The Screened Out

Show notes

Picture the candidate who never gets read. The application goes in, a screening pipeline scores it against a model assembled from resume text, keyword matches, assessment results, and whatever proxy signals the vendor folded in along the way, and the answer comes back before any human is involved: not advanced. When the candidate asks why, the honest answer inside the company is that nobody can say. The feature weights that produced the rejection are distributed across a model nobody on the hiring team built, trained on data they did not curate, and tuned by an objective they cannot fully recite. The decision was made. The account of the decision does not exist. That gap — between an output and an explanation anyone could stand behind — is what this episode is about.

An adverse employment decision is not a recommendation; it is an act that closes a door on someone's livelihood. The older intuition in employment law is that a decision of that weight owes the person on the receiving end a reason they can actually examine — something specific enough to be wrong, and therefore something they could contest. A form rejection fails that standard precisely because it is unfalsifiable. "We received many strong applications and have decided to move forward with other candidates" is not an account of anything; it is a sentence engineered to survive scrutiny by containing nothing. "You weren't selected" tells the rejected person that a judgment was rendered while withholding the only thing that would let them challenge it: what the judgment was based on, and whether that basis was even allowed to count.

The asymmetry is the whole problem. On one side sits a person's ability to work, the most consequential thing many people own; on the other sits a process whose internal reasoning is opaque even to the people running it. The stakes could hardly be higher and the legibility could hardly be lower, and the screening system gets to keep that imbalance because the burden of producing an account has quietly migrated onto the one party with no power to compel it. The candidate is asked to accept as final a decision that the decider cannot reconstruct — which is to say, a decision no one is in a position to defend if it were challenged. A judgment that cannot be defended is not thereby correct; it is merely unreviewed.

This is the warrant frame applied to the hiring funnel. The through-line of the show is that a conclusion is legitimate not because it was reached but because the path to it can be reconstructed and contested — and an automated screen that cannot produce its own reasoning is exactly a conclusion with no admissible path behind it. Standing and contestability are not amenities here; they are the difference between a decision and a verdict no one can appeal. The book this series accompanies, Admissible Reality, takes that demand literally: the goal is not faster screening but screening whose every "no" could survive a hearing, where the person it landed on would have been permitted to ask where the answer came from and whether it was allowed to decide.