The Assessment Gap — Summary
Keiko is building something. Not on contract. On her own time, in the apartment where the folder on her laptop, the one labeled “what the form doesn’t ask,” now contains annotated assessments for eleven deployments across six cities. On a whiteboard propped against the wall is a grid she has been revising for three weeks. One axis is the spectrum from obvious improvements to unacknowledged losses. The other axis is something she does not have a name for yet. She has been calling it “relational load.”
The standard deployment assessment is four pages. It was designed by competent people solving a real problem: municipal decision-makers need numbers to justify autonomous deployments to elected officials and budget committees. Efficiency gains, cost reductions, service reliability, safety outcomes. Each metric is legitimate. Each captures something real. The form measures what the system does. Keiko’s problem is with what the form was not designed to do: measure what the system replaced.
The gap is not an error in the framework. It is a property of what the framework was built to see. The relational function was invisible before automation. Byproducts do not have baselines. They do not appear in before-and-after comparisons. They are visible only through the absence they create after they are gone, and by then the deployment has been approved.
Keiko’s framework asks three questions before deployment: What is the human doing besides the nominal function? Who depends on that function? What happens to them when it ends? The knowledge exists in the people doing the work. Sandra could describe the cups and the orchid and the three flags. Tomás could describe the notebook. The knowledge does not exist in any system a deployment assessment currently consults.
She has shared early versions with program managers. The first response is interest: the framework names something they have felt but could not articulate within the existing structure. The second response is the practical question: what do we do with the information? If the framework identifies high relational load and the deployment is otherwise justified by the numbers, what is the program manager supposed to do? Delay the deployment? Add a companion services contract? Each option has costs that come out of the savings the deployment was supposed to produce.
The resistance is structural, not attitudinal. Knowing creates an obligation the system is not designed to fulfill. A framework that says “this deployment will eliminate the only regular human contact for forty-three homebound recipients” and the deployment proceeds anyway has produced a record of a harm the institution chose to accept. The standard assessment’s silence is protective.
The framework cannot replace what is lost. Even a perfect assessment cannot undo the structural condition: the relational function was never designed, and building the infrastructure to carry it is not the work of a deployment assessment. It is the work of a society that has decided human contact is important enough to design for rather than rely on as a byproduct. Keiko’s framework is a diagnostic instrument applied to a design failure. It can describe the failure with precision. It cannot fix it.