Resources

Reports, templates and checklists.

Everything open except full benchmark PDFs, which ask for an email so we can send the next edition.

Benchmark #1 full report (publishes mid-Nov 2026).

Scope, modes, windows and stop protocol for authorising agent tests.

Trust status, DPA, AI addendum, sub-processors in one place.

The 16 failure classes as a pre-release checklist.

Matrix, GO/NO-GO, narrated repro and limitations.

The calculator behind the pilot exec readout.

How cells are run and judged.

A note you can forward to your PCI assessor.