UnicornWorksA Prismatics System
UnicornWorks
Prismatics

How to run a role

Tutorial

Seven steps, each linked to the place it happens. The sample slate is a real one—same scoring, pinned inputs.

  1. 01

    Describe the role as the work it actually is

    Not a title and a wish list — what the job involves, and which kinds of work it leans on most. You force-rank at least three of the five, because naming only one scores the job as though the rest were irrelevant. How concentrated it is gets counted from your own words rather than assumed.

    Open the role form →
  2. 02

    Add the candidates, with whatever you have

    A profile link, a CV, what you know firsthand. Thin material is fine and is read as thin — the product says what each category rests on rather than filling the gaps in.

  3. 03

    Send invitations where it matters

    For the people you are serious about, ask for their own result. It is the step that moves a read from an inference to something measured, and it is about fifteen minutes of theirs. They keep their own copy, and a result shared for a role is never reused for anything else.

    See your invitations →
  4. 04

    Assess the slate

    Each pairing comes back with how it would work, where it rubs, what would close the distance, and the footing under it. The ranked comparison flags it out loud when the rows are not resting on the same kind of evidence.

    See a real assessed slate →
  5. 05

    Take the probe plan into the interview

    Each place the record disagrees with itself becomes a specific question; each thin category becomes a specific piece of evidence to go and get. You spend the interview where the record is actually unresolved instead of covering ground you already have.

  6. 06

    Feed back what you learned, and re-run

    Debrief notes, work samples, reference conversations. Anything you observe or collect is ordinary input — no further instrument needed. The footing rises, the read sharpens, and what changed is kept.

  7. 07

    Decide, and record what you decided

    You decide. The record is written either way, and adding your own decision beside it is what turns a log of what a tool said into a log of what a human did with it.

    Your roles and their history →

How the read is actually made is on how a read is made; what the product does and does not claim, with the price and the limits, is in the brief. Both are kept in one place each, so this page does not repeat them.

UnicornWorks

UnicornWorks reads each person against one role and ranks their fit to that role—not the people. Each candidate is read on their own, and a Stance360 result is never compared between candidates. That ranking is one input to a decision a human makes, and never the decision. It is not a measure of anyone’s ability or worth, and it does not predict conduct—it describes how a way of working and a role’s demands would meet.

An assessment is only as strong as what it rests on, and every one states its own footing. The strongest rests on a Stance360 the candidate completed and shared for this role, with their acknowledgement recorded; the weakest is inferred from public signal with no input from them at all. Where a reading has no evidence behind it, it reports itself unavailable rather than being filled in. The person is always the best source on themselves.

Where local law regulates automated employment decision tools—including New York City—the employer using UnicornWorks is responsible for the candidate notices and the independent bias audit those laws require. We supply the record they rest on: every ranked slate, the version of the scoring that produced it, and an export built for an auditor. We collect no demographic information about candidates, ever.

A Prismatics System

Powered by Anthropics360