A governance programme harvests 1,400 candidate business terms from dashboard titles and column comments and proposes to define and own all of them. Roughly how many terms can an organisation actually keep true, and what does that number change about the plan?

The assumptions, stated

A definition is true only while someone has recently checked it against the SQL that implements it. So the unit of cost is not writing a definition, it is re-attesting one: open the entry, open the model that computes it, confirm the grain, the filter and the time basis still match, confirm the owner still works here. Call that 30 to 45 minutes when nothing is wrong.

Second assumption: some share of terms are contested, meaning two functions hold different definitions that are each correct for their purpose. Contested terms do not consume review time, they consume decision time from someone with the authority to rule.

The arithmetic

• 1,400 terms at an annual re-attestation of 0.6 hours is roughly 840 hours, about 0.5 FTE, before anyone writes a first draft or resolves an argument.
• Initial authoring is heavier: 2 to 3 hours per term including chasing the implementing model. 1,400 terms is on the order of 3,500 hours, roughly 2 FTE-years of work that expires as it is produced.
• Contested share is typically 10% to 15% in an estate this size. That is 140 to 200 rulings a year, each needing a meeting with two functions and an owner who can decide.

A mid-sized enterprise sustains somewhere between 80 and 150 governed terms, with a plausible range of 50 to 250. Roughly one candidate in ten earns governance. The rest get a description in the catalogue, harvested automatically, with nobody asserting it is correct.

Which assumption dominates the error

The contested share, by a wide margin. At 5% contested you can carry close to 300 terms because the cost is nearly all clerical. At 30% contested you can carry 50, because the constraint stops being hours and becomes the availability of a person entitled to overrule a function head. Review hours scale with headcount; decision authority does not.

What the number rules in and out

It rules out "a glossary of everything", which is the plan that produces 1,400 entries with 40% filled in and destroys the credibility of the ones that were right.

It rules in a two-tier estate with an explicit entry rule. Promote a term to governed only if both are true: it appears in a number reported outside the team that produces it, and two teams have already disagreed about it or a regulator names it. Prefer leaving everything else described but ungoverned, with the catalogue saying so on the page, because an unmarked mixture of governed and harvested text fails in the worst way: readers trust the harvested entries and stop trusting the governed ones when one of them turns out to be stale.

When this is the wrong answer

Under roughly 60 people with a single reporting stack, the whole apparatus is overhead. One owner per domain and a weekly 20-minute triage beats a term budget, because the contested set is small enough to hold in a room. The estimate starts mattering when the number of functions that can legitimately disagree exceeds three.

Common weak answers

• "Define them all, then prune." Authoring cost is the large term and is paid up front; pruning recovers none of it, and every stale entry published in the meantime teaches people not to trust the glossary.
• "Automate the definitions with column profiling." Profiling tells you what is in the column, never what the business means by it. It produces descriptions, which is the ungoverned tier.
• "One steward can own 200 terms." They can own the entries. They cannot rule on 200 terms' worth of cross-functional disputes, which is where the time actually goes.