Is the evidence ready for a controlled AI use case?
Score the evidence foundation before selecting a model. The result is a remediation ledger, not a model score — like checking you have every ingredient and a working oven before judging a recipe, rather than blaming the recipe.
Example: if "traceability" scores low, it means that even when the AI's answer looks correct, a reviewer can't click through to the exact source paragraph to confirm it — like a student handing in the right final answer with no work shown. That gap needs fixing before you evaluate the model, because you'd have no way to catch a wrong answer that merely looks plausible.
WHY THIS INSTRUMENT EXISTS
A capable method applied to unreadable, unindexed or superseded evidence produces confident output about the wrong thing. The evidence foundation — source authority, metadata completeness, access control — determines the ceiling on quality before any model or configuration choice is made.
Scoring readiness before committing budget turns a plausible-sounding project into a testable claim: either the evidence supports the intended use today, or the real first project is closing the evidence gap, not building the tool.