Output scoring - a scorecard before you send

Philosophy

Before you send it, score it - and keep what is checkable apart from what must be judged. An AI output can read smoothly and still be wrong. This skill scores it in two layers: objective (facts you can verify mechanically) and subjective (quality you have to judge against an explicit rubric). Mixing the two is where false confidence comes from - "sounds right" is not "the citations check out".

It scores; it does not write or send. The decision is the lawyer's; the scorecard is the basis.

Layer 1 - objective (mechanically checkable)

Delegate, do not guess:

• Citations and provisions - do they exist in the source verbatim; were they all caught.
• Dockets / identifiers - real and correctly written.
• Completeness - did any RED finding drop out of the summary.