Notes on where models fail on real professional work, and how expert corrections become training signal. These pages describe methods and protocols. They do not report training-run results.


Pillars ·
This table quotes what each company's own terms and privacy pages say about training on customer data, as of the date checked.

Patrick Ojo ·
For finance back office and AI agent products: which expert-corrected sessions matter, what stays out, and how the rights review runs.

Patrick Ojo ·
Defines a correction trace and a matched-budget protocol for testing expert corrections against synthetic data on professional work. No results yet.