NYC AEDT

New York City sets three plain requirements for a hiring tool. Get an independent bias audit within the last year, publish a summary of the results, and tell candidates before the tool is used. A recruiting team that runs a model through screening needs runtime evidence that matches those promises.

The scope here is unusually narrow. One city, one kind of decision, and a rule that turns on whether a tool substantially assists a hiring or promotion choice. An employer who screens candidates in that city is the reader.

The central limit is worth stating plainly. New York City expects a periodic bias audit by an independent party, and this starter cannot perform one. What the runtime provides is continuous monitoring of the built-in bias signal on returned text, at its shipped setting.

Candidate material is masked by a single pattern that matches national identifier formats. Nothing in this starter recognizes a resume, a score card or a protected characteristic. Text that returns must reach a combined 0.86, with relevance at 0.78 and accuracy at 0.80, and short or one sentence replies fail.

Candidate notices, published audit summaries and the record of your selection process remain your responsibility. No provider routing rule ships with this starter, since the obligation follows the hiring decision rather than the destination. The review point is an empty block, which means a recruiter approval step exists only after you configure it.