Diagnostic Instrument/Built In-House

The Initiative Register

One row per AI system, whoever built it — and three things to look for once it is filled in.

When AI is adopted function by function, each system arrives with its own measure, written when it began by whoever was closest to it. Nobody chose to have one ruler per system. The register lays them side by side.

Structure

The Register Columns

One row per system. Where a system is built in-house on a vendor’s model or harness, give each part with a different owner its own row.

SystemIts name, and the one-sentence job it does.

Function servedThe team whose work it does or shapes.

SourceVendor, in-house, or both. If both, which part this row covers.

Championed byThe person or team who pushed for it.

Built byWho wrote its instructions, rules, and test criteria — or, for a vendor part, the vendor.

Run byWho operates it day to day.

Reported on byWho currently says whether it is working.

Measure defined byWho wrote the measure of success, and when — before or after building began.

Baseline set byWho chose the starting point, and whether it has moved since.

Checked byWho verifies it against its measure — and whether they were part of the build.

Next checkThe date, and whether it was set before the system went live.

Findings also go toThe recipient outside the build's reporting line, if one has been named.

A blank cell is an answer. Leave it blank rather than guessing.

Interpretation

Three Things to Look For

One name in three columns

Look along each row at Built by, Reported on by, Measure defined by, and Checked by. Where the same person or team appears in three or four of them, that system's result is a claim its builders are making about their own work. That is a finding about how the role was assigned, not about anyone's honesty.

An empty Checked by, or an empty Next check

Either one means nothing will prompt a look at the system unless something goes wrong first. For an in-house system there is no contract renewal to force one.

How many different rulers there are

Count the distinct entries in Measure defined by and Baseline set by across the whole register. If every system was measured by whoever built or bought it, the numbers were never defined against a common standard, and comparing them compares the builders' choices as much as the systems.

“Can you name, for every system, someone who would be free to say its number had stopped holding? The rows where you cannot are where to start.”