The Human-Decision Ledger
The one-page ledger to complete before AI touches any personnel decision: owner, criteria, bias check.
So here's a question I've started asking executives, and it stops the room every time: if AI helped pick your last layoff list, who signed off on it — and can you prove it? Right now, 26 Meta employees are suing over exactly that gap, alleging that activity metrics and algorithmic rankings chose who got cut from an 8,000-person layoff with no individualized human review. The tactic this week is a one-page ledger you complete before AI touches any personnel decision — layoff selection, performance flags, scheduling, hiring screens. Fill it out and you become the company whose answer to "who decided, and on what basis?" already exists in writing. That's the whole advantage.
Here's the ledger. It works in any doc tool — copy it, fill the brackets, one page per decision:
HUMAN-DECISION LEDGER — [DECISION, e.g., "Q3 support-team restructuring"]
Date opened: [DATE] Type: [layoff selection / performance flag / scheduling / hiring screen]
1) NAMED HUMAN OWNER
Owner: [NAME, TITLE] — accountable for the final call, not the tool.
AI inputs used: [SYSTEM + what it produced, e.g., "ranking model shortlist"]
Owner reviewed each affected person individually: [YES/NO + DATE]
2) DOCUMENTED DECISION CRITERIA
Criteria, written BEFORE anyone saw the AI output: [LIST]
Data the AI used: [LIST]
Data excluded on purpose: [e.g., leave status, activity during protected leave]
Could someone on medical/family leave score well on these criteria? [YES/NO — if NO, revise]
3) DISPARATE-IMPACT SPOT-CHECK
Compared selected vs. retained on: [age / sex / race / disability / leave status]
Selection rates roughly even across groups? [YES/NO — if NO, stop; escalate to HR + counsel]
Anyone selected while on or recently back from protected leave? [NAMES or "none"]
Human re-review of every flagged case: [YES/NO + REVIEWER + DATE]
Sign-off: Owner [INITIALS/DATE] HR [INITIALS/DATE] Counsel notified: [YES/NO]
And a short prompt that runs the spot-check for you — paste it into Claude or ChatGPT with your decision summary:
You are auditing a personnel decision before it is finalized.
Decision and selection data: [PASTE WHO WAS SELECTED, WHO WAS RETAINED, AND THE CRITERIA]
Answer three questions:
1. Which groups (age, sex, race, disability, leave status) show up at higher
rates among those selected than in the overall pool?
2. Which criteria could a person on medical or family leave not have satisfied
by design?
3. What one change to the criteria would most reduce the skew?
Flag anything a plaintiff's lawyer would circle.
Sample output (what the prompt returns on a disguised example decision):
1. Employees on or recently returned from family leave are selected at 3x
their share of the pool.
2. "Trailing-90-day activity score" and "AI tool adoption rate" cannot be
earned while on leave — both structurally penalize a 12-week absence.
3. Score each person's last 90 ACTIVE days instead; the leave-status skew
drops to roughly even.
Flag: no record shows who approved the final list or when.
What changes: You stop reconstructing personnel decisions after the lawyer's letter arrives, and start every AI-assisted one with the owner, the criteria, and the bias check already on paper. Ten minutes per decision buys you the answer everyone else scrambles for.
Where else this works: Promotions, performance-improvement plans, shift scheduling, and vendor-scored résumé screens — same page, different brackets.
The Meta suit, filed this month in federal court in Oakland, claims the metrics behind the cuts "by design" couldn't be earned by anyone on protected leave — and Meta's defense is one sentence: decisions "were and are made by people, not AI." Notice that defense is a ledger entry. It's provable in one page, if the page exists. I'm not a lawyer and this isn't legal advice, so run the template past your counsel before you lean on it — but open your first ledger this week, before the next tool runs.
Your AI Sherpa,
Mark R. Hinkle
Founding Publisher, The AIE Network
Follow me on LinkedIn
