The politics of definitions

Data architecture is never purely technical because definitions allocate visibility and responsibility. A “completed case,” a “valid report,” an “active therapy,” or an “incident” is not only a field value; it is an institutional classification. Bowker and Star showed how classification systems embed histories, negotiations, and exclusions. The platform cannot eliminate this politics, but it can make it governable.

Definitions should therefore be versioned artefacts with owners, rationale, validity intervals, and known consequences. Let a metric definition be

where $f_v$ is the computation, ${\Sigma }_v$ the required input contract, $[t_0,t_1)$ its validity period, and $o_v$ its accountable owner. When a rule changes, historical values need not be silently overwritten. The organisation can compute results under the definition valid at the time or deliberately restate history under the new definition.

This practice replaces the myth of a timeless “single truth” with a more rigorous principle: a single governed account of each definition, including its versions and limits.