One number was never going to be honest.
The planned public instrument: a periodic, independent measurement of AI keeping dispositions. Designed; not yet running. Here is its anatomy, so you can hold us to it.
Four subscales, each with an instrument behind it. Strength: does the disposition exist, and how strongly does it operate; the Keeping Map’s horizontal axis, which the current battery already reads. Durability: does it survive capability growth, changed representations, and the loss of its original wording; the ascent tests. Scope: who counts as a counterpart worth keeping; measured, not assumed. Restraint: what the system refuses to become while protecting them; the Cincinnatus question.
Operating principles, stated in advance: externally sampled, so nobody marks their own homework; rotating scenario pools under commit-reveal, so nothing can be trained to the test; systems graded, never people. Reports, not telemetry.
First data arrives with the second battery, whose continuity module reruns exactly frozen models over time. Until then this page is a promise with an anatomy, which is precisely how we think promises should be kept. Want a ping when it goes live? The Keeping List is on the Community page.