The line that closed on Friday

A small manufacturing shop runs on a revolving line of credit the way a body runs on blood pressure — quietly, continuously, noticed only when it fails. The owner draws on it to buy materials, fulfills the order, gets paid, repays the draw, and does it again the next week. One Friday the line is gone. Not reduced, not flagged for review: closed, by an automated periodic reassessment that ran against the account overnight and concluded the exposure no longer met the bank's risk appetite. The notice arrives by email over the weekend. By Monday the owner cannot buy the materials for the orders already on the books, the supplier wants payment he was going to make from the next draw, and a payroll run is four days out.

Nothing visibly malfunctioned. The reassessment model did what it was built to do: ingest updated signals about the business and the wider sector, re-score the exposure, and act when the score crossed a threshold the bank had set. The bank is within its rights; the line was never a guarantee, and the fine print always said so. And yet a decision has been made here that will be, for this business, effectively impossible to undo in time to matter — and that quality, irreversibility-in-practice, is what the situation is really about.

The decision was final before it was contestable

There is, in theory, an appeal. The owner can call on Monday, wait on hold, reach someone who can open a review, and submit whatever documentation is requested. But the model closed the line on Friday and the harm starts accruing immediately — the missed supplier payment, the order he cannot fill, the standing in front of his own employees. Even a successful appeal three weeks later does not rewind those three weeks. The remedy outlives its own usefulness. How much proof a decision owes the world scales with how hard it is to take back, and this decision is hard to take back not because the line cannot be technically reinstated but because the consequences of its absence cannot be unlived.

This is the trap of decisions that are reversible on paper and irreversible in effect. The institution looks at the appeal channel and sees a fully reversible action: a flag can be cleared, a line restored, a record corrected. The affected party looks at the same channel and sees a remedy that arrives after the damage has already compounded into something money cannot fully reach — a lost customer, a broken supplier relationship, a reputation that does not snap back. When the speed of the decision outruns the speed of the redress, "you can always appeal" stops being a real answer.

A decision you can technically reverse but cannot practically rewind is, for the person living it, simply final.

The proof owed in advance, not after

Here is where the irreversibility changes the obligation. For a decision the institution can cheaply and quickly unwind, a thin basis is tolerable, because the error is correctable before it bites. For a decision that bites the moment it lands, the basis has to have been good enough to justify it before it was taken — because there is no real second chance. The bank cannot discharge that duty by being able to assemble a rationale later, when the owner calls and a human pulls up the account and reconstructs, from memory and partial logs, a story about why the model probably acted. A story assembled after the harm is not the same object as a record captured at the decision. One is testimony shaped by the outcome; the other is evidence fixed before anyone knew there would be a dispute.

So the requirement the case exposes is concrete. A decision of this severity — one that withdraws the operating capacity of a business in a single step — should leave behind, at the moment it is made, a record that could stand as its justification: the signals the model actually weighed, the threshold it crossed and who set that threshold, what changed since the last reassessment to move the score, and the rules in force when it ran. Whether a decision can be proven sound later is decided at the start, by what the system was built to keep. You cannot retrofit that proof onto a Friday-night batch job that threw away its inputs.

What the irreversibility should have required

None of this argues that the bank must keep every line open forever, or that an automated reassessment is illegitimate. It argues that the severity of the action sets the standard of the record, and that some actions are severe enough to demand a pause built into the design — a decision that consequential and that hard to unwind might owe the account holder notice and a window before it takes effect, not a fait accompli discovered over the weekend. At minimum it owes a contemporaneous record robust enough that, when the owner does call, the bank is reading him what the system actually relied on rather than guessing on his behalf. A Decision Receipt for a closure of this kind is not a courtesy to the borrower. It is the only thing that lets the bank claim, honestly, that the call was justified at the moment it shut the line — and not merely that a justification could be improvised once someone complained.

The line closed in the time it took a job to run. The business it kept alive does not recover on that timescale. Between those two clocks sits the whole question the case is asking: when a decision lands faster than it can be undone, the institution owes its proof up front, addressed to the person who will feel it first — or it owes an admission that it acted on a basis it was never prepared to show.

The scenario above is illustrative — a composite drawn to show a pattern, not an account of any real person, company, or event.

— Dispatches · Summit Cognitive · The Casebook