The tenant screening report

Picture an applicant who has done everything the market asks. They have the pay stubs, the references, the deposit in the account. They find a unit within reach, they submit the application, they pay the fee. And a day later a leasing office sends the line that ends it: we’re sorry, you did not meet our screening criteria. No number, no source, no record. Somewhere upstream, an automated screening report returned a score, the score fell on the wrong side of the landlord’s cutoff, and a door closed. The applicant is left holding a verdict without a reason, about a life they know better than anyone, delivered by a process they were never allowed to enter.

What the applicant cannot see is what the score was made of. Maybe it folded in a record that belongs to someone with a similar name and a birth year one digit off. Maybe it carried an eviction filing that was dismissed months later, the dismissal never propagating into the file the way the filing did. Maybe it was a clean case of mistaken identity — the kind that is trivial to create and expensive to undo. The applicant would recognize every one of these errors on sight. But sight is exactly what the process withholds. The one correction that would take a heartbeat is the one the design never allows.

The report that stands in for a person

A screening report is a compression. It takes a scatter of records held in databases the applicant has never seen — court dockets, prior-address histories, collections entries, name-and-date matches of varying confidence — and it reduces them to a compiled figure that a busy leasing agent can read in a glance. The compression is the product. It is what the landlord is paying for, and on its own terms it is defensible: a human could not read all of that, and the applicant pool is large, and a number sorts faster than a file.

But compression is lossy, and here the losses are not random. A report of this kind is probabilistic by nature. It reports likelihoods, matches with confidence bands, records that may pertain to this person. The errors it carries are quiet precisely because the format is confident: a mismatched record does not arrive flagged as doubtful, it arrives as a line item indistinguishable from the true ones. The stale filing that was later dismissed sits next to the current facts wearing the same typeface. Nothing in the artifact tells you which entries are load-bearing and which are ghosts.

Then the landlord’s cutoff does something the report never claimed to do. It converts a probabilistic estimate into a binary outcome. Above the line, a home. Below it, a denial. The report said roughly, with caveats, in the aggregate; the cutoff hears yes or no. All the hedging the score was built with is discarded at the threshold, and a hedge becomes a gate. The applicant does not experience a probability. They experience an eviction from a future they had already begun to picture.

The expert nobody consults

Here is the part that should trouble anyone who cares whether decisions are correct. The applicant is the single best error-correction mechanism in the entire system, and the system is built to route around them.

Think about who knows what. The screening vendor knows how to match records at scale but has no idea whether this record is truly this person’s. The landlord knows the cutoff but not the provenance behind the number. The applicant, and only the applicant, holds the ground truth: that the eviction was dismissed, that the judgment belongs to a stranger, that the address in 2019 was never theirs. They are the cheapest possible auditor — they need no subpoena, no database access, no forensic time. They already have the facts. They would work for free, and eagerly, because their home depends on it. And they are the one party the process never invites to check the inputs before the inputs decide.

This is the receiving-end problem in its sharpest form. A compiled score is a presumption — a default the system adopts about a person from a distance. Standing means that the person on the receiving end can put their particular facts in front of that presumption and have those facts actually override it, rather than being told that the presumption already accounted for everything worth accounting for. A presumption that cannot be rebutted by the one who knows better is not an estimate anymore; it is a sentence. The whole value of the affected party’s knowledge is that it is particular — it is exactly the local, verifiable detail that a statistical match cannot see and cannot correct on its own.

The one person who could have corrected the record in a heartbeat is the one the system was built never to ask — and calls the result objective.

What standing requires here

Standing is not a request for the machine to be kinder, or more accurate, or better supervised in the abstract. It is a structural claim: the person with the most at stake must have a real place in the decision before the outcome hardens. In this domain that resolves into three concrete things, none of them exotic.

First, notice that an adverse screening decision was made, and on what. Not a vague “did not meet criteria,” but the plain fact that an automated report drove the denial and what the operative pieces of it were. You cannot dispute a decision you were never told rested on a document — a point worth dwelling on in its own right. Second, the actual record the score rested on, in time to answer it. Not the conclusion, the inputs: the specific matched records, so the applicant can point at the line that is wrong and say, provably, that is not me, or that was dismissed, here is the order. A conclusion is not contestable; only its underlying evidence is. Third, a correction path that reaches the decision while the decision is still live — before the unit is leased to someone else.

That last condition is the one everything turns on, because a rental market does not wait. In a tight market the unit is gone in days. A dispute process that resolves in three weeks, however scrupulous, resolves into an empty result: the apartment is taken, the family has moved on to worse options, and the eventual finding that the record was wrong corrects a number in a file while changing nothing about the home that was lost. A wrongful denial in housing is irreversible in practice even when it is fully reversible on paper. Standing that arrives after the outcome hardens is not standing; it is an apology with a timestamp.

None of this asks the landlord to abandon screening or to read every file by hand. It asks that the artifact driving the decision be built for the person it decides against — that it carry its inputs, name itself as the cause of an adverse outcome, and open a real window to be rebutted while rebuttal can still change where the applicant sleeps. That is what a record built to be contested looks like when the stakes are a home. A Decision Receipt that exposes what the score was made of, and reaches the affected party in time, turns a silent verdict back into a decision someone can answer for. The point is not to make the machine gentler. It is to give the one true expert in the room a seat before the room empties out.

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