Long-form
Pacing the Enterprise Frontier
The full argument for independent verification of enterprise AI — why self-assessment does not clear the bar, and what evidence an audit committee can actually rely on.
Sample assurance report
A complete redacted deliverable: scorecard, four outcome states, coverage against the frozen denominator, and every confirmed defect with its evidence.
Method & instruments
- Methodology — how an engagement actually runs.
- The rubric — criteria, behavioural anchors and version history.
- The eighteen failure domains — named defect classes with worked examples.
- Standard setting — why a threshold is where it is, and why we do not yet publish one.
- Judge–human agreement — the statistics we commit to publishing about ourselves.
- Known limitations — where the method is weak.
Tools you can use without an account
Bundle verifier
Paste a hash, get the notary record. Includes a published self-test.
ROI model
Interactive, with observable hard value separated from probabilistic risk value — and the assumptions exposed as inputs you can argue with.
Curriculum
156 lessons covering evaluation method, statistics, governance and operations, organised into five role tracks. Free, no account, no email. It is the largest thing we have built and it exists because the category needs practitioners more than it needs another dashboard.
Governance reference
- Standards mapping — NIST AI 100-1, AI 600-1, TEVV, ISO/IEC 42001, IEEE 1012-2024, and the correct EU AI Act dates.
- Glossary — the vocabulary, defined.
- Trust centre — the security pack, including what we do not hold.