When Decisions Leave Their Origin

A steering group settles a delivery date in March. The date is not arbitrary. It was chosen because a certification window closes in the autumn, and missing it costs a year. That reasoning is discussed for twenty minutes and then compressed into a line in the minutes: Delivery confirmed for 14 September.

In June, an engineering lead facing a supplier delay moves the date by three weeks. She has the authority to move dates. She is doing exactly what a competent lead should do with a slipping supplier. Nothing in front of her says the fourteenth was load-bearing.

No single careless act explains the outcome. The steering group omitted a constraint that felt obvious inside the room. The engineering lead exercised authority that looked entirely valid from where she stood. The failure was distributed across what got recorded, what authority was delegated, and what the receiver could reasonably infer.

The date survived the transfer. The reason did not.

At first glance, delegation appears to preserve decisions. A decision is made, authority is assigned, and the decision crosses a team boundary, a reporting line, a system interface, or simply time. It arrives, and work proceeds. The assumption is that what arrives is what was sent.

But delegation does not transport decisions. It transports representations.

What crosses the boundary is an operational representation of the decision: a summary, an instruction, a rationale, a parameter, a record of intent. It rarely carries the full context in which the original commitment became binding. The alternatives that were rejected. The constraint everyone in the room knew and nobody wrote down. The judgment about what mattered and what could give.

The representation carries what was selected for expression. Plenty gets left out because it seemed obvious, or was politically awkward, or would have taken a page.

So on the receiving side, reconstruction is unavoidable. Somebody — a person, a team, a system — has to rebuild a working decision from what they were given. That is not a failure. It is interpretation, and it is what makes delegation useful in the first place, because the receiver knows things the originator never did.

Some degree of interpretation is unavoidable, and most of it is operationally insignificant. It becomes consequential when the receiver has to infer a constraint, an authority boundary or a trade-off that the representation did not preserve.

Often it does not. A decision travels intact when the representation covers what the task needs, when the receiving context is standardised, when the authority boundary is explicit, and when the receiver can pick up a phone and ask. Plenty of delegation works precisely this way, which is why organisations scale at all.

Three conditions make that likely, and they compound.

Context loss. The encoding compresses away a constraint the receiver could not have guessed at. September the fourteenth looked like a date. It was a certification window wearing a date's clothes.

Authority ambiguity. The receiver has to work out whether they are meant to execute, to adapt, or to reopen the question. That determination is itself a decision, made at precisely the moment continuity is most fragile, and usually made alone.

Weak repair. The action is hard to reverse, and clarification is slow, expensive, or socially awkward enough that inferring is cheaper than asking.

Any one of these is usually survivable. Together they turn interpretation into drift.

Drift rarely announces itself. It shows up as symptoms. Decisions reopen, because the propagated representation no longer binds anyone downstream. Escalations multiply, because people facing ambiguous authority quite rationally push the problem upward. Ownership blurs because nobody can say whether the downstream version is still execution of the original decision or a new decision needing fresh authority. Review expands when nobody has specified what each review is meant to establish, so different reviewers test different assumptions and none of them produces a basis for closure.

These get attributed to poor communication or to human inconsistency. Everyone involved is behaving sensibly inside their own environment, working from what they were given. The divergence comes from what was never specified across the boundary.

Which is why apparently settled decisions come back months later in altered form. The commitment did not survive the trip. It survived as a sequence of locally coherent reconstructions, each defensible on its own.

Decision architecture exists to govern that. Not to prevent reconstruction, which would make delegation worthless, but to specify what must survive it.

That turns out to be a more useful question than "was the decision clear?" Every delegated decision has parts that must hold and parts that should be adapted by whoever is closer to the work.

None of which is new. Military mission command formalised a version of it long ago: preserve intent and the binding constraints, delegate execution to the person closest to changing conditions. It works, and it works because of the conditions around it — shared doctrine, common training, repeated dealings, and a commander whose intent can be interrogated when the situation moves.

The harder case is one organisations increasingly face, where the receiver shares none of that. A vendor with different incentives. A team three time zones away with no common doctrine. A model with no institutional membership at all. Mission command assumes shared doctrine and a route back to intent. Where those are absent, the distinction has to be externalised in the workflow itself: in the representation, in the authority rules, in the states a system will accept, and in whatever repair channel is left.

Naming which parts hold and which may move — and stating what conditions require renewed authority — is the core specification. The rest of the work is making it visible where people act: in the states a system will accept, in who is named accountable, in what a review is for. An organisation that has done this can tell a team adapting sensibly from a decision quietly coming apart. An organisation that has not will treat both the same way, with another review.

Communication theory has thought hard about how understanding gets built and repaired, and I am not proposing a replacement for it. Decision architecture asks a narrower question: how interpretation connects to authority, to action, to review, and to consequence. What may be adapted. What must not be. What sends it back.

Most organisations do not describe their coordination problems in these terms. They describe symptoms: decisions that will not hold, escalation chains that expand, coordination that slows despite capable people.

Decisions move. They change as they move. That much is ordinary and mostly harmless.

The trouble starts when nobody specified which parts were allowed to.

---

Sources and influences

Mission command, or Auftragstaktik, for the separation of binding intent from delegated execution. Incomplete-contract theory, particularly Oliver Hart on residual control rights, for who decides in situations a contract could not anticipate. Lucy Suchman, Plans and Situated Actions, for plans as resources for action rather than specifications of it. Herbert Clark on grounding, for how shared understanding gets built and repaired rather than transmitted.