A coverage gap is the portion of a population that something does not reach. The word covers two quite different problems, and conflating them is common enough to be worth separating.
A control coverage gap is an estate that a protective measure does not reach. Endpoints with no agent. Systems outside the backup schedule. Accounts without step-up authentication.
A measurement coverage gap is an estate that your measurement does not reach. Business units not yet onboarded. Cloud accounts no collector has credentials for. Subsidiaries whose data arrives as an annual spreadsheet or not at all.
The first is a security problem and gets attention. The second is a knowledge problem and usually gets none.
Because they hide the first kind.
An endpoint the agent failed to install on is a control gap that the agent console cannot report, because the console only knows about endpoints where the agent installed. The gap is invisible from inside the tool that created it.
Every coverage ratio has this property to some degree. A figure computed over the population your tooling knows about will always look better than one computed over the population that exists. See ratio metric.
Compare independent views of the same population. Network discovery against inventory. Cloud provider APIs against the CMDB. HR headcount against directory accounts. The disagreement is the estimate. See source of truth.
Track which organizational units contribute data at all, separately from what the data says. A metric drawn from four of seven regions is a regional metric being presented as an enterprise one.
Watch for sources that report but report thinly. Partial failure is harder to spot than total failure and produces the same optimistic bias.
Publish measurement coverage next to the metrics that depend on it. One line, showing what share of the known population the figure was computed over.
Where the true population cannot be established, say so rather than reporting a number that implies a boundary it does not have. A metric with a stated limit is usable. One without is a claim about an estate nobody has surveyed. See data completeness.
From the blog