DISPATCHES · Summit Cognitive

← All dispatches

The CasebookAccountabilityJuly 27, 20265 min read

The résumé that was never read

A screening model filters a candidate out before any human opens the file. When you ask who decided, the honest answer is no one — and that is the problem, not the relief it sounds like.

An experienced applicant submits a résumé for a role she is, by any fair reading, well suited to. She never hears back. There is no rejection letter with a reason, because there was no moment anyone would recognize as a rejection: a screening model scored the application against the posting, ranked it below a cutoff, and routed it to the bottom of a pile a recruiter never reaches. No person opened the file. No person decided she was unqualified. The application simply fell below a line, and the line did the rest. When she later asks why — and most applicants never get the chance — the most truthful answer the company could give is that no one in particular turned her down.

That answer is meant to sound exculpatory, and it is the opposite. We tend to picture the unjust hiring decision as a biased manager making a bad call. That failure at least has the shape of a decision: a person chose, the choice can be examined, and the person can be asked to defend it. The failure here is quieter and harder to repair. The applicant was filtered out by a process in which every component did its narrow job correctly — the parser parsed, the model scored, the threshold thresholded, the queue queued — and no single step constitutes the decision to reject her. The rejection is real; she feels all of it. But its authorship has been spread so thin across the pipeline that it has effectively vanished.

An unauthored rejection is an unanswerable one

This is the condition the case exposes precisely: not the biased decision, but the unauthored one. A biased decision has an author who can be wrong and corrected. An unauthored decision has no author at all, and that absence is not a small administrative gap — it is the dissolution of the structure by which we hold a hiring choice responsible. Responsibility presupposes an agent. When the recruiter can truthfully say she never saw the file, the hiring manager can truthfully say the role was filled from the candidates surfaced to him, and the vendor can truthfully say it only supplied a tool the employer configured, the applicant is left holding a real harm and facing a circle of people who can each, honestly, point elsewhere.

And the diffusion feels like fairness, which is what makes it dangerous. No prejudiced gatekeeper, no human caprice — just a model applying the same function to every applicant. But the absence of a visible decider is not the absence of a decision. The decision was made; she bears its consequence. What changed is only that the choice was distributed until no one need own it. A rejection whose author has evaporated is not a more objective rejection. It is one that has been quietly engineered so that no one can be asked to answer for it.

"No one decided not to hire you" is not an alibi. It is a confession that the decision was built so that no one would have to defend it.

What the model could not even be asked

Notice, too, what cannot be examined, because nothing was preserved to examine. Did the model score her down for a gap in her work history that has an ordinary explanation? Did it penalize a credential phrased differently from the ones it was tuned on? Did it lean on a proxy that tracks something the company would never knowingly screen for? These are answerable questions — but only if the system kept, at the moment it scored her, the basis for the score: the features it weighed, the threshold in force, who set that threshold, and the version of the model that ran. Absent that, the company cannot tell her why she was filtered out, cannot tell itself why, and cannot rule out the failure modes it would most want to rule out. The opacity does not just deny the applicant an account; it denies the employer the ability to know what its own pipeline is doing in its name.

Restoring the author

The work is not to put a human back in front of every résumé — at real volumes that is neither possible nor obviously better. The work is to ensure that a consequential outcome, however many automated steps produced it, can be traced to a person or an office that stands behind it and can be made to answer for it. A signature never claimed that one person performed every step; it claimed that one person would answer for the whole. Automation quietly removed the signature line from hiring, and the screen now rejects at a scale and speed no signatory ever vouched for. The remedy is to put the line back — not as a human pulling each lever, but as an owner who deployed the screen knowing what it would do and who therefore owns what it does.

The instrument that restores the author is a record that binds the outcome to its origin: what was decided, on what basis, against which threshold, under whose authority, traceable through which steps. When a screening result carries a record that names the responsible office and reconstructs the path that produced it, the rejection has an author again. The applicant can find the someone who must answer; the someone who must answer cannot dissolve back into the pipeline. The résumé that was never read is the perfectly ordinary shape of the failure — not a dramatic injustice, just a file that fell below a line and a company arranged, without anyone intending it, so that no one ever had to be the one who decided it should.

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

Continue from here

Turn the argument into a practice.

Get new dispatches, assess how your organization handles consequential decisions, or explore Summit Cognitive.