Name the owner of every threshold

Pick any consequential threshold in a system you operate — the fraud cutoff, the risk score that triggers a manual review, the confidence line below which a case is auto-declined — and ask a single question of it: who owns this number? Not who can edit it. Who is accountable for its value being what it is, answerable when it is questioned, responsible for defending it or moving it. In most organizations that question has no answer, and the absence is not an oversight to be tidied up later. It is the ordinary condition of a system that was never built to assign one. An accountable decision with no accountable owner is not a rare failure; it is the default.

This is a different directive from writing the threshold down, and the distinction matters. Recording the operating point — its value, its rationale, its review date — gives you an account of the decision. But an account is not a person. A record can be complete and still name no one who will answer for what it describes. A threshold can be documented to the last decimal and still belong to nobody, because documentation captures the choice and ownership assigns the human. What follows is about the human: for every consequential operating point, a named person of record, accountable for its value, recorded alongside the threshold and its rationale and its next review. Not a team. A person.

The ownerless thresholds already deciding fates

Ownerless thresholds do not arrive all at once; they accumulate, one plausible decision at a time, until a system is run mostly by numbers no living person chose. One cutoff was set three years ago by an engineer who reasoned carefully about it and then left the company, taking the reasoning with her and leaving the number behind. Another was tuned at two in the morning during an incident, a defensible emergency adjustment that everyone meant to revisit once the fire was out, and nobody did — the temporary value became the permanent one by simple inertia. A third was never chosen at all: it is the vendor default that shipped with a component, a line somebody else picked for some other population, inherited wholesale because changing it would have required a conversation nobody had time for. None of these is a scandal. Each is a reasonable thing that happened. Together they produce a system whose most consequential parameters are a sediment of departed intentions.

The failure surfaces the moment the threshold is questioned. A regulator asks why the fraud cutoff sits where it does. An affected person asks why the score they got flagged them for review. A journalist asks who decided that a case below this line would be declined without a human ever seeing it. These are fair questions, and they are answerable in principle — but only if someone owns the answer. When no one owns the threshold, no one can say why it is what it is, which means no one can defend it, no one can responsibly change it, and no one can be asked to stand behind what it does. The number keeps deciding fates with perfect consistency and total anonymity. It has authority over everyone it touches and accountability to no one, and that gap is not neutral. It is the exact shape of a decision that has escaped its author.

Diffuse ownership is no ownership

The reflexive answer to this is to say the team owns it, and that answer is worse than useless because it feels like a resolution while providing none. Diffuse ownership is not a weaker form of ownership; it is its absence dressed as its presence. When a threshold belongs to a team, it belongs to everyone on the team equally, which is to say it belongs to no one in particular, which is to say that when the number is questioned each member can look, with complete sincerity, to the person beside them. Shared accountability is the mechanism by which accountability disappears in plain sight. The point of naming a person is not that a group cannot reason well about a threshold — it often reasons better — but that a group cannot be answerable. Only a person can be asked a question and be expected to have the answer.

A threshold with no owner is not neutral; it is a decision everyone is subject to and no one will answer for.

So the owner is defined by three capacities, and a name that carries none of them is decoration. The owner is the one who can defend the threshold — who holds the rationale and can state, on demand, what trade-off the number strikes and why it is the right one. The owner is the one who can change it responsibly — who has the standing to move the line and the context to know what moving it costs, so that a change is a decision rather than a fumble. And the owner is the one who can answer for it — who does not get to point elsewhere when the number is challenged, because the buck has a named place to stop. Strip any of these away and you have a nominal owner, which is another way of having none. A threshold is a standing decision, applied to everyone it touches for as long as it stands; a decision no person owns is a decision no person can be asked to account for.

Keep the register, make ownerless a defect

The instrument for all this is unglamorous and entirely sufficient: a register of the consequential thresholds and operating points, and for each one a named owner accountable for its value, the rationale of record, and the last and next review date. Not every threshold in the system — the register would drown in constants nobody's fate depends on. The consequential ones: the lines that change who is approved, refused, flagged, escalated, or left to a machine. Building the register is mostly an act of discovery, and the discovery is itself the finding. You will go looking for the owner of a cutoff and learn there isn't one. That is the register earning its keep on the first pass, before it has governed anything, simply by making the ownerless thresholds visible as the liabilities they already were.

Then treat an ownerless consequential threshold as what it is: a defect. This is the governance equivalent of unowned production code — a component in the critical path that no engineer will claim, that no one is on the hook to maintain, that everyone assumes someone else understands. You would not tolerate that in the codebase; you would call it what it is and assign it. Hold the threshold to the same standard. A consequential operating point with no named owner is not a gap to note and move past. It is a fault to be fixed before the system is considered sound, because a fate-deciding number that belongs to no one is a standing decision no one has agreed to defend — and sooner or later, someone will arrive to contest it and find no one home. Name the owner before that day, not after. The number is already deciding. The only open question is whether anyone will answer for it.

— Dispatches · Summit Cognitive