The option value of the record

The strongest argument against keeping a record of your decisions is that you will almost never need it. Most decisions are never questioned. They are made, they take effect, and they recede into the ordinary business of the world without anyone ever asking the maker to account for them. Against that backdrop, the discipline of documenting a decision — capturing what was known, what rules applied, what was concluded and why — looks like effort spent on an event that overwhelmingly will not occur. The record sits in storage, costs something to keep, and is consulted on approximately none of the days it exists. Judged by the average decision, it is waste. That judgment is common, it sounds like prudence, and it is a category error.

The error is in what the record is being measured against. A record is not priced by the typical day, because a record is not a service you consume on the typical day. It is a claim on a future state that may or may not arrive — the state in which someone with standing and motivation contests the decision and demands that it be justified. On every other day the record is dormant. On that one day it is the whole game. To value it by averaging over the many days it does nothing is to average away the only day it was ever for. That is not a subtle mistake. It is the same mistake as pricing a fire extinguisher by how often the kitchen is on fire.

Mispricing the record as a cost

The "we will rarely need it" argument treats a decision record as a consumable — something you buy in order to use, whose value is the use you get from it. On that accounting, a thing used once in a thousand instances is priced at a thousandth of its list. But records are not consumables. A record is a right, held in reserve, that you exercise only if a specific and uncommon event occurs. Its value on an ordinary day is not zero because it went unused; its value is the standing option to use it if the day turns. You are not paying for reconstruction on the days nothing happens. You are paying for the ability to reconstruct on the day everything does.

Averaging misprices this because the distribution is not symmetric. If challenges to decisions were mild and evenly spread — a small, predictable nuisance attached to every decision — then yes, you could sensibly amortize the cost of records across all decisions and ask whether the average one earns its keep. But that is not the shape of the risk. Challenges are rare and, when they come, expensive: a lawsuit, a regulatory inquiry, an audit, a public accusation that the decision was arbitrary or discriminatory or negligent. The cost of being unable to answer such a challenge is not a small increment. It is, at the limit, unbounded — a loss whose size is set by the adversary and the forum, not by you. A record is the instrument that converts that open-ended exposure into a bounded one. What you hold, on every quiet day, is the right to make a ruinous day survivable.

Seen this way, "the average decision is never questioned" stops being an argument against the record and becomes a description of what kind of thing the record is. It is not a cost you recover through frequent use. It is a position you hold against an event whose whole character is that it is infrequent.

An option priced by its disasters

The clean way to say this is that a decision record is a real option. You pay a small premium now — the modest, mostly fixed cost of capturing the decision as it is made — and in exchange you acquire the right, but not the obligation, to exercise reconstruction later, if and only if a rare high-cost event makes it worth exercising. Like every option, it has two states. In the common state it expires unused, and the premium is simply gone, the way an insurance premium is gone at the end of a year with no claim. In the rare state it pays, and it pays in proportion to how bad the alternative would have been. The value of an option does not live in the common case. It lives entirely in the tail.

This is why intuition misprices records so reliably. Intuition reasons from the typical instance, and the typical instance is exactly where an option looks worthless. If you evaluate fire insurance by walking through an average day, you will conclude it is money set on fire, because the average day has no fire. The error is not in the arithmetic; the average day genuinely has no fire. The error is in thinking the average day is what you bought the policy for. You bought it for the day that is nothing like the average — the low-probability, high-severity day whose cost, unhedged, would dwarf every premium you ever paid. Options and insurance are not priced by their expected use. They are priced by the disasters they cover, discounted by how likely those disasters are.

You do not keep the record for the thousand decisions no one questions; you keep it for the one that could sink you, and that one is always in the mail.

A Decision Receipt is this instrument made concrete. Its value on the day it is created is not that anyone will read it — most never will. Its value is that, should the decision it captures ever be dragged into a forum where it must be defended, there exists a reconstructable, contestable account of what was actually done and why. That account is what stands between a bounded, defensible loss and an unbounded, indefensible one. The premium was cheap; the coverage is not. That gap — cheap to hold, decisive when it pays — is precisely the signature of optionality, and it is precisely what the "we rarely need it" objection cannot see, because the objection is looking at the premium and calling it the price.

Hold the option where the tail is fat

None of this makes the option free, and honesty about that is what keeps the argument from collapsing into a slogan. Options carry real premiums. A record costs something to produce, something to store, and something to maintain in a form that stays trustworthy over time. It also carries a genuine liability of its own: holding detailed accounts of decisions has privacy consequences, creates surfaces that must be secured, and accrues its own maintenance debt as formats age and systems change. These are not rounding errors to be waved away by anyone selling records, and I am not going to wave them away. They are the premium, and premiums are paid whether or not the option is ever exercised.

Which is exactly why the discipline is not "keep everything." Option value scales with the severity you are covering, so you buy the option where the tail is fat and let it lapse where the tail is thin. For a consequential, contestable decision — one that lands hard on a specific party, that a regulator could ask about, that a court could be asked to review, that could become the subject of a reputational crisis — the potential loss is large and the premium is trivial by comparison; there, holding the record is not caution but arithmetic. For a genuinely trivial decision, one whose worst plausible outcome is an inconvenience, the tail is thin, the option covers almost nothing, and the premium is not worth paying. Letting the option lapse there is not negligence. It is the same pricing discipline pointed the other way.

So the framing to abandon is the one that asks whether the average decision needs a record. The average decision never needs one; that was never the question. The question is whether, across the decisions you make, there is a tail of them whose failure to be defensible would be catastrophic — and for any organization making decisions that touch other people, there always is. Price the record the way you price every other hedge against low-probability, high-severity loss: not by how often you will use it, but by how much you stand to lose on the day you would have given anything to have it. The saving from skipping it is small, certain, and yours to keep. The cost of its absence is rare, enormous, and arrives on a day you did not choose. That asymmetry is the entire case, and it is why the record is worth holding — worth, in fact, considerably more than the intuition that resists it will ever admit.

— Dispatches · Summit Cognitive