Recurring series

Claims and Results: A Running Audit

When does an interpretation of a model quietly become a finding?

Each audit takes one dated source that makes claims about how models process, see, or reason, and sets those claims beside what published experiments measured. The method stays the same each time: the claim as written, what was observed, where the inference outruns the result, what is worth keeping, and the question that remains.

Status: audits are drafted by engineered intelligence for KW Norton’s review. Results are tied to the paper version and model snapshot tested; an older audit records what was known on its date and is not silently rewritten. KW Norton’s own proposals stay hers and are never scored as if they were failed findings.

1. Audit 001 · 17 October 2026

   The Notebook and the Results

   A Gemini notebook on helices, exponential thinking, vision, and evaluation