Government
Decisions have to stay understandable to people who did not make them.
A government technology decision has to remain defensible years after the people who made it have moved on. That is the constraint that distinguishes public-sector work: not the technology, but the requirement that the reasoning survive the people, the administration and the budget cycle.
Most programmes discover this in the wrong order — asked to justify a decision taken four years ago, reconstructing the reasoning from fragments and the memory of whoever is still in post.
Where the cost shows up
Why provenance is the deliverable
Provenance converts an argument into a record. A programme asked to justify a decision years later is usually reconstructing reasoning from memory and fragments. Where provenance was captured as the decisions were made, the evidence, the reasoning and the reviewers are already attached — and the conversation shifts from whether the decision can be defended to whether it should now change.
Evidence
A federally funded biomedical research programme established audit-ready provenance for every research decision and translated scientific impact into the fiscal accountability language federal evaluators required. The programme survived federal fiscal review; its institutional knowledge became the defence against elimination. Summarised in the case library, with detail available on request.
What makes government different to work in
Accountability is external and adversarial by design. Work has to survive scrutiny by people who were not in the room and are not inclined to be generous.
Authority is distributed across programme, agency and oversight bodies, and a governance model satisfying only one of them is incomplete.
Continuity horizons exceed staff tenure, so anything held only in people is a risk rather than an asset.
Common questions
Can institutional knowledge survive an administration change?
Only if it was captured as an organizational asset rather than held by individuals. That is the object of knowledge intelligence: making programme reasoning queryable by the people who arrive next.
How should a programme govern AI agents?
Define what they may do, on what information, under whose approval, and with what record — and make those controls properties of the system rather than policy documents. A control that depends on someone remembering it will not hold under deadline. See enterprise AI governance.