherence stood up."

Coherence needs a practice: funded, chartered, centralizing the pattern-making while implementation stays where it lives, with a named owner for each join where meaning is known to leak. Not an owner of the CRM and an owner of the help center. An owner of customer, wherever it travels.

Read the full essay

The essay walks the request, the estate, and the ground in full, with figures; grants the industry's own attempts (grounding, citations, evals, data contracts) and shows why they stand on the same floor; and closes on why coherence, as an emergent outcome, compounds unusually well.