INSTRUMENT 03 · THIRTY-SECOND LAB

Can you verify one finding in half a minute?

Compare a vague finding with an evidence-rich presentation. Choose verified or rejected when you are genuinely sure.

This is one worked example, used to demonstrate a point: the same underlying finding, shown two different ways, takes very different time to verify. It is not a test suite — there is one finding here, not ten. Once you have a real AI tool producing real findings, run the same clock-and-decide exercise yourself, offline, against ten of its actual outputs with five different reviewers, and compare the median time to the 30-second target below. That offline repeat is the "ten findings, five reviewers" protocol — it happens at your company, not on this page.

FINDING · WELDED STRUCTURAL PACKAGE

Inconsistency found in material certificate.

00:00

NO-FINDINGS CONVERTER

A statement that no findings were found is incomplete without the population, sources, checks and unresolved exclusions.

WHY THIS INSTRUMENT EXISTS

Verification time is not a theoretical quantity — it is set by whether a finding names the exact rule, the exact location and the exact evidence a reviewer needs to see. A finding that forces a reviewer to re-derive context defeats the purpose of automation even when the underlying conclusion is correct.

Timing a reviewer against a real finding, with a stopwatch and a real deadline, tests the actual interface a reviewer will use in production rather than an assumption about how fast checking should be.

§4.3 Measure the manual baseline first§10.3 Supplier document verification§2.2 Why fluent is not the same as correct§7.3 The reviewer interface is part of the control