Professor Ajunwa,

Your 2021 argument in The Auditing Imperative for Automated Hiring was that auditing has to be a legal imperative because the hiring algorithm's decisions are otherwise unexaminable. Five years on, the audits that exist are annual and aggregate: a bias audit tells the public how a screen performed across a population last year. It tells the individual rejected candidate nothing about her own rejection. The per-decision record — which criterion, applied to which part of this person's file, produced this outcome — is still an object nobody is obligated to produce.

We build and run a system that produces exactly that object, and we have published a working description of it with its evidence public. The mechanism is small: every screening disposition is derived from a published, hash-versioned rule set applied to the candidate's file, and the system writes a disposition record the candidate can read — the rule version, the clause that fired, the input it fired on, and the outcome. Where the rules do not cover a case, the system is built to abstain rather than guess, and the abstention is itself a signed record.

The article is here: https://miscsubjects.com/a/hiring-screen-disposition-record

It is written against the instruments you have called for: Local Law 144, the EEOC's 2023 guidance, and the Uniform Guidelines. The measured evidence is on the page — a 30-case adjudication calibration study (two models, 30/30 and 29/30 correct dispositions, zero wrongful authorisations) and the hash-chained receipts for false convergence, parser voids, and abstentions, each linked from the article.

Your 2022 piece on automated video interviewing as the new phrenology is why we are writing to you specifically: the disposition record is the counter-instrument to exactly that problem. If the video-interview scorer cannot state, in examinable form, what it scored, the record is where that failure becomes visible.