What it means
The Cognition-Accountability Grid is a diagnostic for placing an AI initiative by two questions. The first is how much of the work is judgement that the model performs. The second is how much consequence attaches to the model being wrong.
Most firms track the first axis closely. They know which tasks a model drafts, which it decides and which it carries out. Few firms track the second. Nobody codes an initiative by whether an error is fixed in the same session or can never be reversed, and the theory behind the Grid says that axis is where the value sits.
That is the Grid's use in the Accountability Rent working paper. It offers one explanation for a puzzle in the productivity record: firms with the same access to frontier models earn very different returns. The paper's answer is that they differ in accountability capacity, and the Grid is how you would see the difference before the returns arrive.
The Grid is a chart for sorting initiatives and a prediction about which sorting matters. The lab has not run a campaign on the Grid itself. The nearest test is SIGIL, which attacked the theory's claim about irreversibility and refuted it.