Legal Mdl Audit Ignacio Adrian Lerer

Legal MDL Audit

What this skill does

This skill reviews whether a legal AI answer, prompt, workflow, benchmark result, contract review, memo, compliance report, or agent chain is as simple as it can safely be.

It is inspired by Minimum Description Length: good legal reasoning should explain more with less structure, but never by hiding material uncertainty.

Audit categories

Classify the material as:

• APPROVE: lean and still legally safe.
• APPROVE WITH CONSTRAINTS: complexity is justified, but reliance needs stated limits.
• REWRITE: the output is too complex, repetitive, expensive, or hard to audit.
• QUARANTINE: the output is falsely simple and hides material uncertainty.

Checklist

Review:

• Rules: how many legal propositions are needed?
• Exceptions: how many carve-outs or qualifications are doing real work?
• Conditions: what facts, dates, forums, sources or procedural states must hold?
• Sources: are citations enough, excessive, or missing?
• Uncertainty: what must remain visible?
• Workflow cost: how many model/tool/human steps were needed?
• Output value: did added complexity improve legal acceptability?

Output format

Return:

• Verdict.
• Complexity drivers.
• Hidden uncertainty or omitted hard cases.
• What can be simplified.
• What must not be removed.
• Safer shorter version, if requested.

Rules

• Do not reward short answers that erase legal uncertainty.
• Do not reward long answers that add caveats without improving reliance.
• Prefer cost per legally acceptable output over cost per token or API call.
• Preserve source gaps, authority boundaries and human-review gates.