The cost of the indefensible decision

The reason unaccountable decisions accumulate is that they are cheap. Making a consequential determination and keeping no faithful account of how it was reached costs nothing extra at the moment of the decision; in fact it costs slightly less, since capturing the account takes effort the system could skip. So institutions skip it, not out of recklessness but out of the ordinary economics of doing the minimum that works. The decision fires, the outcome lands, the system moves on, and nothing appears to have gone wrong. This is the trap in its purest form: the cost of the indefensible decision is zero right up until it is the only number that matters, and by then the decision is in the past and the account cannot be created retroactively. What follows is the anatomy of that cost — not to dramatize it, but to show its structure, because the structure is what makes it worth pricing in advance.

The cost is not the wrong decision — it is the unanswerable one

Begin with a distinction that is easy to lose. The exposure here is not the cost of making a bad decision. Bad decisions are a normal, priced, insurable feature of any operation; every institution makes some, and mature institutions have machinery for absorbing them. The exposure is the cost of making a decision — good or bad — that the institution cannot subsequently account for. These are different failures with different price tags, and conflating them is why the risk is chronically underweighted. A defensible bad decision is a manageable event: you show what you knew, what rule you applied, why the outcome followed, and you argue about whether the judgment was reasonable on the facts as they stood. An indefensible decision forecloses that entire conversation. You cannot argue the judgment was reasonable, because you cannot reconstruct what the judgment actually was. The institution is left defending not a decision but a blank.

This is the crux. When a challenge arrives — a complaint, a suit, an inquiry — the challenger's first move is not to prove the decision was wrong. It is to ask the institution to explain it. And the moment the institution cannot produce a faithful account of what evidence was consulted, what rules were in force, and why the outcome followed, the character of the dispute changes. The absence of an account is not a neutral fact. It is read, by every party with standing, as evidence.

An institution that cannot explain its decision is not treated as having no answer. It is treated as having something to hide — and the law, the regulator, and the public all resolve that ambiguity against it.

Three conversions, each one enlarging the loss

The cost compounds through a series of conversions, each of which takes an ordinary, survivable event and makes it worse specifically because no account exists.

The first conversion is from dispute to admission. In an adversarial proceeding, the party that cannot produce the relevant record does not merely lose the point the record would have addressed; it invites an adverse inference. The gap in the account becomes an argument for the other side — the institution had the ability to preserve the evidence, chose not to, and now cannot explain itself, from which a fact-finder is entitled to draw the least favorable conclusion. A defensible decision costs you the argument you actually have. An indefensible one hands your opponent an argument you cannot rebut, because the thing that would rebut it does not exist. The dispute you might have won on the merits becomes an admission you cannot escape on the record.

The second conversion is from inquiry to finding. When a regulator examines an automated decision, the question is frequently not "was this outcome correct?" but "can you demonstrate the process that produced it?" These are separate tests, and the second is often the one with teeth, because a regulator can penalize an institution for being unable to account for its decisions even where no individual outcome is proven wrong. An inability to reconstruct is itself the violation. So the institution that kept no account does not merely risk being found to have decided badly; it risks being found to have operated a system it could not answer for — a finding that attaches to the whole apparatus, not to a single case, and that tends to invite the kind of ongoing supervision that is far more expensive than any one penalty.

The third conversion is from incident to pattern. A single indefensible decision is a problem. The discovery that the institution systematically cannot account for a whole class of decisions is a different order of problem, because it converts one plaintiff into a class, one inquiry into a program, and one bad headline into a narrative about the institution's basic competence. This is the conversion that turns a local cost into an estate-wide liability. If the reason a decision cannot be explained is that the system was never built to explain any of them, then every decision that system made is potentially indefensible, and the exposure is not the one case in front of you but the entire population of cases behind it. The absence of a record does not produce an isolated loss. It produces a correlated one.

The reputational cost sits on top, and it is the one that lingers

Layered over the legal and regulatory costs is a reputational cost with a distinct shape. When an institution cannot explain a consequential decision that affected a real person, the story that reaches the public is not "the institution made a defensible call that some disagree with." It is "the institution did something to someone and cannot say why." That story is uniquely damaging because it maps onto the exact fear that automated decision-making already provokes — the fear of being subjected to a machine that no one can question. An institution caught unable to account for its decisions does not merely take a hit; it confirms the public's worst prior about the whole category, and it does so in a way that no settlement can retract, because the damaging fact is not the outcome but the silence.

The asymmetry is the argument

Set the two numbers side by side. On one side: the cost of capturing a faithful account of a decision at the moment it is made — a marginal, bounded, predictable operating expense, incurred once per decision, whether or not the decision is ever challenged. On the other side: the cost of being unable to produce that account when challenged — an adverse inference, a regulatory finding, a class-wide exposure, a reputational wound, incurred at the worst possible time, on the challenger's schedule, and irreversibly, because the moment to have captured the account has passed. The first cost is small and certain. The second is large and probabilistic, but its probability is not zero and its severity is not bounded.

This asymmetry is the entire case for treating decision accountability as protection rather than overhead, and it is the reason the "we'll deal with it if it comes up" posture is not a neutral deferral but an active bet — a bet that no challenge will land on a decision the institution cannot defend, placed by an institution that, by construction, cannot know which of its decisions those are. The indefensible decision costs nothing to make and everything to be asked about, and the only point at which the institution can change that ratio is before the decision, not after. That is not a reason to record everything at any price. It is a reason to price the gap honestly, because the gap is where the loss lives, and the gap is cheapest to close at exactly the moment it feels least necessary to.

— Dispatches · Summit Cognitive