Consistency Without an Answer

A decision has two properties your institutions often treat as synonyms: consistency and answerability. Consistency means similar cases receive similar results. Answerability means some person or office can explain the rule, own its application, and reconsider it. Automation can increase the first while quietly removing the second.

An automated scheduler calculates four closing shifts from coverage targets, labor costs, recorded availability, and whatever penalty its designers assigned to fatigue. The worker may correct an input, but usually cannot dispute how fatigue was valued. The manager sees the output but lacks authority to alter it; the designers control the rule but never meet the worker. Human judgment has not disappeared. It has been divided so that no human at the point of consequence experiences themselves as the author.

This division does not require a scheme to evade blame. It emerges naturally from scale, specialization, and software purchased from outsiders. One team defines an objective, another configures a product, a third administers cases, and a fourth handles complaints. Each participant can truthfully describe only a small task. The institution collectively decides, while its members individually report that the decision came from elsewhere.

Human discretion is not the lost paradise. A manager empowered to make exceptions can reward relatives, punish enemies, yield to charm, or grant relief mainly to those fluent in complaint. Appeals distribute advantage to the persistent, connected, and eloquent. Automation can make treatment fairer by resisting these pressures and applying one rule across thousands of cases.

But consistency does not establish that the rule is defensible. A procedure can treat every case alike while optimizing for intolerable fatigue, excluding a relevant circumstance, or reproducing an old bias with perfect regularity. Nor does answerability require unlimited exceptions. A rule can remain stable while still having a named owner, an intelligible rationale, and a channel through which its premise—not merely its inputs—can be contested.

The consequential divide is therefore not between human and machine judgment. Human judgment remains in both; the difference is where it becomes reachable. In an answerable system, disagreement can travel back to the level where criteria were chosen. In an unanswerable one, objections are translated into data corrections, support tickets, and instructions to try the portal again. The machine’s most valuable output is sometimes not the decision, but the closed conversation.