Proof without a trial

We picture the record's audience as a tribunal. When we imagine a decision being held to account, we imagine a proceeding: a room, a reviewer, an appellant, a process that draws the record out into the open and walks through it with someone present who built it and can answer for it. That image governs how records get made, and it is almost entirely wrong about how decisions are actually examined — because the overwhelming majority of decisions are never examined at all. They are made, they take effect, and they pass into the past without anyone ever asking them to justify themselves.

This is not a flaw to be corrected. It is the arithmetic of any system that decides at scale. The volume of decisions is enormous and the capacity for review is small, so review is rationed, applied to a vanishing fraction, and the rest go unexamined forever. A record built on the assumption that a trial will eventually arrive to make sense of it is a record built for an event that, for nearly every decision, will not happen. And the rare records that are pulled for review are not a representative sample handled in advance; they are selected after the fact, often years later, by events nobody could have predicted at the moment of decision.

So the design problem inverts. The record cannot depend on the review to complete it, because the review is the exception. It has to be complete on its own — self-sufficient, legible, and capable of standing without a proceeding, a guide, or a living author to fill in what it left out. The question is not will this record survive a trial. The question is can this record explain itself with no trial at all, to someone who simply finds it.

The stranger who arrives later

Picture the actual reader. Not a tribunal — a single stranger, arriving long after the fact, with no relationship to the decision and no access to the people who made it. The author has moved on, or forgotten, or is unreachable; the system has changed underneath; the context that made everything obvious at the time has evaporated. This stranger has the record and nothing else. No one is standing beside them to say what this means is, or you have to understand the situation we were in, or that field is just legacy, ignore it. Whatever the record does not itself contain is simply gone.

A record built for a trial assumes that gap will be filled in the room. It can be terse, it can be allusive, it can lean on shared context, because someone will be there to supply the rest. A record built for the stranger can assume none of that. It has to carry its own evidence, its own active rules, enough state to be replayed, and enough plain legibility that a person with no prior knowledge can reconstruct what happened and judge whether it was sound. The record has to be the witness, because by the time anyone reads it the witnesses are gone.

A record that needs its author present to make sense is not a record. It is a conversation, and the other party has left the room.

This is a higher standard than the trial imposes, and it is the right one precisely because it does not assume help. A record that can stand to an unaided stranger can certainly stand in a proceeding, where there is at least someone to query and context to draw on. The reverse does not hold. A record that only makes sense with its author in the room fails the moment the author is not — and the author is almost never in the room when the record is finally needed, because the need arrives without warning, on its own schedule, long after everyone has moved on.

Build for the review that never comes

The discipline this asks for runs against instinct. It is natural to invest in records only where review is expected, to make the high-profile decision defensible and let the routine one go thin. But review does not arrive where it is expected. It arrives where something went wrong, and what goes wrong is, by definition, what nobody was watching. The decision that gets pulled into the light is rarely the one that was instrumented for it. It is the routine one, the one nobody thought twice about, the one whose record was made thin because no one imagined anyone would ever ask.

Which means the only safe assumption is that any decision might be the one. You cannot know in advance which records will be summoned, so you have to build them all as though each might be — not lavishly, but sufficiently: every consequential decision leaving behind a record self-contained enough to stand alone. This is what a Decision Receipt is for. It is not a document prepared for a hearing. It is the decision's own account of itself, made at the moment of decision, complete enough that no hearing is required to read it — and good enough that if a hearing comes, it is already ready.

The payoff is contestability that does not depend on access. A record that stands on its own gives the affected party — who has no relationship with the system, no line to the author, no standing in any proceeding — something they can actually take up and examine. They do not need to be granted a trial to understand what was done to them. The record holds its own evidence and its own rules, and it can be checked by anyone who finds it. That is what self-sufficiency buys: not just survival under formal review, but legibility to the person who was never going to get a formal review in the first place.

Build the record for the review that never comes. Most never will, and the discipline of writing for the absent reader is what makes a record trustworthy at all. And the one review that does come will come without warning, on a decision you did not flag, read by a stranger you will never meet — and on that day, the record is the only thing in the room. Make it able to speak for itself, because nothing else will be there to speak for it.

— Dispatches · Summit Cognitive