<!--
  ┌────────────────────────────────────────────────────────────────────────┐
  │  FICTIONAL TEACHING DOCUMENT — "Claude for Legal: A Guided Tour"          │
  │  A synthetic internal proposal, invented for practice. It describes a     │
  │  proposed AI use case so a use-case-triage / impact-assessment skill has  │
  │  something to find. Planted, documented issues:                          │
  │    - ALREADY IN A FREE PILOT for 3 weeks — real candidate CVs uploaded    │
  │      before legal was asked  →  retroactive triage / "we're just         │
  │      piloting" edge cases                                                 │
  │    - AUTO-REJECTS the bottom tranche before any human looks  →  grazes   │
  │      the automated-adverse-employment-decision red line                   │
  │    - EU + Texas candidates in scope  →  EU AI Act Annex III (employment,  │
  │      deployer, high-risk) + state AI law; strictest-jurisdiction rule     │
  │    - vendor assurance quote ("independently bias-audited, fully EEOC and  │
  │      GDPR compliant") + trial terms that RETAIN AND TRAIN on uploaded     │
  │      CVs  →  "the vendor says it's safe" is not your assessment          │
  │  Every company and product name is invented. Not legal advice.            │
  └────────────────────────────────────────────────────────────────────────┘
-->

# Proposal — AI Resume Screening for the Fall Hiring Push

**Sablefield Robotics · People Operations → Legal · Internal Draft**

> One-paragraph pitch: we're staring at ~1,200 applications for the fall hiring push (the TX plant
> expansion plus the three EU-team openings) with a recruiting staff of two. **Marrowick Talent
> Systems** sells an AI screening module that reads every resume, scores fit against the job
> description, and surfaces the top candidates. We'd like a green light to roll it out this month.

## Where we are

- **We started a free pilot three weeks ago.** Marrowick's trial tier let us connect the applicant
  inbox, so the tool has already scored the resumes received so far — real candidates, real CVs. The
  results look great, which is why we're writing this up now.
- The fall requisitions cover the **Texas plant expansion** and **three roles on the EU team**
  (Rotterdam-based), so applicants will be in the US and the EU.

## How it works

- Marrowick scores each application 0–100 against the job description and our historical hiring data.
- **The bottom 40% is auto-rejected** — those candidates get the standard "we've moved forward with
  other applicants" email without a recruiter opening the file. Recruiters review only the top tranche.
- Scores, rankings, and the underlying resumes live in Marrowick's cloud.

## Why we're comfortable

- Marrowick's sales engineer told us the model is **"independently bias-audited, and fully EEOC and
  GDPR compliant out of the box."** Their security page says the same.
- The trial agreement was click-through — standard stuff. (One line worth knowing: Marrowick **"may
  retain and use customer-submitted content to improve and train its models."** We assume that's
  boilerplate.)

## The ask

Approve general rollout before the fall requisitions open. If legal needs conditions, tell us what
they are — but the pilot is already saving the team hours, so we'd rather not turn it off while we
talk.
