Most operations that "use AI" score zero on employment. Not because the models are weak — because the structure around them is missing. The Scorecard measures the five requirements.
1. Job description. Does each AI worker have written scope, deliverables, and standing orders — or does it start from a blank prompt every time? A worker without a job description isn't employed; it's improvising.
2. Owned artifacts. Does every file, record, and output have exactly one owner? Shared ownership between agents is how a night's work disappears in two overlapping keystrokes.
3. Standing rules from its own mistakes. When your automation fails, does the failure become a permanent rule — or does it just get retried? The performance-review loop is what separates a workforce from a slot machine.
4. Coded guardrails. Are the things that must never happen structurally impossible, or strongly worded? A prompt is a suggestion; a gate in code is a fact.
5. Escalation contract. Does the machine know exactly when to stop and hand you the pen — and does everything else run without you? Autonomy without a signature line is recklessness; signatures on everything is just you doing the job with extra steps.
While you wait: the rule library is the Scorecard’s source material — six real production failures and the rules they produced.