The Second System

After this lesson, the reader should understand:

• 01Distinguish the documented decision system from the operating one.
• 02Explain why divergence is structural rather than a compliance failure.
• 03State what a policy document can and cannot evidence about practice.

The clause everyone follows differently

Three managers apply the same emergency-leave clause. Their approval rates differ by a factor of four. Which of them is not following policy?

The clause: "Emergency leave may be granted where the employee demonstrates a genuine operational need, having regard to the employee's prior use of emergency leave in the current year." Two undefined terms are doing the actual work — "operational need" and "prior usage" — and over a year of requests, three managers resolve that ambiguity three different ways. One treats "operational need" as an evidentiary bar: a hospital admission, a burst pipe, something documentable. She approves 18% of requests. One treats "prior usage" as the operative test: a first request gets the benefit of the doubt almost automatically, a fourth request gets scrutinised regardless of the reason given. He approves 45%. One reads the clause as a low bar clearing the way for trust in his team's judgement, and approves 72% — four times the first manager's rate, from the identical sentence. All three are compliant. The company's annual HR audit samples a set of decisions from each manager, checks each one against the clause, and clears all three, because nothing in the clause rules out any of these three readings — an evidentiary bar, a usage-history test, and a trust default are all "having regard to operational needs and prior usage." The audit that would surface the divergence does not exist, because it would have to compare manager to manager rather than manager to document, and the document has no canonical meaning to check either manager against. This is what absorption looks like while it is still working^[1]. Each manager's judgement is quietly supplying the definition the clause withheld, and the institution never has to resolve which definition is correct, because no single case ever forces the three readings into contact with each other. An employee whose request is denied by the strict manager does not get to appeal to the permissive manager's standard; the divergence is invisible from inside any one decision, and costless to the institution as long as a person is the one making the call. Then the company automates emergency-leave triage: a system reads the request, checks the same two facts, and returns approve or escalate. The system cannot hold three interpretations open at once. Someone building it chooses — usually without framing it as a choice at all — which manager's pattern the training data or the threshold actually encodes, and that becomes the institution's emergency-leave policy the moment it ships: uniform, immediate, applied to every future case, decided by whoever picked the threshold rather than by anyone with the authority to set leave policy. The three-way divergence that coexisted safely for years does not get resolved by automation. It gets executed, as one specific number, at scale. The strongest objection is that this is not a defect at all. Discretion at the front line is how institutions fit general rules to particular people; the literature on street-level bureaucracy has shown for decades that the people applying a policy effectively make it^[2], and that this is often what lets the policy work. The managers know their teams. A clause loose enough to allow three readings may be loose on purpose. Concede most of this. The argument here is not that discretion should be abolished. It is that discretion should be visible — exercised knowingly, within a declared range, by someone entitled to exercise it — and that automation does not preserve discretion but freezes one instance of it. Structuring discretion, rather than eliminating it, is an old aim of administrative law^[3], and the operating system is where it has to start. One caution on the evidence itself. A fourfold difference in approval rates does not, on its own, prove three different readings. The three teams may receive different requests: a warehouse team on shift rotation and an office team with flexible hours will not generate the same emergencies. Before concluding that the managers disagree about the clause, one has to compare like requests with like — which is exactly the problem the next lectures take up.

• Is the operating system a failure of drafting, or a property of institutions?
• If practice has diverged consistently for a decade, which system is the policy?
• Who is accountable for a rule nobody ever wrote down?
• Can an institution be said to consent to a rule it has never seen stated?
• When the system that replaces these three managers ships, whose approval rate does it inherit — and who actually decided that?
• Which divergences in your institution are discretion working as intended, and which are drift nobody chose?

No responses yet. Be the first to answer one of the questions above.

Locate one divergence you already know about.

1. 01Name a decision where you know practice differs from the written rule.
2. 02Write the rule as documented.
3. 03Write the rule as actually applied.
4. 04Identify what the document leaves undefined that practice has silently defined.
5. 05Note who would have to approve the practised version for it to become policy.
6. 06Check whether the difference could be explained by different cases rather than different readings.