TL;DR
- Organizations usually record what was decided and who approved it, but not the assumptions the decision depended on or the conditions under which it should be reopened.
- Without those two items, later review defaults to outcome judgment: good results are read as good decisions and bad results as bad ones, which confuses luck with reasoning.
- A decision that carries no expiry condition stays in force by inertia, long after the facts that justified it have moved.
- A reviewable decision is one whose premises, rejected alternatives, and revisit triggers were written down while the uncertainty was still live.
The Review That Cannot Be Done
Consider a familiar scene. A regional distributor changes how it routes deliveries, and eight months later margins are lower than expected. A review meeting is called. The records show the date of the change, the approver, and a short summary. They do not show what the team believed about fuel costs, order volume, or driver availability at the time, and they do not show which other routes were considered and set aside.
The review therefore has only two things to work with: the outcome, and memory. Memory is reconstructive. Participants recall the decision as more or less certain depending on how it turned out. The outcome is real but ambiguous, since a sound decision can end badly and a careless one can end well.
The central claim of this article is narrow and arguable: the main obstacle to reviewing decisions is not poor judgment or missing data at review time. It is that decision records are designed to prove a decision was made, not to let anyone judge whether it was made well. Those are different purposes, and most governance routines serve only the first.
What a Record Usually Contains, and What It Omits
The typical decision record is an approval artifact. It exists so that authority can be shown: a ticket, a signed memo, a line in meeting minutes. This is useful for accountability in the narrow sense of who answered for it. It is much less useful for accountability in the broader sense of whether the reasoning held up.
Four elements are commonly absent:
- The premises. The beliefs about the world the decision rested on, stated as claims that could later be checked.
- The rejected alternatives. Without them, a reviewer cannot tell whether the chosen path was best or merely first.
- The confidence level. A decision made at near-certainty and one made as a reasonable bet look identical in a record that omits this.
- The revisit condition. The observable event that would indicate the decision should be reopened.
The absence of the last item is the most consequential and the least discussed. A decision with no revisit condition has no expiry date. It remains in force until someone happens to question it, and questioning a settled decision carries social cost that revisiting a flagged one does not.
Outcome Bias Fills the Vacuum
When premises are missing, reviewers substitute the one piece of information they do have. Behavioral research has long described this tendency as outcome bias, the habit of rating a decision’s quality by its result. The pattern is easy to recognize in organizations without any study at all: a project that succeeds is remembered as well planned, and the same plan, had it failed, would be remembered as reckless.
The practical effect is distortion in both directions.
| Situation | What the outcome-only review concludes | What a premise-based review could find |
|---|---|---|
| Good result, weak reasoning | The decision was sound; repeat the approach | The result depended on a condition that will not recur |
| Bad result, sound reasoning | The decision was a mistake; avoid the approach | A reasonable bet lost; the reasoning remains valid |
| Good result, sound reasoning | Confirmed, but for unstated reasons | Confirmed, with the premises available for reuse |
| Bad result, weak reasoning | Mistake, cause unclear | The specific premise that failed |
Outcome-only review gets the right verdict in the third and fourth rows, but even there it cannot say why. In the first two rows it teaches the wrong lesson. An organization reviewing this way becomes more confident in patterns that were partly luck, which is a mechanism for what WBA has described as epistemic debt: “The gap between what an organization believes to be true and what is actually true, compounded over time by the cost of acting on that gap.”
Decisions Age Even When Nobody Looks
A second failure runs on a different clock. Even a well-made decision rests on conditions that change. A staffing model assumes a certain call mix; a vendor choice assumes a certain price structure; a policy assumes a certain risk environment. The decision does not announce when its premises stop being true. It simply continues to be applied.
This is why the missing expiry matters more than the missing rationale. A rationale tells a later reader why the decision was reasonable then. A revisit condition tells the organization when then has ended. Examples of such conditions are generic and observable: a cost crossing a stated threshold, a volume falling outside a stated range, a dependency changing hands, or a fixed calendar date by which the premises must be re-confirmed.
Calendar-based expiry is the weakest form, since it triggers review whether or not anything changed, but it has one real advantage. It does not depend on someone noticing. Condition-based triggers are sharper but depend on someone watching the metric. Pairing the two covers each one’s weakness, and it addresses the zombie decision: a commitment that persists only because nobody owns the question of whether it still applies.
Why Organizations Skip This
The omission is rarely ignorance. Several structural pressures make it sensible in the short term.
Recording premises exposes the decision to challenge. A stated assumption can be falsified publicly. An unstated one cannot. In environments where being wrong carries penalty, vagueness is protective.
Speed is rewarded at the point of decision, review at no point. The cost of writing down premises is paid immediately by the decision-maker, while the benefit accrues later to someone else.
Ownership of review is unassigned. This echoes the pattern in the correction bottleneck: learning stalls when no one owns the question of why something went wrong. A decision with no reviewer is a decision with no feedback loop.
These pressures are the predictable result of a system that rewards decisiveness and treats reviewability as overhead.
What a Reviewable Decision Looks Like
The argument above points to a small set of properties rather than a heavy process. A reviewable decision is one where, at the moment of choice, the record holds:
- the claim the decision depends on, phrased so it could be shown false;
- at least one alternative that was seriously considered, and the reason it lost;
- a rough statement of confidence, even as a plain “likely” or “a close call”;
- the observable condition, or date, that reopens the question;
- a named owner of that revisit.
This is deliberately modest. It adds minutes to a decision, not days, and it changes what review can accomplish: instead of arguing about outcomes and recollections, reviewers can ask whether the premises held, which is a question with an answer.
The same logic applies where automated systems take part in decisions. A recommendation produced by a model, without a recorded basis, has the same defect as an unrecorded human judgment, only at higher volume. Reviewability is a property of the record, not of who or what made the call.
The Underlying Point
Decisions are bets made under uncertainty, and a bet can only be evaluated against the odds that were believed at the time. Organizations that keep only the result discard the one piece of evidence that separates good judgment from good fortune. The governance question is therefore not whether decisions are approved, but whether they are left in a condition that allows an honest second look.
Research questions on this pattern are welcome through Inquiries.